UNIVERSITY OF CALIFORNIA Los Angeles

Improving Quantification and Detection in PET Imaging: Methods for Image Generation, Analysis, and Consequential Clinical Considerations

A dissertation submitted in partial satisfaction of the requirements for the degree Doctor of Philosophy in Biomedical Physics by Adam Leon Kesner 2008

© Copyrighted by Adam Leon Kesner 2008

The dissertation of Adam Leon Kesner is approved.

_____________________________ Daniel Silverman

_____________________________ Michael McNitt-Gray

_____________________________ Sung-Cheng Huang

_____________________________ Johannes Czernin

_____________________________ Magnus Dahlbom, Committee Chair

University of California, Los Angeles 2008

ii

If accomplishments are poetic plans making time fall in line with a notable beat and come from the grace of an uncertain hand then there's a time to close your eyes and step into the street

If a masterpiece exists before the chisel touches the stone and it whispers a song into life's melody making potential worth pondering in introductions to the unknown then risk is holding on to your right to see

If walking through moments is how we figure out where we stand and while standing for dreams we march to our fate and our dreams touch face of the unknown and unplanned then where you are is the question, and life is the debate

And if it’s in the forest of decision we see who we are in what we do then blessings come from the few stars, whose light is always with you

This dissertation is dedicated to Roy Berzon January 26, 1919 - December 2, 2002

iii

Table of Contents LIST OF FIGURES ...................................................................................................... VII LIST OF TABLES ........................................................................................................... X LIST OF ABBREVIATIONS ........................................................................................XI ACKNOWLEDGEMENTS ......................................................................................... XII VITA................................................................................................................................XV ABSTRACT OF THE DISSERTATION................................................................... XIX INTRODUCTION............................................................................................................. 1 Introduction to PET imaging .................................................................................................................1 PET image generation ...........................................................................................................................2 Methods in PET analysis .......................................................................................................................3 Limitations in PET .................................................................................................................................4 Goals for improvements.........................................................................................................................5

PROJECTS........................................................................................................................ 9 1.

SEMI-AUTOMATED ANALYSIS OF SMALL ANIMAL PET DATA .......................................... 9 Abstract..................................................................................................................................................9 Introduction .........................................................................................................................................11 Methods ...............................................................................................................................................12 Results..................................................................................................................................................18 Discussion............................................................................................................................................20 Conclusion ...........................................................................................................................................23 Figures.................................................................................................................................................25 Tables...................................................................................................................................................31

2.

DECONVOLUTION I............................................................................................................ 37 Introduction .........................................................................................................................................37 Methods ...............................................................................................................................................38 Results..................................................................................................................................................42 Discussion............................................................................................................................................44 Conclusion ...........................................................................................................................................48 Figures.................................................................................................................................................49

3.

DECONVOLUTION II .......................................................................................................... 65 Introduction .........................................................................................................................................65 Methods ...............................................................................................................................................65 Results..................................................................................................................................................68 Discussion............................................................................................................................................69 Conclusion ...........................................................................................................................................71 Figures.................................................................................................................................................72 Tables...................................................................................................................................................83

4. RESPIRATORY GATED PET DERIVED IN A FULLY-AUTOMATED MANNER FROM RAW PET DATA.................................................................................................................................. 87 Abstract................................................................................................................................................87

iv

Introduction .........................................................................................................................................88 Methods ...............................................................................................................................................90 Results..................................................................................................................................................96 Discussion............................................................................................................................................98 Conclusion .........................................................................................................................................103 Figures...............................................................................................................................................104 Tables.................................................................................................................................................118

5.

RECOMBINING RESPIRATORY GATED PET FRAMES (METHOD A).................................. 121 Introduction .......................................................................................................................................121 Methods .............................................................................................................................................122 Results................................................................................................................................................126 Discussion..........................................................................................................................................128 Conclusion .........................................................................................................................................131 Figures...............................................................................................................................................132 Tables.................................................................................................................................................140

6.

RECOMBINING RESPIRATORY GATED PET FRAMES (METHOD B) ................................. 142 Explanation of Method Extension......................................................................................................142 Methods .............................................................................................................................................143 Results................................................................................................................................................144 Discussion..........................................................................................................................................145 Figures...............................................................................................................................................147 Tables.................................................................................................................................................150

7.

RADIATION DOSE ESTIMATES FOR [F-18]5-FLUOROURACIL DERIVED FROM PET-BASED AND TISSUE-BASED METHODS IN RATS................................................................................... 151 Abstract..............................................................................................................................................151 Introduction .......................................................................................................................................153 Methods .............................................................................................................................................155 Results................................................................................................................................................162 Discussion..........................................................................................................................................163 Conclusions........................................................................................................................................169 Figures...............................................................................................................................................170 Tables.................................................................................................................................................178

TIME-COURSE OF EFFECTS OF EXTERNAL BEAM RADIATION ON [18F]FDG UPTAKE IN HEALTHY TISSUE AND BONE MARROW. ................................................................................... 181 8.

Abstract: ............................................................................................................................................181 Introduction: ......................................................................................................................................183 Methods: ............................................................................................................................................185 Results: ..............................................................................................................................................188 Discussion: ........................................................................................................................................190 Conclusion:........................................................................................................................................194 Figures...............................................................................................................................................195 Tables.................................................................................................................................................202

PROJECT COMBINATIONS..................................................................................... 203 9.

RADIODOSIMETRY EXTRAPOLATED FROM SEMI-AUTOMATED ANALYSIS OF SMALL ANIMAL FDG PET SCANS ........................................................................................................ 203 Abstract..............................................................................................................................................203 Introduction .......................................................................................................................................204 Methods .............................................................................................................................................204 Results................................................................................................................................................206 Discussion..........................................................................................................................................207 Conclusions........................................................................................................................................209

v

Figures...............................................................................................................................................210 Tables.................................................................................................................................................211

10. BIODISTRIBUTION OF 5-FU IN RATS MEASURED USING SEMI-AUTOMATED SOFTWARE AND ESTIMATED ATTENUATION CORRECTION ......................................................................... 214 Project Description............................................................................................................................214 Figures...............................................................................................................................................217 Tables.................................................................................................................................................218

11. FULLY AUTOMATED RESPIRATORY GATING AND RECOMBINATION OF GATES IN CLINICAL HUMAN PET............................................................................................................ 219 Project ...............................................................................................................................................219 Implementation ..................................................................................................................................219 Results and Discussion ......................................................................................................................221 Figures...............................................................................................................................................227 Tables.................................................................................................................................................237

DISSERTATION CONCLUSIONS ............................................................................ 242 APPENDIX.................................................................................................................... 243 12.

SOFTWARE .................................................................................................................. 243

Installation Instructions.....................................................................................................................243 User Manual ......................................................................................................................................244

13.

MAJOR PIECES OF CODE ............................................................................................. 252

Respiratory Gating Code ...................................................................................................................252 Scan Simulation .................................................................................................................................280

REFERENCES.............................................................................................................. 284

vi

List of Figures FIGURE 1-1: VOLUMES USED FOR FUSION PROGRAM ......................................................................................25 FIGURE 1-2: SAMPLE SCREEN SHOT FROM GRAPHICAL USER INTERFACE FOR THE FUSION SOFTWARE............26 FIGURE 1-3: DIGITIZED MOUSE MODELS ILLUSTRATING THE IMAGE FUSION PROCESS. ...................................27 FIGURE 1-4: ILLUSTRATION OF METHODS FOR DEFINING LANDMARK POINTS FOR REPRESENTING A REGION OF INTEREST. .............................................................................................................................................28 FIGURE 1-5: ILLUSTRATION OF SMALL ANIMAL PET MOUSE SCAN FUSED WITH ATLAS..................................29 FIGURE 1-6: COMPARISON OF SUV MEASUREMENT METHODS FOR THE LIVER MEASUREMENTS FOR THE NINE MICE. ....................................................................................................................................................30 FIGURE 2-1: EXAMPLE BLURRING PROCESS ILLUSTRATIVE OF BLURRING EFFECTS IN PET IMAGING .............49 FIGURE 2-2: EXAMPLE OF BENEFIT FROM A CLINICAL FDG PET/CT IMAGES................................................50 FIGURE 2-3: CONVOLUTION FUNCTION ..........................................................................................................51 FIGURE 2-4: EXAMPLE OF ITERATIVE STEPS ...................................................................................................52 FIGURE 2-5: PROJECTIONAL IMAGES OF HUMAN SIMULATION VOLUME REGIONS OF INTEREST. .....................53 FIGURE 2-6: DISTRIBUTION OF ORGAN SUV MEASUREMENTS IN 1000 RANDOMIZED DISTRIBUTION SIMULATIONS........................................................................................................................................54 FIGURE 2-7: JASZCZAK PHANTOM. .................................................................................................................55 FIGURE 2-8: ILLUSTRATIONS OF SCAN SIMULATION CORONAL SLICES (WITH BONE ACTIVITY HIGH FOR ANATOMICAL REFERENCE)....................................................................................................................56 FIGURE 2-9: CORRELATION OF MEAN MEASURED ROI VALUE AND TRUE ROI VALUE, BEFORE AND AFTER CORRECTIONAL METHODS, FOR A SAMPLE SIMULATED PHANTOM. .......................................................57 FIGURE 2-10: ERROR MEASUREMENTS FOR 1000 PET SIMULATIONS (~2.6 * 109 COUNTS IN SCAN VOLUME) . .............................................................................................................................................................58 FIGURE 2-11: ERROR MEASUREMENTS FOR 1000 PET SIMULATIONS (~2.6 * 105 COUNTS IN SCAN VOLUME) . .............................................................................................................................................................59 FIGURE 2-12: ILLUSTRATION OF RECONSTRUCTION AXIAL SLICES FOR SCANS TAKEN WITH DIFFERENT COUNT STATISTICS ...........................................................................................................................................60 FIGURE 2-13: CORRELATION BETWEEN TRUE MEASUREMENTS AND UNCORRECTED AND CORRECTED MEASUREMENTS AS A FUNCTION OF COUNT STATISTICS (N=28 ORGANS).............................................61 FIGURE 2-14: CORRECTED AND UNCORRECTED PET IMAGE SLICES OF JASZCZAK PHANTOM.........................62 FIGURE 2-15: CORRECTED AND UNCORRECTED ACTIVITY MEASUREMENTS FOR THE SPHERES IN THE JASZCZAK PHANTOM. ...........................................................................................................................63 FIGURE 2-16: MEAN ROI VALUES VS. ITERATION NUMBER FOR SIMULATIONS CORRECTING A 2D POINT SOURCE FOR PARTIAL VOLUME EFFECTS. ..............................................................................................64 FIGURE 3-1: TWO COMPARTMENT EXAMPLE OF HOW INDIVIDUAL ROI CONTRIBUTIONS ARE DETERMINED...72 FIGURE 3-2: VISUALIZATION OF LINEAR EQUATIONS.....................................................................................73 FIGURE 3-3: ILLUSTRATION OF CONCENTRIC ROIS, USED TO DEFINE ATLAS IN LESION SIMULATIONS............74 FIGURE 3-4: ERROR MEASUREMENTS [(MEASUREMENT – TRUE)/TRUE] FOR 800 PET SIMULATIONS WITH RANDOM BIODISTRIBUTIONS.................................................................................................................75 FIGURE 3-5: PET AND CORRECTED PET VALUES PLOTTED AGAINST TRUE VALUES FOR 800 SIMULATIONS. .76 FIGURE 3-6: SIGNAL RECOVERY AS A FUNCTION OF VOLUME (TOP) AND THRESHOLD (BOTTOM), FOR A SMALL LESION DIAMETER RELATIVE TO CONVOLUTION KERNEL SIZE...............................................................77 FIGURE 3-7: SIGNAL RECOVERY AS A FUNCTION OF VOLUME (TOP) AND THRESHOLD (BOTTOM), FOR A MEDIUM LESION DIAMETER RELATIVE TO CONVOLUTION KERNEL SIZE.................................................78 FIGURE 3-8: SIGNAL RECOVERY AS A FUNCTION OF VOLUME (TOP) AND THRESHOLD (BOTTOM), FOR A LARGE LESION DIAMETER RELATIVE TO CONVOLUTION KERNEL SIZE...............................................................79 FIGURE 3-9: SIMULATED LESION PROFILES (SMALL TO MEDIUM SIZED). ........................................................80 FIGURE 3-10: SIMULATED LESION PROFILES (MEDIUM TO LARGE SIZED)........................................................81 FIGURE 3-11: CORRECTED SPHERE MEASUREMENTS IN A JASZCZAK PHANTOM PET IMAGE DERIVED FROM DIFFERENT BLUR KERNELS....................................................................................................................82

vii

FIGURE 4-1: SIMULATED TIME-ACTIVITY CURVE ANALYSIS FOR SINGLE VOXEL (NOISELESS SIMULATION)..104 FIGURE 4-2: ILLUSTRATION OF VOXEL PROCESSING PRIORITIZATION. ..........................................................105 FIGURE 4-3: FLOW CHART SUMMARIZING MAIN STEPS IN IMAGE PROCESSING LOOP, ILLUSTRATED WITH EXAMPLE CURVES...............................................................................................................................106 FIGURE 4-4: EXAMPLE OF PROCESSING PROCEDURE FOR INDIVIDUAL VOXEL TAC......................................107 FIGURE 4-5: EXAMPLES OF GLOBAL RESPIRATORY SIGNALS DERIVED USING OUR IMAGE-BASED METHODS, SHOWN ALONG CORRESPONDING PRESSURE BELT MEASUREMENTS. ..................................................108 FIGURE 4-6: EXAMPLE SHORT-TERM FOURIER TRANSFORM (STFT) SPECTROGRAMS. .................................109 FIGURE 4-7: SINGLE VOXEL TIME ACTIVITY CURVE: THEORETICAL, SIMULATION, ACTUAL..........................110 FIGURE 4-8: SIGNAL OF COMBINED GLOBAL SIGNAL TRACE (N=500 VOXELS) DERIVED FROM SIMULATED DATA. .................................................................................................................................................111 FIGURE 4-9: SIMULATED VOLUME SLICES. ...................................................................................................112 FIGURE 4-10: SAMPLE SINOGRAM FOR 500 MS ACQUISITION, FROM A PATIENT SCAN...................................113 FIGURE 4-11: SAMPLE IMAGE RECONSTRUCTED FROM 500 MS ACQUISITION, FROM A PATIENT SCAN. .........114 FIGURE 4-12: SIGNAL OF COMBINED GLOBAL SIGNAL DERIVED USING OUR IMAGE-BASED METHOD TRACE (TOP), SHOWN ALONG SIDE THE CORRESPONDING PRESSURE BELT MEASUREMENT (BOTTOM) (SCAN #4). ...........................................................................................................................................................115 FIGURE 4-13: FIGURE ILLUSTRATES CORRELATION BETWEEN FREQUENCY ANALYSIS IF THE IMAGE DERIVED RESPIRATORY TRACE, AND THE CORRELATION SCORE BETWEEN THE BELT DERIVED AND IMAGE DERIVED RESPIRATORY TRACES..........................................................................................................116 FIGURE 4-14: EXAMPLE LESION MOVEMENT ON GATED SCAN DATA, WHICH WAS CREATED USING IMAGE BASED RESPIRATORY TRACE (SCAN #4). .............................................................................................117 FIGURE 5-1: ILLUSTRATION OF NOISE/RESOLUTION TRADEOFF FOR AN EXAMPLE UNGATED/GATED PET IMAGE.................................................................................................................................................132 FIGURE 5-2: TWO VOXEL TIME ACTIVITY CURVES REPRESENTING CANDIDATES FOR STRONG GATED/UNGATED WEIGHTS.............................................................................................................................................133 FIGURE 5-3: UNGATED, GATED, COMBINED GATED SIMULATED LESION VOLUME COMPARISON...................134 FIGURE 5-4: UNGATED, GATED, COMBINED GATED SIMULATED HUMAN VOLUME COMPARISON. .................135 FIGURE 5-5: UNGATED, GATED, COMBINED GATED IEC/NEMA PHANTOM SCAN COMPARISON...................136 FIGURE 5-6: DIFFERENCE IMAGES FOR THE HUMAN SIMULATION, ILLUSTRATING THE NOISE AND RESOLUTION IMPROVEMENTS BETWEEN THE DIFFERENT RECONSTRUCTIONS OF THE SIMULATED HUMAN DATA.....137 FIGURE 5-7: RELATIVE DISPLACEMENT OF THE CENTER OF MASS OF THE 2ND LARGEST SPHERE IN THE IEC/NEMA PHANTOM ON THE MOTION PLATFORM, AS MEASURED FROM THE COMBINED-GATED 4D DATA. .................................................................................................................................................138 FIGURE 5-8: LINE PROFILES 8 ML SPHERE IN THE IEC/NEMA PHANTOM, WHILE PLACE ATOP A PLATFORM UNDERGOING SINUSOIDAL MOTION .....................................................................................................139 FIGURE 6-1: EXAMPLE OF TWO VOXEL TIME ACTIVITY CURVES. ..................................................................147 FIGURE 6-2: UNGATED, GATED, COMBINED GATED SIMULATED LESION AND HUMAN VOLUME COMPARISON. ...........................................................................................................................................................148 FIGURE 6-3: SUMMARY OF LESION LOCATED ON AN INTERFACE WITH AND WITHOUT GATE COMBINATION EFFORTS FOR A LESION WITH DIFFERENT RELATIVE NOISE LEVELS AND DIFFERENT ACTIVITY RATIOS. ...........................................................................................................................................................149 FIGURE 7-1: ILLUSTRATIVE EXAMPLE OF ROIS USED FOR DOSIMETRY CALCULATIONS FROM EXAMPLE SCAN ...........................................................................................................................................................170 FIGURE 7-2 FLOW CHART ILLUSTRATING METHODS USED TO PROJECT HUMAN [F-18]5-FU DOSIMETRY, DERIVED FROM HARVESTING-BASED AND IMAGING-BASED RODENT MEASUREMENTS. .......................171 FIGURE 7-3: ILLUSTRATION OF ERROR WINDOW (MEAN ± SD) USED TO CALCULATE SOURCE ORGAN STANDARD DEVIATION FOR THE LIVER................................................................................................172 FIGURE 7-4: BIODISTRIBUTION MEASUREMENTS BETWEEN DERIVED FROM IMAGING-BASED (HARVEST CALIBRATION) MEASUREMENTS VS. HARVESTING BASED MEASUREMENTS. .......................................173 FIGURE 7-5: PERCENT TOTAL ORGAN DISINTEGRATION MEASUREMENTS (CUMULATIVE ACTIVITY ) OF [F18]5-FU CALCULATED IN HUMANS BASED UPON SMALL ANIMAL IMAGING-BASED AND HARVESTINGBASED ACTIVITY DATA. ......................................................................................................................174

viii

FIGURE 7-6: PROJECTED HUMAN ACTIVITY DISTRIBUTION AND RESULTANT RADIODOSIMETRY, FOR IMAGING (HARVEST-BASED CALIBRATION) AND HARVESTING RAT DATA. .........................................................175 FIGURE 7-7: TIME-ACTIVITY CURVES FOR THE LIVER, FOR HARVESTING- AND IMAGING-BASED METHODS ..176 FIGURE 7-8: SMALL ANIMAL (A) AND HUMAN (B) PET IMAGES ACQUIRED AFTER ADMINISTRATION OF [F18]5-FU. ............................................................................................................................................177 FIGURE 8-1: IMAGES ILLUSTRATING MOUSE IRRADIATION SETUP (AT LINEAR ACCELERATOR).....................195 FIGURE 8-2: DOSIMETRY VISUALIZATION. ...................................................................................................196 FIGURE 8-3: ROI VISUALIZATION.................................................................................................................197 FIGURE 8-4: TIME-COURSE OF MEAN [18F]FDG UPTAKE RATIO IN LIVER. ...................................................198 FIGURE 8-5: TIME-COURSE OF MEAN [18F]FDG UPTAKE RATIO IN LUNGS...................................................199 FIGURE 8-6: TIME-COURSE OF MEAN [18F]FDG UPTAKE RATIO IN FEMUR. .................................................200 FIGURE 8-7: TIME-COURSE OF MEAN [18F]FDG UPTAKE RATIO IN BRAIN....................................................201 FIGURE 9-1: ESTIMATED HUMAN ORGAN ACTIVITIES FOLLOWING FDG INJECTION (MGY / MBQ) ...............210 FIGURE 10-1: BIODISTRIBUTIONS OF 5-FU MEASURED USING DIFFERENT METHODS. ...................................217 FIGURE 11-1: EXAMPLE OF CORRESPONDING RESPIRATORY TRACE AND GATE RELATIONSHIP.....................227 FIGURE 11-2: UNGATED, GATED, AND COMBINED-GATED IMAGE FROM PATIENT #2 PET DATA...................228 FIGURE 11-3: UNGATED, GATED, AND COMBINED-GATED IMAGE FROM PATIENT #4 PET DATA...................229 FIGURE 11-4: UNGATED, GATED, AND COMBINED-GATED IMAGE FROM PATIENT #5 PET DATA...................230 FIGURE 11-5: UNGATED, GATED, AND COMBINED-GATED IMAGE FROM PATIENT #9 PET DATA...................231 FIGURE 11-6: UNGATED, GATED, AND COMBINED-GATED IMAGE FROM PATIENT #11 PET DATA.................232 FIGURE 11-7: PROFILES OF LESIONS MEASURED IN HUMAN SCANS...............................................................233 FIGURE 11-8: SUMMARY OF FWHM MEASUREMENTS TAKEN FROM PROFILES OF LESIONS IN PATIENTS......234 FIGURE 11-9: CONTRAST-TO-NOISE MEASUREMENTS FROM PATIENTS AND SIMULATIONS, FOR UNGATED, GATED, AND COMBINED-GATED MEASUREMENTS. ..............................................................................235 FIGURE 11-10: DISPLACEMENT OF LESION CENTER OF MASS MEASUREMENTS MEASURED FROM 5 PATIENTS STUDIED..............................................................................................................................................236

ix

List of Tables TABLE 1-1 RADIOACTIVITY CONCENTRATION MEASUREMENTS FOR NINE SIMULATED SMALL ANIMAL PET ACQUISITIONS. ......................................................................................................................................31 TABLE 1-2: SUMMARY OF ACTIVITY CONCENTRATIONS MEASURED BY THE SOFTWARE PROGRAM, FOR SIMULATED SCANS OF NINE DIGITAL MOUSE PHANTOMS AND THE ORIGINAL ATLAS PHANTOM. ...........32 TABLE 1-3: PEARSON CORRELATION COEFFICIENTS (R) OF ORGAN ACTIVITIES MEASURED IN ACTUAL PET IMAGES WITH THE SOFTWARE (SSUV) OR MANUALLY (MSUV) VERSUS ACTIVITY MEASURED IN HARVESTED ORGANS (HSUV) USING A WELL COUNTER........................................................................33 TABLE 1-4: CORRELATION COEFFICIENTS (R) FOR DIFFERENT SUV METHODS COMPARED TO HARVESTED ORGAN DATA ACROSS MICE FOR DIFFERENT ORGANS ............................................................................34 TABLE 1-5: REPLICATION CHARACTERISTICS FOR SSUV METHOD .................................................................35 TABLE 1-6: APPROXIMATE SUV CALCULATION TIME FOR SEVEN ORGANS ....................................................36 TABLE 3-1: SUMMARIZED MEASUREMENTS OF TRUE, UNCORRECTED, AND CORRECTED PET LESION SIMULATIONS ........................................................................................................................................83 TABLE 3-2: CORRECTED SPHERE MEASUREMENTS FOR JASZCZAK PET SCAN USING DIFFERENT ESTIMATIONS OF BLUR KERNELS FWHM [MM]...........................................................................................................84 TABLE 4-1: SUMMARY OF LESION UPTAKE MEASUREMENTS ........................................................................118 TABLE 4-2: SUMMARY OF TRACE SCORES. ...................................................................................................119 TABLE 4-3: DATA PROCESSING TIME............................................................................................................120 TABLE 5-1: HUMAN SIMULATION GATING STATISTICS ................................................................................140 TABLE 5-2: SPHERE PHANTOM GATING STATISTICS (2ND LARGEST SPHERE ~8ML) .....................................141 TABLE 6-1 SUMMARY OF STATISTICS FOR A SIMULATED LESION..................................................................150 TABLE 7-1: ACTIVITY DISTRIBUTION MEASUREMENTS (RAT).......................................................................178 TABLE 7-2: PERCENT INJECTED DOSE (PID) ESTIMATED FOR HUMANS (CUMULATIVE ACTIVITY). ...............179 TABLE 7-3: RADIATION DOSIMETRY ESTIMATES (MSV/MBQ). ESTIMATED RADIO-DOSIMETRY RESULTING FROM MEASUREMENTS BASED ON IMAGING, AND HARVESTING (OLINDA OUTPUT);. ........................180 TABLE 8-1: SUMMARY OF MEAN [18F]FDG UPTAKE RATIOS OBSERVED IN FOUR DIFFERENT TYPES OF TISSUE. ...........................................................................................................................................................202 TABLE 9-1: ESTIMATED HUMAN ORGAN ACTIVITIES FOLLOWING FDG INJECTION (MGY / MBQ) ................211 TABLE 9-2: CORRELATION COEFFICIENTS FOR ORGAN DOSE ESTIMATES ......................................................212 TABLE 9-3: MEAN RATIO OF EXPERIMENTAL / EXPECTED DOSE PER ORGAN .................................................213 TABLE 10-1: ATTENUATION CORRECTION FACTORS AT 511 KEV FOR STANDARD MOUSE PHANTOM SCALED TO RAT VOLUME. ................................................................................................................................218 TABLE 11-1: HUMAN #2 GATING IMPROVEMENT STATISTICS ......................................................................237 TABLE 11-2: HUMAN #4 GATING IMPROVEMENT STATISTICS ......................................................................238 TABLE 11-3: HUMAN #5 GATING IMPROVEMENT STATISTICS ......................................................................239 TABLE 11-4: HUMAN #9 GATING IMPROVEMENT STATISTICS ......................................................................240 TABLE 11-5: HUMAN #11 GATING IMPROVEMENT STATISTICS. ...................................................................241

x

List of Abbreviations PET – Positron emission tomography CT – Computed tomography ROI – Region of interest SUV – Standardized uptake value F-18 – Fluorine-18 isotope FFT – Fast-Fourier transform STFT – Short term Fourier transform IDL – Interactive data language

xi

Acknowledgements Chapter 1 is a version of Kesner, A.L., Dahlbom, M., Huang, S.C, Czernin, J., Kreissl, M., Wu, H.M., Silverman, D.H.S., “Semi-Automated Analysis of Small Animal PET data”. Journal of Nuclear Medicine (July 2006). Chapter 2 is a version of Kesner, A.L., Dahlbom, M., Silverman, D.H.S., “Utilization of coregistered PET/CT images for image deconvolution”. Paper presented at the annual conference of the European Association of Nuclear Medicine, Athens, Greece, 2006. Chapter 4 is a version of Kesner, A.L., Detorie, N.C., Dahlbom, M., Czernin, J., Silverman, D.H.S., “Respiratory Gated PET Derived From Raw PET Data”. Paper presented at the IEEE annual medical imaging conference, Honolulu, HW, 2007. Chapter 5 is a version of Kesner, A.L., Dahlbom, M., Czernin, J., Silverman, D.H.S., “Recombining Respiratory Gated PET Frames”, Paper presented at the IEEE annual medical imaging conference, Honolulu, HW, 2007. Chapter 7 is a version of Kesner, A.L., Hsueh, W.A., Yap, C., Czernin, J., Padgett, H., Phelps, M.E., and Silverman, D.H.S., “Radiation Dose Estimates for [18F]5fluorouracil Derived from PET-based and Tissue-based Methods in Rats”. Journal of Molecular Imaging and Biology, In Press (September 2008) Chapter 8 is a version of Kesner, A.L., Lau, V., Speiser, M., Hsueh, W.A., Agazaryan, N., DeMarco J., Czernin, J., Silverman, D.H.S., “Time-course of effects of

xii

external beam radiation on [18F]FDG uptake in healthy tissue and bone marrow”. Journal of Applied Clinical Medical Physics, In Press (May 2008). Personal Acknowledgements: There are many people who have provided support for myself and my research which I would like to note. My dissertation committee members have provided a foundation for all the work I have done: Daniel H.S. Silverman (Advisor) Magnus Dahlbom (Chair) Johannes Czernin Henry Huang Michael McNitt-Gray I have enjoyed being a part of a research family, that has consisted of my lab coworkers and I, and I would like to thank them very much: Nicole Detorie Elena Heckathorne Victoria Lau Erin Siu Cheri Geist Natalie Htet Guesh Cuan Wei-Ann Hsueh There are several people I wish to mention who have made time to play a pivotal role in our research projects: Michael Spieser Nzhde Agazaryan Ralph Bundschuh I would like to thank the microPET team, who has always made an effort to help us carry out our research, which has made quality work possible: David Stout Waldy Ladno Judy Edwards Antonia Luo

xiii

There are several people from the clinic, who have made themselves available and have helped both with technical project aspects and scientific understandings: Larry Pang Richard Eugene Jenny Kusanadi I would like to thank the administrative staff, whom I have often depended on, and have came through for me many times: Amber Luke Terry Moore Reth Thi Thach Soosan Roodbari I would also like to thank the many people, whom have offered help or advice over the years. All of my work has been made possible by the availability of such people. Claudia Kuntner Christiaan Schiepers Martin Auerbach Sibylle Delaloye Christine Wu Michael Kreissl John DeMarco William McBride Qinan Bao Jennifer Cho Erin Angel Sibylle Ziegler

xiv

Vita

January 18, 1980

Born, Madison, Wisconsin

2003

B.S. Physics, Minor: Computer Science, McGill University Montreal, Canada

2003

Fellowship-National Institute of Biomedical Imaging and Bioengineering 2003-2004

2004, 2006

UCLA Quality of Graduate Education Award

2005, 2006

AAPM (Southern California Chapter) Norm Baily award winner

2005

NSF travel award recipient for participation in the NATO Advanced Study Institute, Archamps, France

2005

Master’s degree Biomedical Physics UCLA Biomedical Physics Interdepartmental Graduate Program

2006

Advanced to candidacy

2007

Journal of Nuclear Medicine Alavi-Mandell prize for publication entitled “Semi-Automated Analysis of Small Animal PET data”, Journal of Nuclear Medicine

2007

IEEE conference Trainee Grant recipient

2006

ACMP young investigator presentation award

2007

AMI/SMI joint conference student travel award

2007

UCLA Pharmacology Department travel award

2007

UCLA Sylvia Sorkin Greenfield Award for outstanding research

xv

Publications and Presentations

Kesner, A.L., Dahlbom, M., Pio, B.S., Czernin, J., Silverman, D.H.S., “SemiAutomated Analysis of microPET data Acquired for Radiodosimetry Determination”. Paper presented at the 52nd Society of Nuclear Medicine Annual meeting, Toronto, Canada. 2005. Kesner, A.L., Detorie, N.C., Heckathorne, E., Speiser, M., Czernin, J., Dahlbom, M., “Investigation of the Effect of Respiratory Gating on Signal-To-Noise Recovery in PET”. Paper presented at the 52nd Society of Nuclear Medicine Annual meeting, Toronto, Canada. 2005. Kesner, A.L., Dahlbom, M., Huang, S.C, Czernin, J., Silverman, D.H.S., “Utilization of co- registered PET/atlas images for image deconvolution”. Paper presented at the 53rd Society of Nuclear Medicine Annual meeting, San Diego, California, 2006. Kesner, A.L., Dahlbom, M., Huang, S.C, Czernin, J., Kreissl, M., Wu, H.M., Silverman, D.H.S., “Semi-automated human radiodosimetry extrapolation from small animal PET”. Paper presented at the 53rd Society of Nuclear Medicine Annual meeting, San Diego, California, 2006. Hsueh, W.A., Yap, C., Czernin, J., Padgett, H., Kesner, A.L., Phelps, M.E., and Silverman, D.H.S., “Accuracy of imaging-based determinations of [F-18]-5Fluorouracil uptake in rats used for radiodosimetry calculations”. Paper presented at the 53rd Society of Nuclear Medicine Annual meeting, San Diego, California, 2006. Kesner, A.L., Dahlbom, M., Silverman, D.H.S., “Utilization of coregistered PET/CT images for image deconvolution”. Paper presented at the annual conference of the European Association of Nuclear Medicine, Athens, Greece, 2006. Kesner, A.L., Dahlbom, M., Huang, S.C, Czernin, J., Kreissl, M., Wu, H.M., Silverman, D.H.S., “Semi-Automated Analysis of Small Animal PET data”. Journal of Nuclear Medicine (July 2006). Article. Detorie, N.C., Kesner, A.L., Soldberg, T.D., Dahlbom, M., “Evaluation of Image Noise in Prospectively Respiratory Gated PET”. IEEE Transactions on Nuclear Science, Volume 54, Issue 1, Feb. 2007.

xvi

Kesner, A.L., Dahlbom, M., Czernin, J., Silverman, D.H.S., “Respiratory Gated PET based on Time Activity curve Analysis”. Paper presented at the 54th Society of Nuclear Medicine Annual meeting, Washington D.C, 2007. Kesner, A.L., Lau, V., Speiser, M., Hsueh, W.A., Agazaryan, N., DeMarco J., Silverman, D.H.S., “Time-course of effects of external beam radiation on FDG uptake in healthy tissue and bone marrow”. Paper presented at the 54th Society of Nuclear Medicine Annual meeting, Washington D.C, 2007. Kesner, A.L., Hsueh, W.A., Yap, C., Czernin, J., Padgett, H., Phelps, M.E., and Silverman, D.H.S., “Radiodosimetry Estimates For [F-18]5-Fluorouracil Derived From Imaging-Based and Harvesting-Based Methods in Rats”, Paper presented at the 54th Society of Nuclear Medicine Annual meeting, Washington D.C, 2007. Kesner, A.L., Lau, V., Speiser, M., Hsueh, W.A., Agazaryan, N., DeMarco J., Silverman, D.H.S., “Effects of external beam radiation in FDG small animal PET”, Paper presented at the ACMP annual meeting, Baltimore, MD, 2007. Speiser, M., DeMarco, J., Agazaryan, N., Chetty, I., Solberg, T., Kesner, A.L., “Development and Validation of An MCNPX Monte Carlo IMRT Source Model with Optimized Radiotherapy Interface and Comparative Analysis Software”. Paper presented at the AAPM annual meeting, Minneapolis, MN, 2007 Kesner, A.L., Detorie, N.C., Dahlbom, M., Czernin, J., Silverman, D.H.S., “Data Based Respiratory Gated PET”, Paper presented at the AMI/SMI joint conference, Providence, RI, 2007. Kesner, A.L., Dahlbom, M., Czernin, J., Silverman, D.H.S., “4D Image Improvements Through Recombination of Respiratory Gates”, Paper presented at the AMI/SMI joint conference, Providence, RI, 2007. Kesner, A.L., Lau, V., Speiser, M., Hsueh, W.A., Agazaryan, N., DeMarco J., Silverman, D.H.S., “Time-course of Metabolic Changes Following External Beam Radiation in Normal Tissues”, Paper presented at the AMI/SMI joint conference, Providence, RI, 2007. Kesner, A.L., Detorie, N.C., Dahlbom, M., Czernin, J., Silverman, D.H.S., “Respiratory Gated PET Derived From Raw PET Data”. Paper presented at the IEEE annual medical imaging conference, Honolulu, HW, 2007.

xvii

Kesner, A.L., Dahlbom, M., Czernin, J., Silverman, D.H.S., “Recombining Respiratory Gated PET Frames”, Paper presented at the IEEE annual medical imaging conference, Honolulu, HW, 2007. Kremslehner, R., Gurker, N., Kesner, A.L., Kuntner, C., “Computer-assisted localization of mice organs in micro positron emission tomography”. Paper presented at the Radioactive Isotopes in Clinical Medicine and Research Annual Symposium, Bad Hofgastein, Austria, 2008. Kesner, A.L., Hsueh, W.A., Yap, C., Czernin, J., Padgett, H., Phelps, M.E., and Silverman, D.H.S., “Radiation Dose Estimates for [F-18]5-fluorouracil: A Comparison of PET-based and Tissue-based Methods in Small Animals”. Paper presented at the Alternative Models for Animal Research annual workshop, sponsored by UCLA and the Johns Hopkins Center for Alternatives to Animal Testing (CAAT), Los Angeles, CA, 2008 Kesner, A.L., Dahlbom, M., Czernin, J., Silverman, D.H.S., “To gate or not to gate: Improved combined respiratory gated images through bootstrapped voxel based noise evaluation”, Paper presented at the 55th Society of Nuclear Medicine Annual meeting, New Orleans, LA, 2008. Kesner, A.L., Lau, V., Speiser, M., Hsueh, W.A., Agazaryan, N., DeMarco J., Czernin, J., Silverman, D.H.S., “Time-course of effects of external beam radiation on [18F]FDG uptake in healthy tissue and bone marrow”. Journal of Applied Clinical Medical Physics, In Press (August 2008). Kesner, A.L., Hsueh, W.A., Yap, C., Czernin, J., Padgett, H., Phelps, M.E., and Silverman, D.H.S., “Radiation Dose Estimates for [18F]5-fluorouracil Derived from PET-based and Tissue-based Methods in Rats”. Journal of Molecular Imaging and Biology, In Press (September 2008)

xviii

Abstract of the Dissertation

Improving Quantification and Detection in PET Imaging: Methods for Image Generation, Analysis, and Consequential Clinical Considerations By Adam Leon Kesner Doctor of Philosophy in Biomedical Physics University of California, Los Angeles, 2008 Professor Magnus Dahlbom, Chair

The field of PET imaging has been continually evolving towards better imaging capabilities, and improved clinical diagnosis. At the intersection of several specialties: physics, mathematics, computer science, and medicine, PET imaging has abilities, bases and limitations distributed abound. In this dissertation, contemporary PET imaging is examined, its limitations are categorized, and approaches for improvements are presented in the form of new techniques and standardized methods. More specifically, the subsections in this dissertation contend with respiratory gating, gate combination, image deconvolution, standardization of PET region of interest definitions, radiation dosimetry, and effects of clinical radiation on FDG PET uptake. The work was carried out with an emphasis on applicability, simplicity, and usability, in the hope that this work may be used to help shape the field of PET of tomorrow.

xix

Introduction Introduction to PET imaging The basic principles in PET imaging begin with detecting radioactive decay events, and estimating their origin. PET imaging uses annihilation photon coincidence detection, a method for radiation detection which exploits the fact that a when a positron annihilates, it gives rise to a pair of 511 keV photons are emitted at (nearly) opposite directions. Developments in PET began with early experiments utilizing coincidence detection in the 1950’s and 60’s (1-3). The first positron emission tomography (PET) machine, built in 1978 (4), extended the idea of coincidence detection to create tomographic images using a ring of detectors. PET technology, as well as its establishment in the clinical has continued to develop from the time of its first conception and embodiment through present day. Currently, the role of PET in medical care is continuing to expand (5-9).

PET is a unique tool in that it provides an in-vivo look at molecular pathways in human patients. As such, PET can be clinically utilized to asses many biological processes. The staging and diagnosing of cancer, neurological disease, cardiovascular disease, diseases of the central nervous system, among others, are some of the main uses for contemporary clinical PET.

The utility and value of PET imaging has been established in health science and treatment. At the same time, further research and development promises both a wider

1

utilization of PET in the clinic, as well as improvements in its established uses. It is both in the improvement of established PET applications, and the development of new prospects for the future, that the projects presented in this dissertation were conceived and carried out.

PET image generation In PET imaging, the image signal corresponds to nuclear decay events, distributed throughout space and time. The position of these events in the images correspond to the location of a radioactive isotope, often tagged to a pharmaceutical, residing in the field of view of a scanner.

There are a number of practical steps necessary for the practice of PET imaging, all of which require attention, and include:

-

Generating isotopes, and radio-labeled pharmaceuticals

-

Transporting isotopes or radiopharmaceuticals to an imaging facility

-

Choosing the best dose to obtain an image, balancing good signal statistics, with

detector dead-time effects and radiation dose -

Assaying the intended dose to be administered

-

Administering the radiopharmaceutical to the patient being imaged

2

-

Choosing a priori acquisition characteristics (bed offsets, frames, gates, 2D or 3D

mode, energy windows, etc.) -

Imaging the patient (5-50 minutes)

-

Sorting the raw acquired data into reconstructable sinogram image space

-

Reconstructing tomographic images (choosing the reconstruction algorithm, and

necessary parameters) -

Correcting images (using scale factor, scatter correction, attenuation correction,

randoms correction, arc correction, normalization correction and decay correction)

Following this multi-step process, images are then available for interpretation by researchers/physicians. The methodological landscape at this point, post image acquisition, is still indefinite, with respect to physiologic interpretation, data visualization, and analysis.

Methods in PET analysis PET imaging is utilized in a variety of clinical applications. Uses range from probing for characteristics of interest to characterizing of functional systems. Evaluation of clinical methodology in PET usually uses descriptors such as: -

Sensitivity

-

Specificity

-

Detectability

3

These measurements are dependant on both physiologic reactions to radiotracer pharmaceuticals, and physical characteristics of a PET scan, and the scanner (as mentioned in the image generation process described above).

PET images are analyzed both qualitatively, and quantitatively. Qualitatively, PET images are often utilized to gauge the overall biodistribution of a particular radiopharmaceutical (pharmaceutical tagged with a unstable isotope). Thus areas of low or high uptake may be indicative of a problem and can illuminate a need for clinical attention.

Quantitatively, regional PET uptake measurements are utilized to characterize a region’s reaction to a radiopharmaceutical. These measurements can be factored into the criteria for the diagnosing and staging of disease, or used as a comparable measurement for longitudinal or cross-sectional comparisons, and can further be utilized for anatomic volume measurements, time-activity curve analysis, or kinetic modeling, for radioactivity and radiopharmaceutical measurements.

Limitations in PET Limitations in PET signal accuracy introduce limitations in all of the resultant measurements/conclusions derived from PET images. Many of the steps in image generation have an associated uncertainty, inherent in the application or basic physics of

4

the methodology. For the most part, the manifestation of physical error can be characterized as limits in image noise and/or resolution.

Noise and resolution effects in PET imaging affect detectability, sensitivity, specificity. When using ROI analysis for PET data, noise in the measurements add to their uncertainty and motivate region definition to be larger, encompassing a greater number of voxels, which leads to more issues from resolution. The limitations of resolution can manifest as partial volume effects (10-12), which in PET are certain to add error to resultant measurements from small ROIs. The consequence of this asymmetric error occurring in undefined environments limits the ability standardize ROI definition and ROI uptake measurements in PET.

Goals for improvements In this dissertation, the work presented involves formulating methods for creating better images, drawing standard ROIs, and for generating useful analysis and conclusions from PET data.

Chapter 1 introduces a method for performing rapid semi-automated ROI analysis on PET images. We designed an algorithm to reform and co-register an digitized mouse atlas with a small animal PET mouse scan. Such an algorithm has benefits over traditional ROI methodology: it requires less man-time, creates more standardized uptake measurements across scan populations, and for each scan it produces uptake

5

measurements for 22 separate regions of interest, including those which would be difficult/impossible do define manually. Such methods utilize advances in contemporary computer technology and capabilities, and have potential to make a positive impact in the field of PET analysis.

In chapters 2 and 3, methodology is presented for correcting blurring artifacts in PET resulting from inherent limitations in resolution. Using co-registered information of PET data, anatomical data, and mathematical models of the blurring process, we are able to generate corrected uptake and volume measurements with improved accuracy over uncorrected data. This work utilized both simulations and actual PET/CT scans for exploring the algorithms’ robustness and reproducibility.

In chapter 4 we present a method for extracting respiratory information from routine PET data. This work offers advantages over the current state of the art, in that it is fully software based, and fully automated. We present in this dissertation both the algorithm design, and the transition of the methodology from research to clinical application, where we hope it can make an impact.

Chapters 5 and 6 deal with improving statistics in gated PET images, using novel techniques we developed. Because gated images use only a percentage of the original detected events, the improved resolution they offer come at the cost of noise. Thus interest has been generated in the field to combine the data from different gates, and we

6

are presenting an alternative methodology to the linear and non-linear gate combination algorithms which have been developed. Our methodology is fully automated, robust, easy to implement, and inherently protects itself from generating images any worse than the ungated data set.

Chapter 7 presents methodology for utilizing small animal PET data to estimate radiodosimetry in humans. At present, small animal PET is a fairly new technology, and there is need for establishment of its uses and methodology. Our work compares a traditional non-PET based method for dosimetry calculation with more two methods which utilize PET (derived from fully-image based measurements, and partially image based measurements). Our work presents methodology and procedures for using small animal PET for radiodosimetry and radiotracer development, and presents human radiodosimetry estimates of the radiolabelled anti-cancer agent [18-F]5-FU human radiodosimetry, thus contributing to the forward movement of the field.

In chapter 8 a study is presented utilizing small animal PET to characterize the effects of radiation in mice on FDG PET. This work was made with a cooperation between nuclear medicine and radiation oncology, thus spanning specialty boundaries (as patients do). Using longitudinal and cross sectional data monitoring the effects of radiation treatment, we were able to provide insight into the potential of using PET for monitoring the efficacy of radiation therapy, during therapy treatment regiments. This type of PET

7

monitoring is currently employed for many chemotherapy treatments, but has limitations with radiotherapy which have not been explored sufficiently in contemporary literature.

In all of the projects presented, time and effort was put in to creating a new understanding of problems and following novel ideas and approaches. The projects presented all consisted of multiple steps, which afforded prospects for choices and changes in the development of the methodology. Because of this, the potential for future improvements and developments for much of this work exists, in reformulation and refinement of the methods, as discussed in their respective chapters. Emphasis was given to simplicity, effectiveness, practicality, and robustness, as it is recognized that these are underlying features conducive for meaningful impact in the field.

8

Projects

1. Semi-Automated Analysis of Small Animal PET Data Abstract The objective of the work reported here was to develop and test automated methods to calculate biodistribution of PET tracers using small animal PET images. Methods: After developing software to semi-automatically co-register a digital mouse phantom with a small-animal PET scan, using visually distinguishable organs and other landmarks in the scan, the phantom was elastically transformed to conform to those landmarks in 9 simulated scans, as well as 18 actual PET scans acquired of nine mice. Tracer concentrations were automatically calculated in regions of interest (ROIs) reflecting the whole body and 21 individual organs. To assess the accuracy of this approach we compared (1) the software-measured activities in the ROIs of simulated PET scans to the known activities, and (2) the software measured activities in the ROIs of real PET scans to both manually established ROI activities in original scan data as well as actual radioactivity content in immediately harvested tissues of imaged animals. Results: PET/atlas co-registrations were successfully generated with minimal end-user input, allowing rapid quantification of 22 separate tissue ROIs. The simulated scan analysis found the method to be robust with respect to overall size and shape of individual animal scans, with average activity values for all organs tested falling within the range of 98±3% of the organ activity measured in the un-stretched phantom scan. SUVs measured from

9

actual PET scans using this semi-automated method correlated reasonably well with radioactivity content measured in harvested organs (median r=0.94) and compared favorably with conventional SUV correlations with harvested organ data (median r=0.825). Conclusions: A semi-automated analytic approach involving co-registration of scan-derived images with atlas-type images can be used in small animal whole body radiotracer studies to provide estimation of radioactivity concentration in organs, using a rapid and less labor-intensive technique than traditional methods, without diminishing overall accuracy. Such techniques have the possibility of saving time, effort, and number of animals needed in making such assessments.

10

Introduction The main goal of the work reported here was to develop and test semi-automated methods to estimate rodent radiotracer biodistribution from PET scans, involving image co-registration between actual small animal PET scans and a pre-defined digital mouse phantom (digital atlas) morphed to match designated features in the scans.

Software image registration methods are currently used for various purposes in medical imaging (10, 11). Combining images from modalities such as MRI, CT, PET, or SPECT can aid in assessing physiologic function and anatomical boundaries (12), and can be used for planning therapy such as surgical procedures or radiation delivery (13, 14). Methods of fusing together patient-specific image data with a standard anatomical atlas, similar to the work presented here has so far largely been focused upon brain images (4, 15, 16). The present work aims at further development and testing of methods for fusing actual PET scans with voxelized phantoms requiring minimal end-user input. Such methods have the potential to be helpful in a wide variety of applications, including facilitation of radiation dosimetry measurements, pharmokenetic compartmental modeling, and implementation of certain image processing techniques.

By co-registering a digital mouse phantom with a small animal PET mouse image, general anatomical information can be coupled with animal specific information in assessing biodistribution of tracer activity concentrations. Using a variation of Shepard’s

11

inverse distance weighted method for scattered data interpolations (17), we developed an elastic, feature-based algorithm for aligning corresponding anatomical landmarks, utilizing radioactivity-quantifying data from the PET scan and geometrical data derived from the digital phantom. The methods for scan analysis presented here were evaluated by comparing activity values calculated with use of the implemented algorithms to their true values in simulated PET acquisitions of multi-shaped digitized phantoms, as well as to activity distributions actually measured in animal PET studies using both manual region of interest (ROI) image analysis methods and determination of radioactivity content of tissues harvested from animals sacrificed immediately after imaging.

Methods A program was developed on a PC-based platform to allow rapid calculations of tracer activity biodistribution, following fusion of a three dimensional MOBY digital mouse phantom (18) with small animal PET mouse volumes (Figure 1-1). The aims in the development of the program were to make a user friendly tool which would enable the user do interact with the data in a intuitive way. And example screen shot of the graphical user interface can be seen in Figure 1-2.

In the operation of the program, the end-user is initially asked to semi-automatically (see below) define as many organs as possible on the small animal PET scan. A threshold ROI definition method is used for this purpose with threshold intensities and scan

12

smoothing values selected by the end-user for each otherwise automated ROI rendering. This whole process takes a few seconds for each organ. When each organ is defined, a characterizing three dimensional grid is overlaid on each organ (Figure 1-3) based upon an explicit dimensional integration center of mass weighting method we developed for this purpose (Figure 1-4). These points act as landmarks within the small animal PET scan, and correspond to similar pre-defined landmarks in the digital phantom. The two images are essentially ‘pinned together’ at these common points. While it is possible to employ more grid points to define each organ, in initial pilot runs of the program it was determined that a grid of 15 points was sufficient for delineating the general size and shape of most organs, and represented a practical compromise to avoid unduly hindering the speed with which the algorithm is executed.

A representation of these points of definition can be seen in Figure 1-3A. The actual fusion process consists of two further steps. First, the phantom is stretched in the anterior-posterior and left-right directions to best match the dimensions of the actual PET scan. No explicit rostral-caudal stretching is introduced because of complications arising from the variation on the extent of the animal inside the scanning field of view. However, stretching in this dimension occurs implicitly when landmark points are defined at the rostral and caudal ends of the scan (e.g. the head and the kidneys, respectively). Second, two sets of n landmark points pi and qi, i = 1,…,n in two 3-dimensional image representations are generated, where p represents the defined points on the phantom and q represents the defined points on the scan. The two sets can be used to make a new set of

13

n landmark shift vectors r, representing the shifts necessary to apply to the phantom landmark points to map them on top of the scan:

ri = pi − qi

i = 1..., n Equation 1-1

The algorithm will then analyze all the points ck in the phantom (where k=1,…, number voxels in the phantom) and calculate a displacement vector dk for each point. The displacement vector for each point is created as a combination of all the n landmark shift vectors, with the shift vectors within a closer proximity of point ck influencing its displacement vector more; n

dk =

∑w i =0 n

r

ki i

∑w i =0

ki

Equation 1-2

where wki is the weighting factor controlling the influence of each of the landmark shift vectors for a particular point ck, and is based on the distance between the threedimensional point coordinates (ckx, cky, ckz) and the coordinates of the phantom landmark points (pix, piy, piz). The wki values are calculated as:

wki =

1 1 = 2 2 (dist. between c i and p i ) (ckx − pix ) + (cky − piy ) 2 + (ckz − piz ) 2 Equation 1-3

14

The weighting factor was limited to a maximum value of 100, to avoid computational errors associated with possible values of infinity. Displacement vectors are rounded to integer values for each dimension, thus originating and terminating at the center of a voxel (rounding error assumed negligible since voxel dimensions are sub-resolution). Once the displacement vectors are calculated, every voxel can be translated from the phantom space to the PET space by its displacement vector. Figure 1-3B illustrates a grid in the phantom space being stretched and warped by this method. Completing this process will create a transformation map between the PET space and the phantom space. All the tissue and organ-specific voxels in the phantom space are then projected into the PET space and assigned a tissue type in the PET volume at each voxel. A relatively small number of voxels in the scans, not assigned a tissue type in the PET space due to under-sampling along the most stretched portions of the warped phantom, are assigned values corresponding to those of the tissue type represented with greatest frequency among the immediately adjacent voxels.

From the fused PET/phantom images, corresponding PET information is assigned to each particular phantom organ. The software also allows the user to visually inspect the image co-registration as a quality control measure. Figure 1-3C provides a graphical representation of the fused volumes.

To explore the accuracy of this image co-registration approach we examined the PET activity measurements it generated using 1) simulated PET acquisitions of digital mice of

15

different shapes and sizes, and 2) biodistributions determined through traditional scan analysis of nine actual small animal PET studies, as well as data acquired from the immediate harvesting of mouse organs following the scans, with the radiotracer content directly measured in a well counter.

For the first comparison we evaluated how accurately the approach could be used to calculate activity distributions in digital phantoms of varying dimensions, with predetermined biodistributions. Using the MOBY phantom software (18) several such phantoms were generated. Measurements of the average animal lengths in each of three dimensions and their standard deviations were made of the nine mice used in the small animal PET scans later evaluated in this study, and then nine phantoms were generated within a range of body dimensions representing the mean dimension lengths ± two standard deviations (see Table 1-1). Once the phantoms were generated, 2D PET acquisitions were simulated from each of them. Axial slices were forward projected into sinograms, convolved with a 1.8 mm blurring kernel, adjusted to model the effects of Poisson noise (total image counts scaled to 1.2*107 trues), and then reconstructed using filtered backprojection and a ramp filter. Attenuation was considered corrected for and not modeled. No post-processing filtering was used.

This process was carried out for the nine generated digital phantoms of varying sizes, plus once more for the un-stretched phantom used as the digital atlas. After repeating this simulation process for the digital atlas (initially assigned the same uniform radioactivity

16

concentrations per tissue type as for the set of nine digital phantoms), we were able to measure the degree of accuracy lost due solely to the blurring and noise introduced in the simulation of the PET acquisition and independent of the fusion process, since no elastic transformation step was required in conjunction with that simulation.

Secondly, we looked at nine actual small animal PET scans and compared the standardized uptake values (SUV) quantifying tracer concentration in each organ using the ROIs defined by the software program (sSUV) with the SUVs based upon manually defined ROIs (mSUV). In both cases, SUVs were calculated relative to measured wholebody tracer concentrations as follows: sSUV = radioactivity concentration in software defined organ ROI / radioactivity concentration whole body volume; mSUV = radioactivity concentration in manually defined organ ROI / radioactivity concentration whole body volume. SUV ratios were used as the unit of comparison since this is the most common ratio unit used in published PET literature.

As a criterion standard, the SUVs derived from these images were compared to an analogous unit (hSUV) derived from measuring the actual radioactivity content with a well counter in organs harvested from animals sacrificed immediately upon completion of imaging, as previously described (19) (hSUV = radioactivity concentration measured in harvested organ / average radioactivity concentration in all harvested tissue). For these studies, nine non-fasted mice were imaged twice in a small animal PET scanner from the tip of the nose to the caudal extent of the pelvis, as described previously (19): once with

17

FDG (10-15 MBq), and once with [18F] fluorocyclophosphamide (F-CYP) (15MBq synthesized as recently described (20)). The two scans were acquired within one week of each other using different radiotracers having different routes of excretion, in order to identify as many organs as possible. The actual radioactivity concentrations reported in the Results section refer to those associated with the F-CYP tracer.

As an indication of the reproducibility and inter-observer variability, software analysis measurements several times for a single FDG mouse scan, using different thresholds to define organs.

Results As a first test of the co-registration implemented in our PET biodistribution software program we studied nine simulated PET acquisitions of mice of varying dimensions. Table 1-1 and represent the data from the calculated organ activities measured for the nine simulated mouse scans, and their variance, along with (in, column 3) the activity measured in the simulated scan of the atlas phantom. The data in Table 1-1 are represented as ratio of organ activity measured in each simulated scan to organ activity measured in the simulated atlas phantom scan. The known organ activities (i.e., before application of blurring kernel and reconstruction algorithms, displayed in the final column of Table 1-2 were set proportional to those found in typical mouse FDG scans in arbitrary units, with activities in relatively low-concentrating organs set equal to 100.0. It

18

is seen that generally close correspondence exists between the values measured in the simulated scan set and the simulated atlas phantom scan, with comparable activity losses occurring in both (before application of any recovery coefficient factor), relative to the known activities.

The next test involved calculating organ SUV measurements in actual mouse scans. Figure 1-5 demonstrates the fusion of the spaces achieved for an example mouse. Table 1-3 provides Pearson correlation coefficients and associated p values, for each of the SUV definition methods relative to the values obtained from harvesting of organs (heart, brain, lungs, liver, spleen, large intestines and kidneys). Correlation coefficients for sSUV vs. hSUV measurements were higher in seven of the nine mice (median r = 0.943 with range 0.483 – 0.999) than those for the mSUV vs. hSUV measurements (median r = 0.778 with range 0.083 – 0.992), indicating that data generated with the semi-automated program for calculating mouse organ activities from small animal PET correlated well with harvested data, and compared favorably to data generated by manual ROI analysis of the same images. We also evaluated how well the image-based SUV values were correlated with hSUV values across the mouse scans on an organ-by-organ basis, as shown in Figure 1-6 and Table 1-4. If there were no error in the methods, we would expect to see in the figure the image measurements relate to the harvested measurements at a ratio of 1-to-1, however we see deviation from this for the software based measurements, and more so for the manual image based-measurements. From the table, we see some organs (e.g. lungs and liver) contributed to the improved correlation more

19

than others (e.g. brain and kidneys). Overall the correlation coefficients for sSUV vs. hSUV measurements were higher than for mSUV vs. hSUV measurements for six of the seven organs examined. The spleen correlation with hSUV values was poor using either image-based method (possibly due to its relatively high blood content in images, lost with harvesting), but worse for sSUVs.

The variability of the software based measurements are summarized in Table 1-5. We saw that when varying the threshold used in the organ definition step in the software method the correlation between the sSUVs and hSUVs ranged between 0.759 - .996 with a standard deviation of 0.082.

Discussion We approached the analysis of PET data through utilizing a co-registered digital mouse atlas phantom applied to small animal PET mouse scans, generating predefined ROIs to be overlaid on the actual images. After image co-registration was concluded, a complete set of 22 organ ROIs could be rapidly defined, including those that were not manually definable on the original PET image. This semi-automated approach requires two to five minutes of human interaction (for organ definition) and four to six minutes for computer processing (using a 3.4 GHz Pentium 4 processor), as summarized in Table 1-6. Thus a researcher, in the time span of about ten minutes, can evaluate a small animal PET mouse scan and obtain biodistribution and time activity data for 22 organs in each animal.

20

In assessing the robustness of the image co-registration method per se, we compared activity measurements from nine simulated PET scans of mice of varying x, y, and z dimensions to measurements from a simulated scan of a (non-stretched) atlas. These data, displayed in ratio form in Table 1-1, are quite close to unity (mean = 0.99, standard deviation = 0.02), indicating the ability of the program to perform well over a typically varying range of mouse dimensions. In Table 1-2 the simulation measurement statistics are summarized. The differences in actual and measurements (both positive and negative) stem from noise and blurring effects and misalignment of the co-registration.

Compared to the criterion standard of harvested organ data derived from nine mice, sSUV data were found to correlate favorably relative to mSUV data (Table 1-3,Table 1-4, and Figure 1-6). The scans associated with the most affected data sets (e.g., of mouse #1 and mouse #6) tended to be of relatively lower image quality, likely making accurate ROI definitions especially difficult manually. As the sSUV method was less perturbed by this, it permitted ROI measurements to be made

with reasonable accuracy in those

scans. This reflects the fact that with the image fusion process used in the sSUV method, every pixel that is mapped on top of the small animal PET scan is influenced by many other points of information and definition of the individual organ ROIs with this method is thus less critically affected by individual organ uptake levels. A potential advantage of this approach, then, is to allow reasonable estimation of organ radioactivity contents even in images where scan quality or target-to-background ratios are not conducive to precise manual definition of organ ROIs.

21

The image transformation and coregistration aspects of this work used an adapted method stemming from Shepard’s inverse distance weighted interpolation function (17). In our implementation, the tissue values from the digital phantom were projected into the PET space. Alternatively, the projections could have been calculated from the PET space into the phantom space. Upon closer inspection, both approaches yielded very similar measurements, affecting less than one percent of the voxels.

There have been many alternate methods published for achieving generalized image coregistration, which can be explored in future work. A few such methods are projectional (21) and bi-linear mapping (22) algorithms, spline-based algorithms (23-25), and mutual information-based algorithms (26). Algorithms have been compared in several studies (27-30), which generally concluded that different methods perform better under different circumstances. In our work we used a landmark-based function, as PET is a relatively low resolution imaging modality, particularly with respect to features having minimal accumulation of tracer above background level, and thus well suited to a distanceweighted interpolation function approach that makes maximal use of the most welldefined regions in each scan. The drawbacks of our algorithm, however, is that they are dependant on decent definition of landmarks. If these landmarks are not available or hard to define, the final accuracy will suffer.

22

The establishment of procedures to facilitate analysis of small animal PET, as well as human data PET, could prove useful in many areas. The assignment of voxels within a PET scan to specific tissue types provides a platform for dosimetric calculations, and potentially for achieving rapid patient-specific dosimetry. Better scan analysis techniques can also play a role in new radiotracer development with the possibility of taking advantage of image-based measurements in animals to replace some of the invasive measurements currently performed, thus speeding up initial testing and reducing the number of animals needed for such purposes.

Conclusion The main goal of this work was to develop and test methods with which to accomplish semi-automated image co-registration of a digital mouse phantom with small animal PET data, applied to calculation of activity distributions in PET mouse scans. Using these methods, we have been able to calculate radioactivity distributions with an accuracy at least as good as with conventional ROI analyses, with minimal end-user input. The coregistration of images not only helps with PET activity calculations, but also provides a three dimensional voxelized tissue-based model of the individual animal. Such data have the potential to aid in many applications of image analysis, including radiation dosimetry calculations, new tracer development, compartmental modeling, attenuation correction and partial volume correction (as discussed in chapter 2 and 3).

23

24

Figures

A

B

Figure 1-1: Volumes used for fusion program (A) Example Coronal slice of a standard PET scan. (B) Moby digital mouse phantom, used to generate mouse Atlas.

25

Figure 1-2: Sample screen shot from graphical user interface for the fusion software.

26

Figure 1-3: Digitized mouse models illustrating the image fusion process. Frame A shows a PET (left) and digital phantom volume (right) to be fused together (as 2D projectional views). Overlaid on the images are the points delineating the organ volume grids for several organs (note: in 2D view only 9 of 15 points used per organ are visible). The points in these grids define the parameters used in the transformation of the phantom matrix when fusing it to the PET scan. The organs are defined by the user (see text), while a set of points are automatically defined at the tip of the head (shown in orange) at the most rostral boundary of the whole body ROI, to ensure proper rostral-caudal stretching. Frame B illustrates the spatial warping applied to a grid in the phantom space (right) when fused to the PET space (left) for a given slice. The figure in frame C illustrates the fusion of the stretched phantom coregistered with the PET scan.

27

Figure 1-4: Illustration of methods for defining landmark points for representing a region of interest. The 3D object (for this example a kidney) is first spatially integrated over one dimension (A), and then integrated a second time over a second dimension, yielding a 1 dimensional volume distribution (B). A center of mass point is defined at the center of this distribution. Dimensional high and low points are defined at each end of the distribution so as to leave 5% of the volume outside of these bounds on each side. These three points provide a representation of the volume in one dimension. This process is repeated for each of the 3 spatial dimensions. From these points an overlying grid encasing the ROI is constructed (C). The corners of the grid are defined by the dimensional high-low boundaries, and additional landmark points are placed at the center of mass and its orthogonal projections on each of the faces of the rectangular grid.

28

Figure 1-5: Illustration of small animal PET mouse scan fused with atlas. Once the two data sets are fused, organs and regions of interest can be easily visualized in superimposed spaces. This figure illustrates 24 different organs view from different angels (rotated around rostral-caudal axis).

29

Line of inflection B

1.4

1.4

1.3

1.3

1.2

1.2

1.1

1.1

mSUV

sSUV

A

1.0 0.9 0.8

1.0 0.9 0.8

0.7

0.7

0.6

0.6

0.5

0.5

0.4

0.4 0.4

0.5

0.6

0.7

0.8

0.9

1.0

1.1

1.2

1.3

1.4

0.4

hSUV

0.5

0.6

0.7

0.8

0.9

1.0

1.1

1.2

1.3

1.4

hSUV

Figure 1-6: Comparison of SUV measurement methods for the liver measurements for the nine mice. Both sSUV (A) and mSUV (B) measurements are shown relative to the hSUV measurements (considered the criterion standard.). Regression lines are shown on the charts in black

30

Tables Table 1-1 Radioactivity concentration measurements for nine simulated small animal PET acquisitions. The dimensions assigned to the nine mice are shown in the table, along with the organ software measurement ratios, defined as the ratio of the organ activities measured in the simulated mice scans, to the organ activities measured in the simulated atlas phantom scan. This latter scan was obtained by applying identical reconstruction parameters to the atlas, as were applied to the other data sets (see text).

1

2

3

Simulated Mouse 4 5 6

7

8

9

Mean

SD

Dimension [mm] X

31.8

25.4

31.8

28.6

28.6

25.4

28.6

28.6

28.6

28.6

2.3

Y

22.9

16.9

19.9

22.9

19.9

19.9

16.9

19.9

19.9

19.9

2.1

Z

90.9

88.0

89.5

89.5

90.9

89.5

89.5

88.0

89.5

89.5

1.0

Organ Software Measurement Ratios Heart

0.97

1.03

0.98

0.99

0.98

1.03

1.01

1.02

1.03

1.00

0.03

Liver

1.00

0.99

0.99

1.00

0.99

0.99

0.99

1.00

0.99

0.99

0.01

Lung

0.99

1.03

1.00

0.99

1.01

1.02

1.01

1.03

1.04

1.01

0.02

Stomach

1.00

1.03

1.01

1.00

1.01

1.02

1.01

1.02

1.02

1.01

0.01

Kidney

1.02

0.97

0.99

1.00

0.97

0.96

0.99

0.97

0.97

0.98

0.02

Spleen

1.00

0.98

1.03

0.98

0.99

1.00

0.99

1.00

1.01

1.00

0.02

Intestine

0.98

1.00

0.99

1.00

1.00

0.98

1.00

0.99

0.98

0.99

0.01

Bladder

0.96

0.96

0.93

1.00

0.98

0.96

0.97

0.99

0.97

0.97

0.02

Testes

1.00

0.99

1.00

1.00

1.00

1.00

1.00

0.99

1.00

1.00

0.00

Rib

0.97

0.97

0.97

0.99

0.97

0.96

0.99

0.99

0.98

0.98

0.01

Spine

0.98

1.01

1.00

0.99

1.00

1.00

1.00

1.00

1.00

1.00

0.01

Brain

0.99

0.91

0.95

0.97

0.97

0.95

0.95

0.93

0.96

0.95

0.02

Body

0.98

1.00

1.01

1.01

1.00

1.00

1.01

1.03

0.98

1.00

0.01

31

Table 1-2: Summary of activity concentrations measured by the software program, for simulated scans of nine digital mouse phantoms and the original atlas phantom. Columns I and II show mean activity measurements and their standard deviation for each organ for the nine simulated PET images. Column III shows the activity measurements in the simulated atlas phantom. Column IV shows the pre-blurred organ activities, before simulated image reconstruction (i.e. before convolution with the blurring kernel and noise added). Activity values are in arbitrary units.

I

II

III

IV

Mean Measured Activity Actual Organ

Measured Standard

in Atlas-

Deviation

based

Activity

Activity in

(pre-blurred)

Phantom Set Phantom (n=9) Heart

174.2

4.6

173.4

225.0

Liver

128.0

0.7

129.1

160.0

Lung

116.7

2.2

115.1

145.0

Stomach

87.3

0.7

86.1

100.0

Kidney

337.4

7.3

343.9

500.0

Spleen

133.7

2.1

133.7

185.0

Intestine

86.2

0.7

86.8

100.0

Bladder

1561.8

32.6

1612.0

2500.0

Testes

83.0

0.3

83.3

100.0

Rib

83.3

1.0

85.2

100.0

Spine

83.3

0.8

83.7

100.0

Brain

165.3

4.1

173.2

235.0

Thalamus

132.0

19.9

100.1

100.0

Body

97.1

1.3

96.9

121.7

32

Table 1-3: Pearson correlation coefficients (r) of organ activities measured in actual PET images with the software (sSUV) or manually (mSUV) versus activity measured in harvested organs (hSUV) using a well counter.

mouse

sSUV vs. hSUV

mSUV vs. hSUV

r

p

R

p

1

0.943

0.001

0.083

0.859

2

0.957

0.001

0.973

0.000

3

0.947

0.001

0.956

0.001

4

0.893

0.007

0.825

0.022

5

0.908

0.005

0.560

0.191

6

0.999

0.000

0.276

0.550

7

0.934

0.002

0.915

0.004

8

0.483

0.272

0.424

0.343

9

0.996

0.000

0.992

0.000

33

Table 1-4: Correlation coefficients (r) for different SUV methods compared to harvested organ data across mice for different organs

Organ

sSUV vs. hSUV (r) mSUV vs. hSUV (r)

Heart

0.579

0.358

Brain

0.331

0.237

Lungs

0.799

0.303

Liver

0.589

0.192

Spleen

0.038

0.193

0.777

0.350

0.636

0.513

Large Intestine Kidneys

34

Table 1-5: Replication characteristics for sSUV method

Mouse # 2 Software organ definition threshold ( % max)

Correlation Coefficient

0.9

0.989

0.8

0.759

0.8

0.995

0.8

0.986

0.7

0.990

0.6

0.994

0.5

0.996

user defined best fit

0.990

(sSUV vs hSUV)

35

Table 1-6: Approximate SUV calculation time for seven organs

1 mouse

10 mice

sSUV

~ 5 min

~50 min*

hSUV

~45

~7.5 hours

mSUV

~30

~5 hours

*Time necessary for human involvement. Automated processing takes 10-20 minutes per mouse

36

2. Deconvolution I Introduction Blurring occurs in PET images, limiting details of a true activity distribution of a radiopharmaceutical within the field of view of a PET scanner. Blurring in imaging occurs several reasons. The main contributing limitations in resolution are inherent from uncertainties in positron range, annihilation photon non-colinearity, and detection location uncertainties (31). Blurring is a non-reversible function, that is it represents a loss of information, and thus and increase uncertainty for subsequent measurements or image representations derived from the data.

The extent to which blurring in clinical PET effects measurements or detectability varies relative to the geometry of the subject being scanned. The size, shape, and activity relative to background activity within the different organs all effect the resolution of the final PET image, and information about the true activity distribution is lost (Figure 2-1).

Since it is possible to a priori understand and characterize the blurring which happens in PET, researchers have made efforts, in several forms, to correct PET imaging measurements. The simplest approaches involve look up table correction factors (15, 3235), while more complicated approaches are embedded in reconstruction (36), or use advanced mathematical techniques(37).

37

We aimed to develop a method to correct regional activity concentration values for partial volume and other blurring effects utilizing geometrical information gained from image fusion techniques with small animal PET and phantom or CT images. PET/CT images are currently becoming a standard in the clinical PET imaging, and doctors utilize this coregistration of images qualitatively for making diagnosis, staging (38), and therapy planning (example PET/CT in Figure 2-2). But there is yet to be methods for quantitative utilization of this merge in modalities. These new methods in image co-registration along with contemporary computing power are opening the door for a new generation of PET image deconvolution.

Methods The main ideas behind our methods is, for a given scan, if we know what the boundaries of the true activity distribution are, and we can classify the process of blurring in the image, then we can estimate what true activity distribution would result in the blurred image we are seeing, and we can do this using iterative guesses to get closer to this solution.

The blurring which occurs in a PET image can be approximated mathematically by a convolution function (39) (Figure 2-3). Using this model, we are presenting the idea that what we see in a PET image can be thought of as a true activity distribution which has been convolved with a blurring kernel, with the kernel being specific to the type of the

38

PET scan and the particular scanner. And when we make a method to correct for blurring effects, we are trying to find what true activity, when blurred in this environment, produced activity distribution observed in our given PET image.

The method we are presenting aims to correct regional activity concentration values for partial volume and other blurring effects utilizing geometric information gained by the fusion of PET with an atlas volume, or a CT scan which can be used to create an atlas volume. The merger of these two type of volumes allows us see what type of geometry produced a given blurred distribution. And with the assumption that different regions defined on the atlas image had uniform activity distribution, we gain a constraint we can utilize to iteratively correct the blurred image.

Beginning with a PET image (i.e. a blurred activity distribution) and a co-registered atlas, every region in the whole PET volume can associated as belonging to a particular region of interest (ROI). Our methods work by processing (i.e. correcting) single ROIs one at a time. With the atlas and the image co-registered, the ROIs are defined and the atlas regions can be overlaid on the PET image, where we can calculate a ROI mean activity.

The true regional activity values for each organ ROI (region of interest) were estimated through iterative adjustments converging to match the convolution of the estimated true ROI value with the observed PET scan ROI value. The process for estimating the true

39

activity distribution has the iterative form: new estimate = old estimate * correction factor. This process is repeated until a stopping criteria is met (Figure 2-4).

The specific for of the equation we used to deconvolve each of the ROIs is shown in Equation 2-1. Corrective multiplication factors were applied at each iteration to the estimated “true” organ activity. This corrective factor was calculated by blurring the estimated “true” organ and vicinity, and then comparing the mean ROI uptake to the ROI uptake in the actual PET image. Iterations continued until the two values were within 0.1% of each other.

New estimate

( ) Est. true mean

i+1

= old estimate * correction factor =

( ) Est. true mean

( Mean PETo ) i

i

x kernel (Est. mean )

i

Stopping criteria: |Corr. Fact.| < 1.001 Or i > 50 Equation 2-1

40

This process was repeated for all of the organ ROIs, to generate images with estimated true activity distributions.

To test our method we used simulated human scans and PET/CT phantom scans.

The human scans were simulated by using the NCAT phantom (18). Activity distributions were generated by assigning uptake values to 28 different organs/regions within the human atlas (Figure 2-5). Associated PET 2D scan simulations were generated by (1) creating emission volumes from the atlas volumes, (2) scaling the number of events in the volume (to simulate realistic statistics), (3) forward projecting the axial slices into sinograms, (4) blurring the sinograms using a convolution function (2D Gaussian distribution with FWHM = 0.7 cm), (5) adding Poisson noise to each individual voxel, (6) reconstructing images using Filter back projection (ramp filter). Images and phantoms used voxels with a size of 0.3125cm3 and the field of view covered from the neck to the top of the legs (Figure 2-5). The blurring kernel used for the correctional methods was taken from simulating the distribution of a point source with the same reconstruction parameters as the scans had (to account for subtle differences). To asses the success and robustness of our methods we generated 1000 PET/phantom simulations. Activity concentrations within the different organs varied randomly within realistic boundaries (Figure 2-6) for each simulation.

41

For our actual PET/CT scans we used a Jaszczak Phantom (Figure 2-7). PET scans were acquired using F-18 with an sphere to background activity ratio of approximately 10:1. Scans were reconstructed using ordered-subset expectation maximization algorithm (6 iterations, 16 subsets), with attenuation correction applied (derived from co-registered CT), and reconstructed with a voxel size of 1.06mm x 1.06mm x 2.0mm. A CT scan was acquired of the same Jaszczak phantom on a Siemens SOMATOM Sensation 64 Slice CT scanner and reconstructed with a voxel size of 0.6mm3. A separate PET scan was also acquired using a single point source in the field of view, in the center of the field of view, approximately offset 5 cm radial center. The FWHM of the spread measured x=5.2mm, y=5.3mm, z=4.5 mm, looking similar to a Poisson distribution. This distribution was used to define the blurring kernel used with our algorithm (Figure 2-7).

Following scan acquisition, the CT scan was segmented using region growing algorithms so as to define the boundaries for each of the spheres and the background, thus creating an atlas for our image.

As a measure of accuracy we were able to compare estimated true sphere activity values with their know concentrations.

Results The correctional methods were applied to our data sets, and the subsequent uptake measurements were improved.

42

Simulations:

An example simulation can be seen in Figure 2-8 and the SUV measurement improvements for in Figure 2-9. In 1000 simulated PET scans, the correlation between the corrected organ activity measurements with the true measurements (mean: 0.991, range 0.985-1.000) improved over the correlation of the uncorrected data (mean: 0.964, range 0.931-0.999) in 100% of the human scan simulations. The mean normal absolute error between the corrected PET measurements and the true measurements (|PETtrue|/true) was 0.109 ± 0.078 (mean ± SD, 95% percent confidence interval 0.007-0.456). Before the correctional methods were applied to the mean error had a value of 0.289 ± 1.66 (95% percent confidence interval 0.002-1.180). The spread of the error measurements is shown in Figure 2-10. We can see, however, that an increase in the simulated noise changes some of the relative percent error measurements changed (Figure 2-11). Simulations were repeated with several different noise levels (Figure 2-12, 200 resimulations) and the improvements between the corrected and uncorrected images were characterized (Figure 2-13).

PET/CT phantom scans:

The corrected images correlated much better than the uncorrected images with the true activity concentrations as shown in Figure 2-14 and Figure 2-15 . The mean normal

43

absolute error between the corrected PET measurements and the true measurements (|PET-true|/true) was 0.019 ± 0.015 (mean ± SD, range: 0.001-0.045) for the six measurements. Before the correctional methods were applied to the mean error for sphere uptake measurements had a value of 0.27 ± 0.129 (range: 0.13-0.472). This measure of accuracy improved for 100% of the sphere measurements.

One of the things we noticed while developing this algorithm, was that with the more a priori information we can include as constraints, the better our results appeared. A simplified example of this can be seen in Figure 2-16, where we are correcting a point source for blurring by using an unknown and known background ROI. In this example, we se that when we had the background area defined, we were able to converge to a solution for the foreground more accurately, and more rapidly (less iterations). This extends to more realistic situations, where we can force free space (outside the body) to zero activity, or the heart blood pool to background body activity, for example.

Discussion We developed a practical method for accomplishing estimated recovery of some of the PET information lost by typical image degradation processes using simulated PET/Atlas co-registration. We have found in our simulated phantoms, that the correctional methods improved the activity measurements for all of the defined organs, indicating that the

44

methodology robustly works at the scale of human PET, that is at the volumes of the organs, relative to the size of the blurring kernel.

By utilizing PET/CT image co-registration for a Jaszczak sphere phantom, we diminished average error in the pixel intensity distribution matrix typically arising in acquiring whole body PET images, by more than three-fourths of its uncorrected level. The significance of extending our mythology beyond simulation to actual PET/CT coregistered scans, is that this combination of modalities is becoming commonplace in clinical imaging, and methodology which can utilize this already has a platform for practically applying deconvolution.

One common use of PET is for following uptake measurements in lesions for staging or monitoring treatment (40-45). The size of these lesions are such that their uptake is often distorted partial volume effects. Errors by partial volumes effects can also be compounded by the changes in lesion size and shape over time, having the effect to have uneven distortions over time, thus making longitudinal data less comparable. In the work we are presenting here we scanned the Jaszczak phantom and evaluated the abilities of the deconvolution method to deconvolve spheres of different sizes against a background activity, which can be analogous to clinical PET. We have found that we were able to both recover a better estimate of true activity within the sphere, as well as get this recovery consistently for spheres ranging in sizes 0.5mm3 – 16mm3. If it is these

45

phantom results can be extended to human PET images, they can be useful both in the clinical and in research.

The extension of the work we did here, to human PET/CT is dependant on the ability to extend the basic assumptions we used. Our method uses the assumption the PET image is correctly co-registered with an image atlas, the activity in the image atlas defined ROIs is uniform, and that we can model the blurring process.

We have shown (with our phantom PET/CT data) that the CT scan is capable of producing a atlas which can be used to for deconvolution. To generate our atlas we had a researcher define thresholds to use in region growing algorithms which best segmented the CT. This however is a subjective step in our methods, and the effect of variations in the quality of the segmentation was not studied in this work. Segmenting a human CT scan will certainly prove more complex than a phantom, whether its performed manually or by an automated algorithm. As it is now, image segmentation is being used for PET, CT, and PET/CT measurements (46-50). It is still uncertain how the variability of these type of measurements will affect an algorithm such as the one we are presenting. Additionally, the question of segmentation becomes a different question if one is trying to perform a local deconvolution (i.e. for a single lesion) or a apply the methods for a global deconvolution (as we have done in the simulated PET scans). The emerging development of PET/MRI (51-53) also provides another opportunity for improved image segmentation.

46

There are several assumptions used by our algorithm. We are assuming that the signal within individual ROIs are uniform. This is an assumption often used in image analysis in the clinic and in research (54-57). We are also making an assumption that we know what is happening when our true distributions are getting blurred, and that we can model this to a mathematical convolution function, and assume that we know what blurring kernel to use in the convolution function. We used a single blurring kernel to describe the blurring throughout the field of view. More work could be put in to more accurately model local blurring functions (which are expected to vary slightly throughout the field of view), possibly creating a blurring kernel map. However, similar work has shown that small variations in kernel size do not make a significant difference in recovery (39).

Another point of importance for global deconvolution is the order in which the different ROI’s are processed, along with the amount of a priori information we can apply to the deconvolution model (i.e. force the heart blood pool to have background activity). When we processed the 28 different organs in our phantom, we ordered the background organs to be processed first (lungs), and then processed the organ ROIs in order of magnitude of mean signal. We have seen in Figure 2-16, the more we know about an organs surrounding activity distribution, the more accurate an estimation we can obtain for the organs corrected activity.

47

All of this work was also repeated for simulated small animal PET mouse scans, using a digitized mouse atlas, and a smaller blurring kernel. The results from these simulations illustrated very similar potential for application and improvements.

Conclusion We developed a practical method for accomplishing estimated recovery of some of the PET information lost by a typical image degradation processes using simulated PET/Atlas co-registration. Our Initial experiments using a sphere phantom for PET/CT deconvolution have achieved similar improvements relative to what we saw in the PET/Atlas work. The recent industry embrace of combination PET/CT machines offer a renewed opportunity in the utilization of PET/CT data for practical deconvolution methodology. Our work presented here has evaluated simulations and PET/CT scans, aiming to present a base for further methodological development.

48

Figures

Activity in Body Slice

PET Slice

X Figure 2-1: Example blurring process illustrative of blurring effects in PET imaging

49

Hot spot

Hot lateral node

PET

PET/CT

Figure 2-2: Example of benefit from a clinical FDG PET/CT images. In the PET scan alone we see a non-characteristic region with increased uptake. In the combined PET/CT image, we are better able to characterize the location (lateral node) and size of the hyper-metabolic region.

50

= Figure 2-3: Convolution function

51

PET

CONVOLUTED

KNL =

Cost function

PET

C (corr fact)

Figure 2-4: Example of iterative steps

52

0

1

2

3

4

5

6

7

8

9

10

11

12

13

14

15

16

17

18

19

20

21

22

23

24

25

26

27

28

29

Figure 2-5: Projectional images of human simulation volume regions of interest. (0 – Air, 1 - Heart (Left Myocardium), 2 - Heart (Right Myocardium), 3 - Blood Pool (Left Heart), 4 Blood pool (Right Heart), 5 – Body, 6 – Liver, 7 - Gall Bladder, 8 – Lung, 9 – Stomach, 10 – Kidneys, 11 – Spleen, 12 - Vertebral Bone (head), 13 - Vertebral Bone (process), 14 - Pelvis Bone, 15 - Wet Rib Bone, 16 - Bone Cartilage, 17 - Abdom. Artery, 18 - Abdom. Vein, 19 - Bladder, 20 – Prostate, 21 - Asc. large int., 22 - Asc. large int. contents), 23 - Desc. large int. 24 -Desc. large int. (contents), 25 –Rectum, 26 - Seminal Vessicles, 27 - Vas deferens, 28 – Testes, 29 – Whole body projection)

53

Figure 2-6: Distribution of organ SUV measurements in 1000 randomized distribution simulations. (0 - Heart (Left Myocardium), 1 - Heart (Right Myocardium), 2 - Blood Pool (Left Heart), 3 - Blood pool (Right Heart), 4 – Body, 5 – Liver, 6 - Gall Bladder, 7 – Lung, 8 – Stomach, 9 – Kidneys, 10 – Spleen, 11 Vertebral Bone (head), 12 - Vertebral Bone (process), 13 - Pelvis Bone, 14 - Wet Rib Bone, 15 - Bone Cartilage, 16 - Abdom. Artery, 17 - Abdom. Vein, 18 - Bladder, 19 – Prostate, 20 - Asc. large int., 21 - Asc. large int. contents), 22 - Desc. large int. 23 -Desc. large int. (contents), 24 –Rectum, 25 - Seminal Vessicles, 26 - Vas deferens, 27 – Testes)

54

A

B

Blur kernel CT volume

PET volume (with ct)

Figure 2-7: Jaszczak phantom. (A) Sample image. (B) PET/CT setup illustration

55

(projected on two planes) (enlarged)

Emission volume

Scan simulation

Corrected scan

100%

Emission scan

Emission corrected

0%

Figure 2-8: Illustrations of scan simulation coronal slices (with bone activity high for anatomical reference). Top row – volumes used for true, uncorrected, and corrected measurements. Bottom row – subtraction images illustrating effects of blurring and correction (color scaled between 0 and max emission value).

56

3.00E+06

2.50E+06

2.50E+06

2.00E+06

2.00E+06

True uptake

True uptake

3.00E+06

1.50E+06

1.00E+06

1.00E+06

5.00E+05

5.00E+05

0.00E+00 0.00E+00

1.50E+06

0.00E+00 0.00E+00 4.00E+05 8.00E+05

1.20E+06 1.60E+06

1.00E+06

2.00E+06

3.00E+06

2.00E+06

Corrected measured uptake

Measured uptake

R = .837

R = .996

Figure 2-9: Correlation of mean measured ROI value and true ROI value, before and after correctional methods, for a sample simulated phantom. Each point on the charts represents the measurement of mean uptake in a particular organ

57

A

B

Figure 2-10: Error measurements for 1000 PET simulations (~2.6 * 109 counts in scan volume) . (A) Error measurements [(measurement – true)/true] for 1000 PET simulations with random biodistributions. (B) Mean and 95% confidence interval. (0 - Heart (Left Myocardium), 1 - Heart (Right Myocardium), 2 - Blood Pool (Left Heart), 3 - Blood pool (Right Heart), 4 – Body, 5 – Liver, 6 - Gall Bladder, 7 – Lung, 8 – Stomach, 9 – Kidneys, 10 – Spleen, 11 - Vertebral Bone (head), 12 - Vertebral Bone (process), 13 - Pelvis Bone, 14 - Wet Rib Bone, 15 - Bone Cartilage, 16 - Abdom. Artery, 17 - Abdom. Vein, 18 - Bladder, 19 – Prostate, 20 - Asc. large int., 21 - Asc. large int. contents), 22 - Desc. large int. 23 Desc. large int. (contents), 24 –Rectum, 25 - Seminal Vessicles, 26 - Vas deferens, 27 – Testes)

58

A

B

Figure 2-11: Error measurements for 1000 PET simulations (~2.6 * 105 counts in scan volume) . (A) Error measurements [(measurement – true)/true] for 1000 PET simulations with random biodistributions. (B) Mean and 95% confidence interval. (0 - Heart (Left Myocardium), 1 - Heart (Right Myocardium), 2 - Blood Pool (Left Heart), 3 - Blood pool (Right Heart), 4 – Body, 5 – Liver, 6 - Gall Bladder, 7 – Lung, 8 – Stomach, 9 – Kidneys, 10 – Spleen, 11 - Vertebral Bone (head), 12 - Vertebral Bone (process), 13 - Pelvis Bone, 14 - Wet Rib Bone, 15 - Bone Cartilage, 16 - Abdom. Artery, 17 - Abdom. Vein, 18 - Bladder, 19 – Prostate, 20 - Asc. large int., 21 - Asc. large int. contents), 22 - Desc. large int. 23 Desc. large int. (contents), 24 –Rectum, 25 - Seminal Vessicles, 26 - Vas deferens, 27 – Testes)

59

101

102

103

104

105

Average counts / 3.1 mm slice Figure 2-12: Illustration of reconstruction axial slices for scans taken with different count statistics Low to high – from left to right.

60

Mean Measured vs. True Correlation

A

Uncorrected

Corrected

1.2 1

Correlation

0.8 0.6 0.4 0.2 0 2.55E+03

2.55E+04

2.55E+05

2.55E+06

2.55E+07

-0.2 Counts in Scan

B

Mean Measured vs. True Correlation Uncorrected

Corrected

0.985 0.98 0.975 Correlation

0.97 0.965 0.96 0.955 0.95 0.945 0.94 0.935 2.55E+05

2.55E+06

2.55E+07

Counts in Scan

Figure 2-13: Correlation between true measurements and uncorrected and corrected measurements as a function of count statistics (n=28 organs). (A) 5 levels of noise, increasing in steps of 10x. (B) 3 levels of noise, increasing in steps of 10x (zoomed in relative to A).

61

0%

110%

PET data before correction

PET data after correction

Figure 2-14: Corrected and uncorrected PET image slices of Jaszczak phantom.

62

Activity ( Measured / True )

Normalized Sphere Mean ROI Values

1

corrected uncorrected

0 16

8

4

2

1

0.5

3

Sphere Volume (mm )

Figure 2-15: Corrected and uncorrected activity measurements for the spheres in the Jaszczak phantom.

63

A

B

Mean ROI value

Estimated PET Convolution of est. PET Pet Value

90 70 50

True value

30

Background value

10 -10

1

2

3

4

5

6

7

8

9

Mean ROI value

110

110

90 70

50 30

10 -10

10

0 1 2 3 4 5 6 7 8 9 0' 1' 2' 3' 4' 5' 6' 7' 8' 9'

Iterations map

Iterations

blurred image

map

blurred image

Figure 2-16: Mean ROI values vs. iteration number for simulations correcting a 2D point source for partial volume effects. (A) Case where only point source is segmented (atlas with one ROI). (B) Case where both point source and background are segmented (atlas with two ROIs).

64

3. Deconvolution II

Introduction Following our original deconvolution work (last section) we decided to explore a similar, but new approach. Once again the underlying premise was that if we have a blurred activity distribution, and we have a region of interest (ROI) map and a model for how the blurring occurs, we could find a somewhat unique solution to the inverse blurring problem. To do this, we are approaching the iterative correction from our previous work by utilizing well developed linear algebra properties and theory.

Methods Our methods have the form of sorting out the data so that the true activity for all the ROIs are presented in a system of linear equations, for which we can then solve using developed and established methods.

The create the linear system, the contribution of spill-in and spill-out factors from every defined ROI, to every voxel, were determined. This was done by assigning unit values to every voxel within the each defined ROI (while each one is being evaluated separately), and null values everywhere else, and looking at the then blurred version of the volume (example in Figure 3-1). This blurring process was done by a convolution function, and

65

represents the blurring which occurs when the PET image is degraded. This process is repeated for every ROI, and linear equations can then be set up from every voxel from the measured contributions from this ROI blurring process, as illustrated in Figure 3-2.

An evaluation of all the useful voxels within a typical PET volume can provide ~105 linear equations and x unknowns (x = number of ROIs), which could be solved using linear algebra (using a singular value decomposition (58)). This whole process can be done rapidly (seconds-minutes) and provides a estimation of a true activity distribution derived from a blurred activity distribution.

Human PET simulations: To test our methods, we generated 800 PET simulations with random activity distributions. Each PET simulation was made by convolving a emission volume (derived from the NCAT phantom (18)), and convolving it with a blurring kernel (FWHM = 1 cm in x, y, and z dimensions.

Lesion threshold definition simulation: We also tested the impact of ROI definition on signal recovery. To do this, we ran our correctional methods for several spherical lesion/background situations. Using a 5-to-1 lesion to background ratio, we simulated spheres with diameters 2, 6, 10 times the FWHM of the blurring kernel, while using different % maximum thresholds to define the lesion ROI.

66

Concentric ROI lesion simulations: Once we ran the tests as mentioned above, we found the methodology working effectively and robustly (as discussed in results section), so extensions of the methods were then explored.

We observed variation in the accuracy of the methods results relative to ROI definition. ROIs drawn too small were over compensating, while ROIs too large were under compensating, but correct ROIs are working very well with our methods. We reasoned to ourselves that if the algorithm had the flexibility to avoid over and under compensation it would handle it well. We then decided to take a different approach towards deconvolution by no longer framing a lesion ROI using a single ROI, but rather with many ROIs, organized by concentric shapes defined with region growing algorithms using different threshold values (Figure 3-3). This approach works because while we don’t know which of the ROIs are inside the of the true lesion volume, we know that the activity within each ROI should be uniform, and one of the RIO contours should closely approximate the true lesion contour. By doing this approach, we hoped to get both the corrected activity for lesion ROI, as well as a estimated size and shape. We tested this size/activity recovery method by simulating PET scans of lesions of several sizes, and applying our correctional methods.

67

Jaszczak PET simulations: Following simulations, we used the same lesion deconvolution algorithm for actual PET images. Using the Jaszczak sphere phantom images (from the first approach), we extended all the methodology used in the above simulations, and took measurements of the results.

Results Human PET simulations: The corrected images correspond very well with the true images and activity measurements (Figure 3-4 and Figure 3-5). The corrected ROI values in the human simulations (n=29) improved in error measurements (|true-measured| / true) an average of 0.17±0.28 (mean±SD).

Lesion threshold definition simulation: The relationship between accuracy of the correctional methods, and ROI definition based on threshold ROIs derived from the PET image are summarized in Figure 3-6, Figure 3-7, and Figure 3-8.

Concentric ROI lesion simulations:

68

Lesions of several sizes were simulated, and corrected using ROIs defined as contours of the blurred image. The measured improvements are summarized in Table 3-1, Table 3-2 and illustrated in Figure 3-9, and Figure 3-10.

Jaszczak PET simulations: When we tested our methods using the Jaszczak sphere phantom, we were no longer able to perfectly simulate the blurring process with a known convolution function. It became apparent that the size of the blurring kernel (in 3 dimensions) effected the corrected ROI values. Figure 3-11 illustrates the variation in corrected sphere values when estimating the PET scan blurring with differently sized blurring kernels. Table 3-2 summarizes the data from which these figures were created. The table and figure provide an indication of the magnitude of the error which results from incorrect blurring kernels used in the algorithm. From the table, we can see the most appropriate global blurring kernel to use corresponds to a Gaussian distribution with a FWHM in the x, y, and z dimensions equal to 5.88, 5.88, and 5.25 mm, respectively. This is the kernel which corresponds to corrected measurements with the least deviation from the true sphere values.

Discussion Our algorithm we are presenting is based upon the premise that the combination of a blurred image and a map of the uniform regions of activity allows for the recovery of unique solution for the inverse of the blurring function.

69

When we simulated random distributions to be blurred and corrected, we realized that our algorithm was able to correct the data with excellent accuracy. We then set out to further examine the capabilities and limitations of such an algorithm. We found that when the ROIs were defined correctly, or close to correctly (within 10%-30% of true volume depending on size relative to resolution), we were able to accurately recover true activities, but when boundaries were incorrectly defined it was possible for the uncorrected data to be more accurate (summarized in Figure 3-6, Figure 3-7, and Figure 3-8). Thus the accuracy of defining regions of similar uptake is important.

In lesion/background scenarios, we were able to correct both uptake and lesion diameter. by making an assumption that activity was uniform within threshold specific ROIs. Thus we are presenting a fully automated algorithm to correct for partial volume effects in lesion uptake measurements. This type of algorithm has potential to create standardized measurements across studies, scanners, researchers, and centers, as the variability is published SUV measurements is quite large.

By correcting our image using many differently sized kernels (Figure 3-11, and Table 3-2), we were able to solve for a best effective kernel. More traditional approached for the characterization of the blurring kernel involve measurements from the scanning of a point source. Using our methods to find the blurring kernel for an image may have

70

advantages over using a point source in that they contain contributions to the blurring from the actual activity distribution for the image being analyzed (relative to unique biodistribution, scatter, deadtime). Additionally, with further methodological development, it can be possible to solve for the size and shape of the blurring kernel in the same system of equations presented in this method, as there are already many more equations than unknowns (would probably take place in frequency space where convolution turns into multiplication).

Conclusion By using blurred images, blurring models, and distribution maps, along with other available a priori information, we have been able to solve for inverse blurring with an accuracy superior to uncorrected images in simulations and phantom PET scans.

71

Figures

Figure 3-1: Two compartment example of how individual ROI contributions are determined. Each ROI within the image map is blurred individually, and the subsequent relative grey-scale in each voxel is used in the methodology as a contribution factor for the particular ROI.

72

Ax

+B y+ Ax

Ax +

By +

Ax + Cz

+…

+B

y+

=

By

+…

Cz

=

=

Va lu e

Ax + Cz

+C z+ …

+…

=

By +

Va lu e

Figure 3-2: Visualization of linear equations.

73

Va lu

Cz

Va lue

e

+…

=

Va lu e

ROI 6 ROI 5 ROI 4 ROI 3 ROI 2 ROI 1 Figure 3-3: Illustration of concentric ROIs, used to define atlas in lesion simulations.

74

Figure 3-4: Error measurements [(measurement – true)/true] for 800 PET simulations with random biodistributions. (0 - Heart (Left Myocardium), 1 - Heart (Right Myocardium), 2 - Blood Pool (Left Heart), 3 - Blood pool (Right Heart), 4 – Body, 5 – Liver, 6 - Gall Bladder, 7 – Lung, 8 – Stomach, 9 – Kidneys, 10 – Spleen, 11 Vertebral Bone (head), 12 - Vertebral Bone (process), 13 - Pelvis Bone, 14 - Wet Rib Bone, 15 - Bone Cartilage, 16 - Abdom. Artery, 17 - Abdom. Vein, 18 - Bladder, 19 – Prostate, 20 - Asc. large int., 21 - Asc. large int. contents), 22 - Desc. large int. 23 -Desc. large int. (contents), 24 –Rectum, 25 - Seminal Vessicles, 26 - Vas deferens, 27 – Testes)

75

Figure 3-5: PET and corrected PET values plotted against true values for 800 simulations.

76

Sphere diameter / FWHM resolution = 1 / 0.5

Uncorrected Corrected

Uncorrected Corrected

Figure 3-6: Signal recovery as a function of volume (top) and threshold (bottom), for a small lesion diameter relative to convolution kernel size.

77

Sphere diameter / FWHM resolution = 3 / 0.5

Uncorrected Corrected

Uncorrected Corrected

Figure 3-7: Signal recovery as a function of volume (top) and threshold (bottom), for a medium lesion diameter relative to convolution kernel size.

78

Sphere diameter / FWHM resolution = 5 / 0.5

Uncorrected Corrected

Uncorrected Corrected

Figure 3-8: Signal recovery as a function of volume (top) and threshold (bottom), for a large lesion diameter relative to convolution kernel size.

79

Figure 3-9: Simulated lesion profiles (small to medium sized). Left – true (black), uncorrected PET simulation (red), and corrected PET simulation (blue). Right – profile of concentric ROIs used in deconvolution process, separate levels represent separate ROIs.

80

Figure 3-10: Simulated lesion profiles (medium to large sized). Left – true (black), uncorrected PET simulation (red), and corrected PET simulation (blue). Right – profile of concentric ROIs used in deconvolution process, separate levels represent separate ROIs.

81

Figure 3-11: Corrected sphere measurements in a Jaszczak phantom PET image derived from different blur kernels. Spheres 0-5 correspond to sphere approximate volumes of 16, 8, 4, 2, 1, and 0.5 ml. Top row: ROI max values, bottom row: ROI mean values. Corresponding data shown in Table 3-2.

82

Tables

Table 3-1: Summarized measurements of true, uncorrected, and corrected PET lesion simulations

Test

Pixel Size

Resolution (kernel FWHM)

Target Sphere Diameter

FWHM ([pixels] -from central axis profile)

Mean Value (mean of voxels above 50% max)

Center Value

True

Uncorr

Corr

True

Uncorr

Corr

True

Uncorr

Corr

1

0.15

0.5

0.5

6.25

7.18

2.65

5.00

2.70

0.00

5.00

2.70

5.07

2

0.15

0.5

0.75

6.25

7.16

6.25

5.00

3.94

5.00

5.00

2.90

5.00

3

0.15

0.5

1.0

8.25

8.73

8.25

5.00

4.79

5.00

5.00

3.36

5.00

4

0.15

0.5

1.25

10.25

10.58

10.25

5.00

4.96

5.00

5.00

3.61

5.00

5

0.15

0.5

1.5

12.25

12.57

12.25

5.00

5.00

5.00

5.00

3.83

5.00

6

0.15

0.5

2.0

16.25

16.00

16.25

5.00

5.00

5.00

5.00

3.94

5.00

7

0.15

0.5

2.5

18.25

18.84

18.45

5.00

5.00

4.99

5.00

4.15

4.77

8

0.15

0.5

3.0

22.25

22.80

22.62

5.00

5.00

5.00

5.00

4.25

4.80

9

0.15

0.5

4.0

28.25

28.88

28.54

5.00

5.00

5.00

5.00

4.41

4.94

10

0.15

0.5

5.0

36.25

36.59

36.53

5.00

5.00

5.00

5.00

4.50

4.82

11

0.15

0.5

6.0

42.25

42.88

42.37

5.00

5.00

5.00

5.00

4.57

4.86

12

0.15

0.5

10.0

68.25

68.89

68.34

5.00

5.00

5.00

5.00

4.73

4.98

83

Table 3-2: Corrected sphere measurements for Jaszczak PET scan using different estimations of blur kernels FWHM [mm]. All values were normalized using the mean of the center ROI of the largest sphere in the uncorrected scan. (Best value at kernel FWHM = 5.88,5.25, shown in bold and underlined). Mean values are defined as the mean of a region grown to 90% of the maximum value in the PET scans.

FWHM

Sphere (normalized maximum values)

Sphere (normalized mean values)

X,Y

Z

1

2

3

4

5

6

1

2

3

4

5

6

2.45 2.45 2.45 2.45 2.45 2.45 2.45 2.45 2.45 2.45 2.45 2.94 2.94 2.94 2.94 2.94 2.94 2.94 2.94 2.94 2.94 2.94 3.43 3.43 3.43 3.43 3.43 3.43 3.43 3.43 3.43 3.43 3.43 3.92 3.92 3.92

1.75 2.10 2.45 2.80 3.15 3.50 3.85 4.20 4.55 4.90 5.25 1.75 2.10 2.45 2.80 3.15 3.50 3.85 4.20 4.55 4.90 5.25 1.75 2.10 2.45 2.80 3.15 3.50 3.85 4.20 4.55 4.90 5.25 1.75 2.10 2.45

1.01 1.01 1.01 1.01 1.01 1.01 1.01 1.02 1.02 1.02 1.02 1.01 1.01 1.01 1.01 1.01 1.01 1.02 1.02 1.02 1.02 1.02 1.01 1.01 1.01 1.01 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02

1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.03 1.03 1.03 1.03 1.02 1.02 1.02 1.02 1.02 1.03 1.03 1.03 1.03 1.03 1.03 1.02 1.02 1.02 1.03 1.03 1.03 1.03 1.03 1.03 1.03 1.04 1.02 1.03 1.03

0.96 0.96 0.96 0.97 0.97 0.97 0.97 0.97 0.98 0.98 0.98 0.96 0.97 0.97 0.97 0.97 0.97 0.98 0.98 0.98 0.98 0.98 0.97 0.97 0.97 0.97 0.98 0.98 0.98 0.98 0.98 0.98 0.98 0.97 0.97 0.98

0.92 0.92 0.92 0.92 0.93 0.93 0.94 0.94 0.94 0.95 0.95 0.92 0.93 0.93 0.93 0.94 0.94 0.94 0.95 0.95 0.96 0.96 0.93 0.93 0.94 0.94 0.94 0.95 0.95 0.96 0.96 0.97 0.97 0.94 0.94 0.95

0.76 0.76 0.76 0.77 0.77 0.78 0.78 0.78 0.79 0.80 0.81 0.77 0.77 0.78 0.78 0.78 0.79 0.79 0.79 0.80 0.81 0.82 0.78 0.78 0.79 0.79 0.79 0.80 0.80 0.81 0.81 0.82 0.84 0.79 0.79 0.80

0.58 0.58 0.58 0.59 0.59 0.60 0.61 0.61 0.62 0.63 0.64 0.58 0.59 0.59 0.60 0.61 0.62 0.63 0.64 0.65 0.67 0.68 0.59 0.59 0.60 0.61 0.62 0.64 0.65 0.67 0.69 0.71 0.73 0.60 0.60 0.61

1.01 1.01 1.01 1.01 1.01 1.01 1.01 1.02 1.02 1.02 1.02 1.01 1.01 1.01 1.01 1.01 1.01 1.02 1.02 1.02 1.02 1.02 1.01 1.01 1.01 1.01 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02

1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.03 1.03 1.03 1.03 1.02 1.02 1.02 1.02 1.02 1.03 1.03 1.03 1.03 1.03 1.03 1.02 1.02 1.02 1.03 1.03 1.03 1.03 1.03 1.03 1.03 1.04 1.02 1.03 1.03

0.96 0.96 0.96 0.97 0.97 0.97 0.97 0.97 0.98 0.98 0.98 0.96 0.97 0.97 0.97 0.97 0.97 0.98 0.98 0.98 0.98 0.98 0.97 0.97 0.97 0.97 0.98 0.98 0.98 0.98 0.98 0.98 0.98 0.97 0.97 0.98

0.92 0.92 0.92 0.92 0.93 0.93 0.94 0.94 0.94 0.95 0.95 0.92 0.93 0.93 0.93 0.94 0.94 0.94 0.95 0.95 0.96 0.96 0.93 0.93 0.94 0.94 0.94 0.95 0.95 0.96 0.96 0.97 0.97 0.94 0.94 0.95

0.76 0.76 0.76 0.77 0.77 0.78 0.78 0.78 0.79 0.80 0.81 0.77 0.77 0.78 0.78 0.78 0.79 0.79 0.79 0.80 0.81 0.82 0.78 0.78 0.79 0.79 0.79 0.80 0.80 0.81 0.81 0.82 0.84 0.79 0.79 0.80

0.58 0.58 0.58 0.59 0.59 0.60 0.61 0.61 0.62 0.63 0.64 0.58 0.59 0.59 0.60 0.61 0.62 0.63 0.64 0.65 0.67 0.68 0.59 0.59 0.60 0.61 0.62 0.64 0.65 0.67 0.69 0.71 0.73 0.60 0.60 0.61

84

3.92 3.92 3.92 3.92 3.92 3.92 3.92 3.92 4.41 4.41 4.41 4.41 4.41 4.41 4.41 4.41 4.41 4.41 4.41 4.90 4.90 4.90 4.90 4.90 4.90 4.90 4.90 4.90 4.90 4.90 5.39 5.39 5.39 5.39 5.39 5.39 5.39 5.39 5.39 5.39 5.39 5.88 5.88 5.88 5.88 5.88 5.88 5.88

2.80 3.15 3.50 3.85 4.20 4.55 4.90 5.25 1.75 2.10 2.45 2.80 3.15 3.50 3.85 4.20 4.55 4.90 5.25 1.75 2.10 2.45 2.80 3.15 3.50 3.85 4.20 4.55 4.90 5.25 1.75 2.10 2.45 2.80 3.15 3.50 3.85 4.20 4.55 4.90 5.25 1.75 2.10 2.45 2.80 3.15 3.50 3.85

1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02

1.03 1.03 1.03 1.03 1.03 1.03 1.04 1.04 1.03 1.03 1.03 1.03 1.03 1.03 1.03 1.04 1.04 1.04 1.04 1.03 1.03 1.03 1.03 1.03 1.04 1.04 1.04 1.04 1.04 1.04 1.04 1.04 1.04 1.04 1.04 1.04 1.04 1.04 1.04 1.04 1.04 1.04 1.04 1.04 1.04 1.04 1.04 1.04

0.98 0.98 0.98 0.98 0.98 0.99 0.99 0.99 0.98 0.98 0.98 0.98 0.99 0.99 0.99 0.99 0.99 0.99 1.00 0.99 0.99 0.99 0.99 0.99 0.99 0.99 1.00 1.00 1.00 1.00 0.99 0.99 0.99 1.00 1.00 1.00 1.00 1.00 1.00 1.01 1.01 1.00 1.00 1.00 1.00 1.00 1.00 1.01

0.95 0.95 0.96 0.96 0.96 0.97 0.98 0.99 0.95 0.95 0.96 0.96 0.96 0.97 0.97 0.97 0.98 0.99 1.00 0.96 0.97 0.97 0.97 0.97 0.97 0.98 0.98 0.99 0.99 1.00 0.98 0.98 0.98 0.98 0.98 0.99 0.99 0.99 0.99 0.99 1.00 0.99 0.99 1.00 1.00 1.00 1.00 1.00

0.80 0.81 0.81 0.81 0.82 0.83 0.84 0.87 0.80 0.80 0.81 0.81 0.82 0.82 0.83 0.84 0.85 0.87 0.90 0.81 0.81 0.82 0.82 0.83 0.83 0.84 0.85 0.86 0.89 0.92 0.90 0.91 0.91 0.92 0.93 0.94 0.95 0.94 0.93 0.90 0.94 1.04 1.04 1.05 1.07 1.09 1.10 1.11

0.62 0.64 0.66 0.68 0.71 0.74 0.76 0.79 0.60 0.61 0.62 0.63 0.65 0.68 0.71 0.74 0.78 0.82 0.85 0.64 0.64 0.64 0.64 0.66 0.69 0.73 0.77 0.82 0.86 0.90 0.72 0.72 0.72 0.71 0.69 0.70 0.75 0.79 0.84 0.89 0.93 0.84 0.85 0.84 0.83 0.81 0.78 0.74

85

1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02

1.03 1.03 1.03 1.03 1.03 1.03 1.04 1.04 1.03 1.03 1.03 1.03 1.03 1.03 1.03 1.04 1.04 1.04 1.04 1.03 1.03 1.03 1.03 1.03 1.04 1.04 1.04 1.04 1.04 1.04 1.04 1.04 1.04 1.04 1.04 1.04 1.04 1.04 1.04 1.04 1.04 1.04 1.04 1.04 1.04 1.04 1.04 1.04

0.98 0.98 0.98 0.98 0.98 0.99 0.99 0.99 0.98 0.98 0.98 0.98 0.99 0.99 0.99 0.99 0.99 0.99 1.00 0.99 0.99 0.99 0.99 0.99 0.99 0.99 1.00 1.00 1.00 1.00 0.99 0.99 0.99 1.00 1.00 1.00 1.00 1.00 1.00 1.01 1.01 1.00 1.00 1.00 1.00 1.00 1.00 1.01

0.95 0.95 0.96 0.96 0.96 0.97 0.98 0.99 0.95 0.95 0.96 0.96 0.96 0.97 0.97 0.97 0.98 0.99 1.00 0.96 0.97 0.97 0.97 0.97 0.97 0.98 0.98 0.99 0.99 1.00 0.98 0.98 0.98 0.98 0.98 0.99 0.99 0.99 0.99 0.99 1.00 0.99 0.99 1.00 1.00 1.00 1.00 1.00

0.80 0.81 0.81 0.81 0.82 0.83 0.84 0.87 0.80 0.80 0.81 0.81 0.82 0.82 0.83 0.84 0.85 0.87 0.90 0.81 0.81 0.82 0.82 0.83 0.83 0.84 0.85 0.86 0.89 0.92 0.81 0.81 0.82 0.82 0.83 0.83 0.84 0.85 0.87 0.90 0.94 0.81 0.81 0.81 0.82 0.82 0.83 0.83

0.62 0.64 0.66 0.68 0.71 0.74 0.76 0.79 0.60 0.61 0.62 0.63 0.65 0.68 0.71 0.74 0.78 0.82 0.85 0.61 0.61 0.62 0.64 0.66 0.69 0.73 0.77 0.82 0.86 0.90 0.60 0.61 0.62 0.64 0.67 0.70 0.75 0.79 0.84 0.89 0.93 0.59 0.60 0.61 0.63 0.66 0.70 0.74

5.88 5.88 5.88 5.88 6.37 6.37 6.37 6.37 6.37 6.37 6.37 6.37 6.37 6.37 6.37 6.86 6.86 6.86 6.86 6.86 6.86 6.86 6.86 6.86 6.86 6.86 7.35 7.35 7.35 7.35 7.35 7.35 7.35 7.35 7.35 7.35 7.35

4.20 4.55 4.90 5.25 1.75 2.10 2.45 2.80 3.15 3.50 3.85 4.20 4.55 4.90 5.25 1.75 2.10 2.45 2.80 3.15 3.50 3.85 4.20 4.55 4.90 5.25 1.75 2.10 2.45 2.80 3.15 3.50 3.85 4.20 4.55 4.90 5.25

1.02 1.02 1.02 1.02 1.03 1.03 1.03 1.03 1.03 1.03 1.03 1.02 1.02 1.02 1.02 1.03 1.03 1.03 1.03 1.03 1.03 1.03 1.03 1.03 1.03 1.02 1.03 1.03 1.03 1.03 1.03 1.03 1.03 1.04 1.04 1.04 1.04

1.04 1.04 1.04 1.04 1.04 1.04 1.05 1.05 1.05 1.04 1.04 1.04 1.04 1.04 1.10 1.05 1.05 1.05 1.05 1.05 1.05 1.05 1.05 1.05 1.11 1.14 1.06 1.06 1.06 1.06 1.05 1.05 1.05 1.05 1.11 1.15 1.19

1.01 1.01 1.01 1.02 1.01 1.01 1.01 1.01 1.01 1.01 1.01 1.01 1.02 1.02 1.03 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.03 1.03 1.03 1.03 1.03 1.03 1.03 1.03 1.03 1.03 1.03 1.03 1.03

1.00 0.99 0.99 0.99 1.01 1.01 1.01 1.02 1.02 1.02 1.01 1.01 1.01 1.03 1.04 1.03 1.03 1.04 1.04 1.04 1.04 1.03 1.04 1.07 1.11 1.14 1.06 1.06 1.06 1.06 1.06 1.06 1.06 1.09 1.14 1.20 1.25

1.11 1.09 1.05 0.98 1.24 1.24 1.26 1.28 1.31 1.33 1.35 1.36 1.34 1.29 1.21 1.52 1.53 1.56 1.59 1.63 1.67 1.70 1.71 1.70 1.65 1.56 1.00 1.01 1.02 1.05 1.08 1.14 1.20 1.29 1.41 1.57 1.76

0.79 0.84 0.88 0.92 1.01 1.02 1.02 1.02 1.01 0.97 1.01 1.05 1.08 1.11 1.12 0.93 0.94 0.97 1.01 1.05 1.09 1.13 1.16 1.18 1.19 1.17 1.04 1.06 1.10 1.15 1.20 1.45 1.52 1.56 1.58 1.59 1.32

86

1.02 1.02 1.02 1.02 1.03 1.03 1.03 1.03 1.03 1.03 1.03 1.02 1.02 1.02 1.02 1.03 1.03 1.03 1.03 1.03 1.03 1.03 1.03 1.03 1.03 1.02 1.03 1.03 1.03 1.03 1.03 1.03 1.03 1.03 1.03 1.03 1.03

1.04 1.04 1.04 1.04 1.04 1.04 1.05 1.05 1.05 1.04 1.04 1.04 1.04 1.04 1.03 1.05 1.05 1.05 1.05 1.05 1.05 1.05 1.05 1.05 1.04 1.03 1.06 1.06 1.06 1.06 1.05 1.05 1.05 1.05 1.04 1.04 1.04

1.01 1.01 1.01 1.02 1.01 1.01 1.01 1.01 1.01 1.01 1.01 1.01 1.02 1.02 1.03 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.02 1.03 1.03 1.03 1.03 1.03 1.03 1.03 1.03 1.03 1.03 1.03 1.03 1.03

1.00 0.99 0.99 0.99 1.01 1.01 1.01 1.02 1.02 1.02 1.01 1.01 1.00 0.99 0.98 1.03 1.03 1.04 1.04 1.04 1.04 1.03 1.02 1.01 0.99 0.98 1.06 1.06 1.06 1.06 1.06 1.06 1.06 1.04 1.03 1.00 0.98

0.84 0.86 0.90 0.95 0.79 0.79 0.79 0.80 0.80 0.80 0.81 0.81 0.83 0.87 0.94 0.76 0.76 0.76 0.76 0.76 0.76 0.76 0.77 0.78 0.83 0.91 1.00 1.01 1.02 1.05 1.08 1.14 1.20 1.29 1.41 1.57 1.76

0.79 0.84 0.88 0.92 0.57 0.57 0.58 0.60 0.62 0.97 1.01 1.05 1.08 1.11 1.12 0.93 0.94 0.97 1.01 1.05 1.09 1.13 1.16 1.18 1.19 1.17 1.04 1.06 1.10 1.15 1.20 1.39 1.42 1.42 1.40 1.34 1.24

4. Respiratory Gated PET Derived in A FullyAutomated Manner From Raw PET Data

Abstract Respiratory motion in PET degrades images and limits detectability of small or lowcontrast lesions. Although image quality can be improved using respiratory-gating hardware, this adds to the complexity and expense of acquiring PET data. We aimed to develop a data-driven method, based on individual voxel signal fluctuations, for accomplishing electronic respiratory gating of PET data acquired in a clinically practical manner, requiring no additional hardware or end-user input. We tested our methods using both simulated PET scans and actual human PET acquisitions. For the simulations, our methods correctly identified the start frame of each respiratory cycle defined for the phantom. Resultant gated images demonstrated improved effective resolution and increased SUV uptake for lesions scattered in the thorax. For human PET data, we were able to recover respiratory phase information with a large signal-to-noise ratio. Fully automated voxel-based respiratory gating of PET images may be achieved without the need for gating hardware or additional user input, in a manner capable of improving effective resolution and increasing lesion detectability.

87

Introduction A large source of image degradation in thoracic PET images can be attributed to respiratory motion, which causes blurring of the torso region (over the multi minute scan acquisition). This blurring can be difficult to characterize, and thus has the effect to limit detectability, particularly of small lesions or lesions with low tracer uptake, as well as to diminish accuracy of the measurements for the lesions which are visible.

Respiratory gating in PET is an approach to lessen the image degradation from respiratory motion by separating the breathing cycle into different phases and generating images from data corresponding to respiratory phases. In the past few years there has been much research in developing this type of imaging, with the hope that this can increase the quality of diagnosis derived from PET imaging. The consensus in literature is that the respiratory gating of PET images presents a feasible solution to the image degradation introduced by respiratory motion. Researchers have studied the use of respiratory gated PET with respect to improved quantification (59-61), lesion detectability and artifacts (62-67), image-coregistration accuracy (68), and the use of gated PET/CT in radiotherapy treatment planning(69, 70). A variety of methodology has been presented in the above literature for triggering of respiratory gates including techniques utilizing cameras, pressure belts, thermometers, point sources, pneumatic sensor systems, and mechanical ventilation (in dogs).

88

Our methodology is inspired by the notion that there is much information in raw PET listmode data which is not being utilized. When a PET image is reconstructed by integrating the acquired histogram data over several minutes (ungated PET), there are many details about what events happened when and where that are being discarded, while those data contain information about respiratory motion. The question we are trying to address, the same question the handful of other papers on data driven gating have tried to address, is how do we sort out the respiratory signal from the relatively large amount of noise?

The approach we are presenting here begins by taking the PET scanner’s listmode output, and sorting out the repetitive cycles contained within it. In the reconstructed image voxels alone, noise and sensitivity are largely limiting factors for finding cyclic information. To get around this we are looking at many voxels, and filtering out the frequencies in the signal which are less relevant. There are over a million voxels inside a normal PET scan field of view, potentially resulting in a million different phase sensors. Each one outputs a signal relating a level of activity in a small volume in space. Rather than using previously presented image-based methods (71, 72) of characterizing the movement of structures to gate the scans, we look directly at the fluctuation of signal within each voxel, caused in part by respiratory movement, and then combine the information into a global respiratory signal.

89

Methods To obtain gated scans, we begin by evaluating the time activity patterns in individual voxels (Figure 4-1). To get time-activity information from individual voxels, scan listmode data is resorted and reconstructed into images corresponding to short time bins (short relative to the respiratory cycle - we used 0.5 seconds). The specific parameters for reconstruction are discussed below.

Before we begin looking at the voxels, we must first define the order in which we will process them. We do this for two reasons: the global signal (discussed below) is cumulative and will grow more accurately and quicker when the voxels with the most useful information are used first, and secondly, to sort out which of the millions of voxels are useful to look at all. To do this, every voxel within the scan volume is scored by its activity gradient in the cranial-caudal direction on the summed reconstructed image (Figure 4-2), where we detect rapid changes in activity distribution throughout the respiratory cycle. The absolute value of this score is used as a priority value for the voxel processing order.

Once the order is defined for processing the voxels, the algorithm begins to look at voxels individually and combines information from them to create a global signal trace. The global signal trace is a summation of individual voxel TACS, and summarizes the estimate of respiratory motion.

90

The processing of individual voxels and combining them with the global signal trace consists of several steps: 1) Non-respiratory frequencies (periods <2 seconds and >9 seconds) are filtered out of the voxel time activity curve (TAC) 2) Three possible new global traces are presented (A) global trace (unchanged) (B) global + voxel TAC (C) global - voxel TAC 3) Of the three, the scenario with the highest standard deviation (normally B or C) is chosen to be the new global signal trace (i.e. the trace with the greatest difference between peaks and valleys). These steps for processing voxels are summarized in Figure 4-3.

The purpose of step two is to determine the best contribution an individual voxel can make to the global signal trace. Scenarios using addition and subtraction are included to account for the fact that voxels may be in or out of phase with the global signal trace, depending on whether they lie superior or inferior to gradients of motion.

To initiate the processing, the global signal trace is defined as the TAC of the voxel slated to be evaluated first. With each iteration of this process and for each new voxel processed, the global signal trace either remains the same, or is improved (case example in Figure 4-4).

91

This process can be repeated for hundreds to thousands of voxels. Final phase information can be extracted from the timing of the peaks and dips in the global signal trace.

We tested these methods using both simulated data and actual clinical data.

Simulations: PET scan simulations were generated based on the 4D NCAT torso phantom. The NCAT phantom was used to generate eight 0.5 second frames, altered to represent respiratory motion during a 4 second breathing cycle. The steps used in scan simulation are as follows: Emission volume for typical FDG scan -> voxels scaled to clinical size

(x ,y, and z = 0.533385, 0.533385, and 0.2 cm,

respectively) -> all activity scaled to 370 MBq (sensitivity = 0.005, % in FOV = 50%) -> forward projection (2D) -> attenuation -> blurring with 0.9cm blurring kernel -> Poisson noise added -> correction for attenuation -> image backprojection (using FBP)

92

Attenuation maps derived from individual gates were used for both attenuation and attenuation correction. The ungated image was defined as all the gated images summed together.

The scan simulation was re-simulated in back to back breathing cycles, to represent a 5 minute scan acquisition (with and without uniform periodicity). One simulation was generated with the phantom as defined above, and another one was generated with the addition of six lesions placed within the lungs at scattered locations, each with a diameter of 10 mm. The first simulation was used as the case to test our methods in a relatively more robust manner, while the second (periodically uniform) phantom with the lesions was used to measure the effects and benefit of the gating.

The data were gated as described in the above methods, and a global signal trace was acquired.

PET Acquisitions: Raw list mode data were acquired for twenty-four patients, all with possible lung lesions, following a routine clinical FDG PET protocol (320-507MBq, Biograph Sensation 16 scanner (Siemens Medical Solutions (72)).

Patient scan data were separated into 0.5 second time frames for a total scan time of 10 minutes (mean counts per frame ± standard deviation: 173,000 ± 35,000 prompts, 78,000 ± 20,000 randoms, n=24 patients). Images were reconstructed using 3D OSEM

93

reconstruction (2 iterations, 8 subsets) into a 128x128x47 matrix (voxel x, y, and z size = 5.30673 x5.30673 x 3.375 mm, respectively).. No attenuation correction was used because corrected images are not necessary for our algorithm, and can disturb the algorithm as attenuation maps derived from non-gated CT images (standard) are aligned with some phases and not others.

While being scanned, patients wore an Anzai pressure belt (Anzai Medical Co., LTD, Japan) to monitor respiratory motion. This signal was also used as an external comparison for our image derived signal.

As a way to quantitatively measure the correlation between the nine acquired dataderived traces, and the associated belt-derived traces we set up a scoring method as defined in Equation 4-1:

SCORE

=

∑ (D

∑ (D

image

∗ D belt

image

∗ D belt

time

)

)

time

Equation 4-1 Dimage = slope of image-derived global signal trace Dbelt = slope of belt-derived signal trace

94

The slopes of the traces are defined as the change in the trace signal (at every discrete time point), and were used as a quick and standard way to get rid of baseline shifts present in the belt signal, while preserving frequency and amplitude information.

This scoring method takes advantage of the fact that constructive interference of two waves can be attributed to similar frequency makeup. Using this scoring method, two traces which combine 100% constructively when multiplied by each other, i.e. very similar waves, would receive a score of 1, while traces which have nothing to do with each other or are out of phase will be scored very close to zero.

As an assurance of alignment between the two traces, correlation was calculated at all possible offsets. The correct offset was chosen to be the offset with the highest correlation, reasoning better correlation should not occur randomly.

In addition to the correlative score mentioned above, efforts were made to characterize the accuracy of the image-based gating methods directly from image-based traces. We created a quality factor to help users predict the usefulness of the output signal. The quality factor is a frequency-based measurement that determines the relative amount of ordered and random frequencies present in the signal trace.

The image-based traces were used to create respective short-term Fourier transform (STFT) spectrograms. More expressly, at each discrete point in time (time bins = 0.5

95

seconds) a 60 second frequency-power spectrum was determined using the fast Fourier transform, thus contributing a line to the spectrogram. Once completed, the spectrogram presents the relative amounts of peak frequencies and background frequencies in the signal. Analogously, the former could be seen as relative amounts of ordered and random contributions to the signal. These relative amounts can be summarized in the mean of the spectrogram, as a percentage, and that was how we characterized the quality factor of the traces. Example spectrograms are shown in Figure 4-6.

Although the purpose of this paper is centered on identifying the respiratory traces, we extended our methods for one of our scans to sorting the listmode data into gated bins and reconstructing gated images, to illustrate the potential application for our methods. Respiratory cycle triggers were defined by the image based respiratory traces at local maxima (±1 second), and the timing of the gates were derived by dividing the time between respiratory triggers into 8 bins. Gated images were reconstructed using OSEM (2 iterations, 8 subsets) reconstruction.

Results Sample voxel time activity curves can be seen in Figure 4-7, illustrating the levels of noise present in these signals.

96

For the simulations we acquired respiratory traces (Figure 4-8) measured using our image-based methods, which exhibited the expected respiratory frequency (period = 4 seconds for the periodic simulation).

The start frame of each respiratory cycle could be identified by the peaks in the respiratory trace. The gated images demonstrated improved resolution and organ boundaries, and allowed for increased detectability of lesions in the lungs. Regions of interest (ROI’s) were drawn over the lesions in the reconstructed images using the known true location of the simulated lesions. Using the gate closest to full expiration, the average activity concentration measured in the six lesions increased 60% in the gated vs. ungated images. The improvements in the gated image are illustrated in Figure 4-9 and summarized in Table 4-1.

For our clinical scans, our traces were derived from images acquired over 500 ms windows. An example sinogram and image used in the construction of the traces are shown in Figure 4-10 and Figure 4-11 , respectively. From the clinical data, we were able to extract image derived respiratory signals and compare them to hardware derived signals (example in Figure 4-12).

A summary of the correlative scores (from Equation 4-1) for the image-derived and beltderived respiratory traces for the twenty four human scans we evaluated can be seen in Table 4-2.

97

The absolute Pearson correlation coefficient between our image-derived trace derived quality factor and the scores was 0.83, and is illustrated in Figure 4-13. The individual quality factors are presented in Table 4-2.

The reconstructed human gated PET images are displayed in Figure 4-14, and illustrate lesion motion captured from image-based respiratory signal.

Discussion For our set of 24 scans, 19 scored relatively well (>.5), while 5 of the scans appear to have scored poorly (<.25) (Table 4-2).

Example respiratory traces derived using both image-based and hardware-based methods are shown in Figure 4-5. The figures illustrate qualitative correlation between the data sets. In the figure, we can see examples of both image-derived traces and pressure beltderived traces preserving periodic signal, relative amplitudes in the signal, and irregularities in the respiratory patters.

We speculate that there are three scenarios that led to a decrease in the assigned scoring presented above. Firstly, in this analysis we used the belt derived traces as a gold standard, while they themselves are prone to error. On closer inspection of the 24 belt derived traces, there is an abundance of unexpected non-sinusoidal motion. This may be

98

resultant from error in patient setup, hardware malfunction, or just the pressure belt concept/design. Secondly, we noticed one patient had many respiratory periods falling outside the 2-9 second respiratory window we were filtering the data for, having respiratory cycles lasting as long as 12 seconds during their scan. For this patient, their score went from 0.16 to 0.76 (quality factor 13 to 6.5) when the frequency filter window was changed to 4-15 seconds. We choose the window a priori to encompass most respiratory frequencies. Thirdly the reason some scans may not score well is that the patients may not make good candidates for respiratory gating no matter what method used to capture their respiratory trace, as has been observed before (18). Patients receiving PET scans often have respiratory problems (e.g. thoracic disease, lung cancer patients, pneumothorax, etc.) which may cause them to breath shallowly, and irregularly. The patients we looked at who scored poorly all had “messy” belt-derived traces, with a lot of non-sinosoidal deviations.

In this work we described a quality factor, which is derived directly from the acquired image-based respiratory trace, and we showed there is a correlation between this quality factor and the correlative scoring factor. Having such a measure available allows users of these methods to gain relative confidence when they acquire an image based respiratory trace, which may aid in the decision of weather or not to use the gated PET images clinically for a specific scan acquisition.

99

There are several key factors used in the algorithm presented here that have an important impact on the final image-derived respiratory trace quality.

The global signal trace is made up from combining information from many voxels. When combining our data, we used the measure of standard deviation to evaluate the improvement each voxel contributed to the global signal, and when found beneficial, added the voxels information to the global signal. This played an important role in deciding which information to use and which information to discard, and how we are using it. It should be noted that these are possible points of improvement for the algorithm. Additionally, since the global signal trace is made in a cumulative manner, the order with which we decided to process the voxels does make some difference in the final result. In the methods we presented here we prioritized voxels by their gradient in the cranio-caudal dimension.

Another possible point of improvement is the manner in which unwanted frequencies are filtered out of the voxel TACs. We have observed that the tighter the frequency window, the nicer the final global signal trace appears, along with an increase in the speed with which it is built up. In the work here we are setting the window a priori allow periods of 2-9 seconds, which appeared to work for most of our scans, but not all of them. Currently, the algorithm is fully-automated, which we believe is one of its valuable features. This frequency filter however will be examined for further methodological

100

improvement with the possibility of incorporating the quality factor in iterative cycles for choosing a more appropriate/scan specific frequency window.

If one is willing to set aside the advantages of using our methods for a fully-automated approach, our image based methodology could be used to compliment existing methodology already in commercial use. Patients can have double traces acquired on their scans, covering each other if one fails and providing quality assurance and measurements of dependability on a system or on a scan by scan basis. Similarly they can be used in concert; specifically the belt can be used to define the frequency window for filtration in the image based methods and the image derived trace can be used to define the respiratory phases.

Another advantage of using image based methods which should be noted, is the assurance of the temporal alignment of the respiratory trace and the PET data. Human or software errors can manifest as misalignment of the data, as we have seen with our data acquisitions. Misalignment between the two data sets of as little as a few seconds could ruin potential benefits of the gating.

To generate time-activity curves, as is necessary for data-driven gating methods, many short-frame images need to be sorted out and reconstructed. This effects processing time. We have found that for a ten minute scan the total time to extract a respiratory trace is

101

around 12 hours on a single PC, as summarized in Table 4-3. This time could be decreased using a cluster of computers, along with methodological development.

There are evident advantages with having a fully-automated data-based algorithm for creating gated PET images. The algorithms are image based, and thus machine independent, and can be used with existing scans, or scanners. The images are generated without any extra effort or deviation from routine clinical procedures. They come at no “cost”, other than processing time, and can be generated along side traditional non-gated PET. This practicality for image generation can make gated images more available in both clinical and research settings, pushing PET gating forward in the industry.

This work is still in its early stages. There are still details which can be explored and tuned, such as investigating how each of the following parameters effects the results of our method: •

voxel size

•

reconstruction algorithm

•

temporal bin size

•

smoothing filters

There is also room for reformulating and refining the method with respect to: •

Weighting and statistical filtering for signal combination

•

Voxel processing order

•

Volumes used for signal analysis (could be ROI on sinogram or image)

102

•

Signal (could be statistic of activity concentration)

•

Addition of local confidence weighting

•

Methods of validation used to characterize the limitations of such

algorithms.

In addition to improvements in the algorithm, we can project that as PET technology improves (i.e. detector sensitivity and resolution) our algorithms will be able to exploit these improvements, thus expanding the advantage gap between this software based methodology and more common hardware based gating methods.

Conclusion We have presented methods to characterize human PET scans for respiratory motion, using both simulations and actual human PET data. There have been many methods proposed to reduce motion artifacts by “gating” the acquisition of images, most commonly based on an external measure of motion. Our methodology employs a fully automated software algorithm which is image based, requires no external equipment or hardware, and presents a practical method for characterization of respiratory movement in PET. The methods could potentially be applied to existing PET data, and used with existing PET scanners. Similar methods could also be extended to improve imaging modalities other than PET, and periodic motion artifacts caused by processes other than respiration (e.g. cardiac).

103

Figures

Coronal Projection

Sagittal Projection

Amplitude

Trigger

Time Figure 4-1: Simulated time-activity curve analysis for single voxel (noiseless simulation). (Top left) volume projections, (Top right) sample voxel, with grey scale corresponding to activity in scan, (bottom) time-activity curve of voxel.

104

Coronal

Sagittal

PET (summed image)

Activity gradient on z-axis

Combined image Figure 4-2: Illustration of voxel processing prioritization. (Top) – PET slice, (Middle) – Voxel prioritization, (Bottom) – PET/prioritization co-registered.

105

Set global respiratory trace to equal first voxel TAC to be processed

Evolving global respiratory trace

Choose scenario with greatest SD to be new global signal trace

global trace Global trace

global global trace + voxel TAC TAC + voxel

global global trace voxel TAC - -voxel TAC

SD = 1.0 SD = 1.0

SD = 1.3 SD = 1.3

SD = 0.8 SD = 0.8

Check if stopping criteria met

Yes (stop)

Create 3 possible scenarios

No (continue)

Filter out nonrespiratory frequencies

Return time global respiratory trace

Retrieve next voxel TAC for processing

Figure 4-3: Flow chart summarizing main steps in image processing loop, illustrated with example curves. With each new voxel processed, the global signal trace is updated. (SD = standard deviation).

106

fequency filtered voxel TAC

original global trace

(A) original global trace

SD = 1.0

(B) original global trace + voxel TAC

SD = 1.2

(C) original global trace - voxel TAC

SD = 0.7

Max standard deviation: 1.2 (Scenario B) Figure 4-4: Example of processing procedure for individual voxel TAC. The steps for processing a voxel TAC are illustrated above. The process begins with the original global signal trace, and the frequency-filtered voxel TAC (for the voxel being processed)(Top). Next, three possible final scenarios are created and evaluated (Middle) using no change, addition, or subtraction. In the last step of the process, the scenario with the greatest standard deviation is chosen to be the new global signal trace (Bottom).At this point the process is complete and the algorithm may move on to process another voxel.

107

Respiratory Traces Example A 16.0

Relative Amplitude

14.0 12.0 10.0 8.0 6.0 4.0 2.0 0.0 500

510

520

530

540

550

Time (s)

Example B 14.0

Relative Amplitude

12.0

10.0

8.0

6.0

4.0

2.0

0.0 515

525

535

545

555

565

575

585

Time (s)

Example C

Relative Amplitude

14.0

12.0

10.0

8.0

6.0

4.0

2.0

0.0 300

310

320

330

340

350

360

370

380

390

400

Time (s)

Image-Derived (top)

Belt-Derived (bottom)

Figure 4-5: Examples of global respiratory signals derived using our image-based methods, shown along corresponding pressure belt measurements. Example A illustrates case with good correlation between belt and image based signals. Example B illustrates some situation where we see conservation of relative amplitude (in signal over time) for signals derived using both methods. Example C illustrated behavior of the image-derived signal during periods of non-cyclical motion (as measured by the belt).

108

Frequency

Quality factor = 5%

Quality factor = 11%

Time Figure 4-6: Example short-term Fourier transform (STFT) spectrograms. Top and bottom images illustrate good and bad patient samples, respectively. Images are in grey scale, scaled from min to max, and that the images have been stretched horizontally to fit within the paper.

109

Noiseless Simulation

Simulation with Noise

Human Scan

0

10

20

30

Time (s) Figure 4-7: Single voxel time activity curve: theoretical, simulation, actual. Time activity curves corresponding to first voxel process for each method.

110

40

50

Data Derived Respiratory Trace for Simulation 200 150

Relative Amplitude

100 50 0 0

2

4

6

8

10

12

14

16

18

20

-50 -100 -150 -200 Time (s)

Relative Amplitude

Data Derived Respiratory Trace for Non Periodically Uniform Simulation

Data derived respiratory trace Phase zero of simulated cycle

100

120

140

160

180

200

Time (s)

220

240

Period Simulated Measured 5 5 6 6 9 8 10 10 8 9 7 7 5 6 6 5 6 6 9 8 10 10 8 9 7 7 5 6

Figure 4-8: Signal of combined global signal trace (n=500 voxels) derived from simulated data. Top – Periodically uniform motion simulation. Bottom – Non-periodically uniform signal. Line represents respiratory trace measured with software, the black dots represent the frequency of the simulated cycles. Respiratory cycles were simulated by interpolating a respiratory cycle to have different periods, back to back, and thus with instant changes in the speed of phase change. Measured periods (right) were derived from peak to peak distances on the trace.

111

Gated slice

Ungated slice

Single 0.5 second time frame

Volume projection

Figure 4-9: Simulated volume slices.

112

Sinogram Volume Projection

Sinogram Slice

Histogram of Counts in 500ms Sinogram 10000000 1000000 100000 10000 1000 100 10 1 0

1

2

Counts Figure 4-10: Sample sinogram for 500 ms acquisition, from a patient scan.

113

3

Image Volume Projection

Axial Image Slice

Histogram of Counts in 500ms Image 1000000 100000 10000 1000 100 10 1 0

7.7304E-06

1.5461E-05

Counts Figure 4-11: Sample image reconstructed from 500 ms acquisition, from a patient scan.

114

2.3191E-05

Image-Derived Belt-Derived

Respiratory Traces Relative Amplitude

140 120 100 80 60 40 20 0 100

110

120

130

140

150

Time (s) Figure 4-12: Signal of combined global signal derived using our image-based method trace (top), shown along side the corresponding pressure belt measurement (bottom) (scan #4).

115

Mean Percentage of Short-Term Fourier Transform Spectrogram 1.2

1

Score

0.8

0.6

.

0.4

0.2

0 0%

R=0.83 R=0.83

5%

10%

15%

20%

Quality Factor

Figure 4-13: Figure illustrates correlation between frequency analysis if the image derived respiratory trace, and the correlation score between the belt derived and image derived respiratory traces. The mean percentage of the short termed Fourier transform spectrogram is a measure of the non-random frequencies contained in the signal. The correlation implies information about the correctness of the fit is contained within the extracted image based signal.

116

Gated 1

2

3

4

5

6

7

8 6.8 cm

Ungated

100%

0% Figure 4-14: Example lesion movement on gated scan data, which was created using image based respiratory trace (scan #4).

117

Tables Table 4-1: Summary of lesion uptake measurements

Ratio

Tumor (10 Location

mm diameter

Mean (gated /

Displacement

SUV

sphere)

ungated)

Superior

1

1.50

Inferior

2

2.44

Posterior

Anterior Medial Lateral

3

4 5 6

1.63

2.51 1.95 2.01

118

gated

5.87

Ungated

4.00

gated

6.61

ungated

3.55

gated

4.99

ungated

3.75

gated

5.22

ungated

3.13

gated

5.67

ungated

3.49

gated

5.81

ungated

3.51

1.47 1.86 1.33 1.66 1.63 1.66

Table 4-2: Summary of trace scores. The score indicates a correlative measure of the belt-derived and image-derived respiratory traces. The quality factor is a measure derived solely from the image-derived respiratory traces.

Scan # Score

Score

Quality Factor

(with frequency

(image trace derived)

filtered belt signal)

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24

0.98 0.97 0.95 0.95 0.95 0.95 0.95 0.93 0.91 0.88 0.88 0.88 0.88 0.84 0.78 0.74 0.70 0.56 0.55 0.21 0.16 0.14 0.11 0.09

0.98 0.98 0.99 0.98 0.98 0.95 0.94 0.96 0.92 0.95 0.93 0.90 0.92 0.87 0.83 0.80 0.75 0.60 0.58 0.39 0.17 0.17 0.12 0.10

119

6.2 5.7 6.8 7.0 9.6 6.6 7.3 7.4 7.1 9.1 8.7 6.5 10.7 7.7 11.6 11.6 10.7 8.6 9.0 11.8 13.2 15.8 11.7 14.1

Table 4-3: Data processing time.

single 3.3 GHz PC Sinogram rebinning

10 frames / minute

OSEM reconstruction 2 frames / minute (2 iterations, 8 subsets) Voxel analysis

100 / minute

10 minute scan, 0.5 second frames (listmode -> respiratory trace) takes ~ 12 hours

120

5. Recombining Respiratory Gated PET Frames (method A) Introduction In contemporary PET imaging, a large source of image degradation can be attributed to respiratory motion in the thorax, and there have been several approaches developed to gate the images. Once the images are gated, there is a tradeoff in image quality between using gated images which have higher resolution and higher noise, and ungated images having lower resolution and lower noise (Figure 5-1). Traditional approaches for combining the separately gated data include non-elastic(73-75) and elastic transformation algorithms (76-79). Combining gates using non-elastic transformations has the disadvantage that the final images can only be locally aligned. Elastic algorithms offer a more robust alignment, but they often contain many levels of complexities necessary to achieve practical processing time, making them difficult to characterize. We are proposing an alternate method for utilizing gated data, to create a superior final image, based on a voxel-by-voxel signal evaluation. The algorithm works by characterizing the extents of benefit and degradation introduced when using gated signals to define a voxel’s value, and designating a most appropriate value for the voxels final value.

121

Methods The method is based upon the concept that for a given 4D (3D volume plus time) gated data set, an images quality can be optimized using a combination of gated and ungated voxel values in the image. In PET imaging, information is presented as voxel intensity, contained in a 3D field of voxels. If we look at voxels individually it becomes apparent that some voxels benefit largely from the gating of the PET data (achieving improved resolution) while gating comes as a detriment to the signal of other voxels (i.e. the noise in the gated data outweighs the intended advantage). Our algorithm looks at data in a voxel-by-voxel analysis, and determines the relative contribution of respiratory signal to noise for each voxel, and creates a final voxel value formed as a weighted combination of the gated and ungated voxel value. This is repeated for every voxel in the volume.

The overall idea of the algorithm is that we want the voxels in the motion path of high spatial activity gradients to be gated, to preserve the resolution in these areas, and the voxels in uniform or noisy regions to be ungated, to preserve the statistics in these areas. To determine the magnitude of the gated and ungated contributions to each final voxel value, we look at the distribution of frequencies that make up its individual voxel’s time activity curve (TAC). Thus voxels having a majority of signal in frequencies associated with respiratory motion likely lie near two structures with a discernable activity gradient; our algorithm assigns a greater weight for this voxel to tend towards the voxel’s gated value. Voxels containing no dominant respiratory frequencies or respiratory frequencies

122

with very small amplitudes will be assigned a value weighted closer to that of the ungated value for the voxel.

The respiratory and non-respiratory frequencies are found by simulating reproductions of the data acquisition over the respiratory cycle. The pre-reconstructed gated image data using the bootstrap method (80) and reconstructed into images, which thus simulates scans of respiratory cycles acquired back to back. Next, the frequencies in each individual voxel’s TAC (example in Figure 5-2) are analyzed using the fast Fourier transform. The contribution of the voxel signal stemming from respiratory motion is defined from the amplitude of the frequencies which correspond to respiratory frequencies (i.e. those with integer number of cycles per respiratory period). The contribution of the noise in the voxel is derived from the relative magnitude of the nonrespiratory frequencies in the signal.

Voxel specific weight factors used for determining the relative amounts of the gated and non-gated values in the final voxel value are calculated using the following equations:

(

Volumex , y , z = (TAC x , y , z ( gate) * WeightFactor ) + TAC x , y , z * (1 − WeightFactor )

)

Equation 5-1

123

WeightFactor =

∑ FFT (TAC ) ∑ FFT (TAC )

i =respiratory frequencies

i

i

i =all frequencies

Equation 5-2

| FFT(TAC)|i = magnitude of the ith frequency component of the fast Fourier transform of the voxels time activity curve.

To test our methods, we used simulations, and phantom PET scans.

For the simulations, emission volumes of typical clinical activities were simulated and blurred using a convolution function and Poisson noise was added. Two scenarios were simulated: a doughnut-shaped activity source atop a dual background interface, moving in sinusoidal motion (Figure 5-3), and a 4D anthropomorphic digital phantom (NCAT phantom (18)), simulating human PET including respiratory motion. For all the simulations, resultant images were accompanied by standard deviation images. The standard deviation images were made by repeating simulations 100 times, and mapping out the voxels associated standard deviation.

Our phantom scan consisted of a IEC/NEMA sphere phantom atop a motion platform, which repeatedly moved in sinusoidal motion (displacement 1.2 cm and 1.8 cm in the interior-posterior and superior-inferior directions, respectively) with a period of 4.6

124

seconds (simulating respiratory motion). The spheres had volumes of ~ 0.5, 1, 2, 4, 8, and 16 ml. All of the spheres in the phantom where filled with F-18 at a ratio of 10:1 activity, except for the largest one, which was filled at a ratio of 6:1. The PET system used in the phantom acquisition was an ECAT EXACT HR+ (CTI, Knoxville, TN). The ECAT EXACT HR+ is a BGO PET system with an intrinsic resolution of approximately 4.5 mm FWHM (isotropic) at the center of the field of view.

All periodic cycles were divided into 8 time gates/cycle. All data consisted of three separate cycles (1 original data set, and two additional sets generated using the bootstrap method). Since the simulations did not involve forward and backprojection in the reconstruction, bootstrapped data set were simulated by recreating the images with the same noise statistics as the original data.

For the IEC/NEMA phantom scan data, it was the sinograms that were recreated using the bootstrap method, which were then reconstructed into images using OSEM reconstruction (2 iterations, 8 subsets). The sinograms were derived from a image-based respiratory gating signal (81).

Ungated images were derived from the mean of all 24 frames of the image data (24 bootstrapped frames equaling 3 respiratory cycles). The mean of the gated images was used so that the gated/ungated images could be compared directly, irrespective of

125

reconstruction algorithm. All of the simulated frame were used so that all the gated and ungated numbers were derived from the same data set.

For the computer simulations, resultant images were accompanied by standard deviation images. The standard deviation images were made by repeating all simulations 100 times, and mapping out the voxels associated standard deviation.

Results The resultant images using combined time frames demonstrated better resolution than the corresponding ungated images, and better signal to noise than the gated images (Figure 5-3, Figure 5-4, and Figure 5-5). From the images we can see that areas with high activity gradients were weighted toward gated voxel values, as we expected, and the noise (standard deviation) was higher in these areas. The overall noise was never worse in the gated-combined images than in the gated image, and often much better. The total activity in the combined-gated images was conserved (difference = 0.01%).

The image statistics for the human simulation are summarized in Table 5-1. We can see that the relative % standard deviation for an ROI drawn in the liver for the ungated, gated, and combined-gated image was 1, 1.98, and 1.06, respectively. The associated values for the mean uptake of the highest 80% of the voxels found in an ROI drawn around a moving lesion was 1, 1.40, and 1.33.

126

The image statistics for the IEC/NEMA phantom PET scan are summarized in Table 5-2. We can see that the relative % standard deviation for an ROI drawn in the background for the ungated, gated, and combined-gated image was 1, 2.57, and 1.53, respectively. The associated values for the mean uptake of the highest 80% of the voxels found in an ROI drawn around the 2nd largest sphere was 1, 1.25, and 1.15.

The differences among the ungated, gated, and combined-gated image are illustrated in Figure 5-6.

As a double check that our IEC/NEMA phantom PET scan data was correctly gated and recombined, we defined a threshold ROI for the 8ml sphere, using the mean of the top 80 percent of the voxels within a general ROI as the threshold (from Table 5-2). Figure 5-7 illustrates the measured displacement (relative to most end inspiration position) from the combined-gated image set. This provides verification that we are in fact working with gated data that does contain sinusoidal movement.

While there are several ways to quantify resolution, we included a profile analysis for the 8 ml sphere in the IEC/NEMA phantom (Figure 5-8) along the axis of movement. The FWHM measurements for the ungated, gated, and combined-gated sphere uptake peaks were 16.4, 10.5 and 12.2 mm, respectively.

127

Discussion We have developed a fully automated algorithm which can be applied to gated image data, which utilizes the information from the time voxel activity curves, to create a superior image. While there are other gate combination methods presented in literature, our methodology offers a new approach. Our algorithm is straightforward, and can be summarized in simple equations.

In our implementation, the software was written in the IDL programming language, and took ~ 2 hours to run for a matrix of 128x128x81x8 elements on a p4 pentium processor, and combined-gated volumes were generated for all respiratory gates.

Defining, quantifying and presenting uptake measurements in PET is already a source of confusion because of effects of noise and resolution (55, 82). Combined-gated images offer an advantageous balance of resolution and noise between ungated or gated images alone. These images will be more dependable/reproducible when used for image quantification/volume definition.

Another added benefit, of producing less noisy, gated images which we did not explore in this study, may be in the added accuracy with utilization of attenuation correction. Clinically, PET scans are commonly corrected for attenuation. Once gated PET data and gated attenuation data (4DCT) are coregistered, corrected PET images can be generated

128

(83), offering a large potential for additional improvements in resolution, thus improving lesion detectability.

If we compare our method to non-linear gate combination algorithms, it is hard to make a systematic comparison, because non-linear algorithms are hard to systematically characterize. Usually they are complex (76, 84), proprietary (50, 75, 85, 86), and can behave irregularly with different patient/image properties. The concept of these algorithms is to translate, rotate, stretch and morph one volume to fit atop another, in an effort maximize a quality function. However, improvements can be limited to a cost function, based either on the activity data in the PET or attenuation data in the CT. A hard-to-notice lesion may likely not be accounted for well in the cost function, and thus may be lost/reduced in the data thrown out during interpolation. This is especially counter-productive as the biggest utility of combining gated PET may be in identifying hard to see lesions. Our algorithm presented in this paper, presents an alternate method for combined gated PET data. The algorithm does not scale, stretch, or add any data, rather it deciphers how much of the available gated data is advantageous to utilize, and presents an algorithm where 300 lines of code can work comparably with programs consisting of 30,000.

Our algorithm comes with one particularly advantageous attribute. The algorithm protects the final image against being any worse that the original ungated image. If the PET data was poorly gated, or contains low count statistics, or for any reason does not

129

make a good candidate for gating, the algorithm detects this and weights the data towards an ungated image. Thus, using this algorithm, one can feel confident they are not risking poor diagnosis resulting from image degradation.

It is important to note that our algorithm is based on comparing the amplitude of respiratory and noise within individual voxels TAC. To do this we used the weight factor as defined in Equation 5-2. This weighting factor was decided upon after developing and testing several different approaches. The weighting factor can be reformed and weighted towards noise or resolution improvements. It is also possible to develop a different approach to frequency analysis or even a whole new weighting factor to improve the performance of this algorithm. The ideal weighting factor needs to determine the contribution of noise and true signal in the observed signal, and can be based on resorting data, Poisson statistics, or advanced signal analysis. Ideally we would like to see as much uniformity as possible in the weighting factor within regions of expected similarity (i.e., liver, lesion). We can speculate that creating more bootstrapped data (more than the 3 used here), or using a non-linear weighting for the ratio for noise to respiratory frequencies, would improve this regional consistency.

A particular strategy we propose for further method development would be in the determination of the voxel noise contributions, an integral part of our gated/ungated voxel weighting factor. In the methods we presented here we have used bootstrapped data sets. We would propose forgoing the bootstrapping off any data, and approximating

130

noise from the assumption of Poisson statistics, either in image or sinogram space. This could speed up the processing time, and possibly improve the accuracy.

Conclusion We were able take 4D gated image data sets, degraded by respiratory motion, and create images with improved noise characteristics and comparable resolution relative to gated images. Our methods are fully-automated, straightforward to understand and implement, and protect against image degradation resulting from the implementation of respiratory gating. Being fully automated, our methods and can be combined with image-based fully automated gating algorithms (as we’ve done) to generate gated clinical PET images at no cost in man time and with no changes in routine clinical PET procedures. Future work will include utilizing these methods both pre- and post- reconstruction, extending these methods to be applied to clinical PET images, as well as incorporating these methods in other modalities, such as the development of low dose 4D CT, and for other biological processes (cardiac motion).

131

Figures

Ungated

Gated

Figure 5-1: Illustration of noise/resolution tradeoff for an example ungated/gated PET image.

132

Gating Candidate 350

Magnitude

300 250 200 150 100

Original Bootstraped

50

0 0

5

10

15

20

25

Phase

Averaging Candidate 140

Magnitude

120 100 80 60 40

Original Bootstraped

20 0 0

5

10

15

20

25

Phase

Figure 5-2: Two voxel time activity curves representing candidates for strong gated/ungated weights. The top curve represents a voxel which is a good candidate for gating, while the right curve represents a voxel which can be left ungated.

133

Ungated

Combinedgated

Gated

C-g weight distribution Image

SD Image

(Top row) Ungated, gated, and combined-gated simulated images.(Bottom row) associated standard deviation images. Distribution of weighting factors are also displayed on the right, used to weight the voxels in the combined-gated image between an gated and ungated value.

134

ted ga

d te ga Un

Figure 5-3: Ungated, gated, combined gated simulated lesion volume comparison.

Ungated

Combinedgated

Gated

C-g weight distribution Image

SD Image ga

t ga ed

(Top row) Ungated, gated, and combined-gated simulated images.(Bottom row) associated standard deviation images. Distribution of weighting factors are also displayed on the right, used to weight the voxels in the combined-gated image between an gated and ungated value.

135

ted

Un

Figure 5-4: Ungated, gated, combined gated simulated human volume comparison.

Ungated

Combined-gated

Gated

C-g weight distribution gated

Ungated

Figure 5-5: Ungated, gated, combined gated IEC/NEMA phantom scan comparison. Ungated, gated, and combined-gated simulated images.

Distribution of weighting factors are also

displayed on the bottom right, used to weight the voxels in the combined-gated image between an gated and ungated value.

136

Gated – Combined gated

Ungated - gated

Ungated – Combined-gated

Figure 5-6: Difference images for the human simulation, illustrating the noise and resolution improvements between the different reconstructions of the simulated human data.

137

Figure 5-7: Relative displacement of the center of mass of the 2nd largest sphere in the IEC/NEMA phantom on the motion platform, as measured from the Combined-gated 4D data.

138

Simulation - Sphere Phantom Profile

Relative Amplitude

Ungated Gated Gated-Combined

0

20

40

60

80

100

120

140

mm Figure 5-8: Line profiles 8 ml sphere in the IEC/NEMA phantom, while place atop a platform undergoing sinusoidal motion

139

Tables Table 5-1: Human Simulation Gating Statistics mean %SD of

mean

Gated

lesion

Value

/SD liver

lesion / Mean

Std Dev

% SD

mean liver Ungated Lesion

66.51

23.64

0.36

0.25

21.96

0.61

Gated Lesion

93.19

58.16

0.62

0.62

15.42

0.85

Combined-Gated Lesion

88.17

53.17

0.60

0.57

27.30

0.80

Ungated Liver

109.28

3.03

0.03

0.03

Gated Liver

110.01

6.04

0.05

0.05

Combined-Gated Liver

109.59

3.23

0.03

0.03

Normalized to ungated value: mean %SD of

mean

Gated

lesion

Value

/SD liver

lesion / Mean

Std Dev

% SD

mean liver Ungated Lesion

1

1

1

1

1

1

Gated Lesion

1.40

2.46

1.76

2.46

0.70

1.39

Combined-Gated Lesion

1.33

2.25

1.70

2.25

1.24

1.32

1

1

1

1

Gated Liver

1.01

1.99

1.98

1.99

Combined-Gated Liver

1.00

1.07

1.06

1.07

Ungated Liver

140

Table 5-2: Sphere Phantom Gating Statistics (2nd largest sphere ~8ml) mean %SD of

mean

Gated

lesion

Value

/SD liver

Std Mean

lesion / % SD

Dev

mean liver

Ungated Lesion

0.63

0.19

0.29

0.24

17.31

2.10

Gated Lesion

0.79

0.27

0.34

0.34

8.42

2.48

Combined-Gated Lesion

0.73

0.21

0.29

0.26

13.06

2.36

Ungated Background

0.30

0.04

0.12

0.11

Gated Background

0.32

0.09

0.29

0.29

Combined-Gated Background

0.31

0.06

0.18

0.18

Normalized to ungated value: mean %SD of

mean

Gated

lesion

Value

/SD liver

Std

lesion / % SD

Mean Dev

mean liver

Ungated Lesion

1

1

1

1

1

1

Gated Lesion

1.25

1.44

1.15

1.44

0.49

1.18

Combined-Gated Lesion

1.15

1.12

0.97

1.12

0.75

1.12

1

1

1

1

Gated Background

1.06

2.57

2.43

2.57

Combined-Gated Background

1.03

1.53

1.49

1.53

Ungated Background

141

6. Recombining Respiratory Gated PET Frames (Method B)

Explanation of Method Extension In the previous section (chapter 5) a method was presented for recombining respiratory gated PET. As discussed in that section, the most important part of the algorithm is in the determination of the weighting factor, and one can conceive of a number of ways to do this. In the work presented in the above chapter, the noise determination, as thus the weighting factor, was determined through utilization of bootstrapped data, and this was the most effective manner with which to design the algorithm.

While we were developing the methods we explored many approaches. One approach and line of thinking that stood out was creating gate combination algorithms which utilized only the available gated image data (n images for n gates), and thus not relying on the availability of raw data (e.g., list mode data or sinogram data) to create bootstrapped images, possibly making the method application more practical. Additionally, such an algorithm readily lends itself for application with other imaging modalities which create/utilize gated digital data (CT, SPECT, MR). It is for these reasons a similar, second method for the recombination of gated PET frames is included in this dissertation, based solely on data available only in gated images.

142

Methods To determine the magnitude of the gated and ungated contributions to each final voxel value, we are looking at the distribution of frequencies that make up its individual voxel’s TAC. Thus voxels having a large component of signal in a frequency close to that of the respiratory cycle will probably lie near two structures with a high activity gradient; our algorithm will assign a greater weight for this voxel to tend towards the voxel’s gated value. Voxels containing no dominant respiratory frequencies or frequencies with very small amplitudes will be assigned a value weighted to be close to that of the ungated value for the voxel (Figure 6-1). Weights are calculated using the following equations:

(

Volumex , y , z = (TAC x , y , z ( gate) * WeightFactor ) + TAC x , y , z * (1 − WeightFactor )

)

Equation 6-1

WeightFactor =

[

]

max (amplitude f 1 ) or (amplitude f 2 )

(max(TAC ) − min(TAC ) ) 2

Equation 6-2

f1 = 1 cycle /respiratory period frequency f2 = 2 cycles /respiratory period frequency

The algorithm utilizes the fast Fourier transform to separate the frequency components for each individual TAC. The weight factor takes into account at the amplitude of the

143

frequencies of the parts of the signal that have one or two cycles per respiratory period relative to the amplitude of the total displacement of the signal. One or two cycles per respiratory period was chosen to represent respiratory frequencies. These correspond to cases where one or two activity gradients pass though a voxel during a respiratory cycle, for example a voxel with an activity changing from low to high during a respiratory period, or a voxel with activity changing from low to high then back to low. The final value of the voxel being processed will be some combination of the gated and ungated values, often being close to one end or the other.

To test our methods we generated 4D image models, simulating a spherical lesion, twiceonce with an empty background, and once with a centrally necrotic (low activity) region and located on edge of a dual background interface. Both Poisson noise and blurring (using a convolution function) were added to all the images. The system underwent sinusoidal linear motion in the direction of the opposing background planes. In addition to these simple movement simulations we also simulated 4D human images using the NCAT phantom. All periodic cycles were divided into 8 time gates/cycle.

Results The resultant images using combined time frames demonstrated better resolution than the corresponding ungated images, and better signal to noise than the gated images (Equation 6-2 and Figure 6-2). The total activity in the combined images was conserved (difference = 0.01%).

144

We re-simulated the lesion data using different count statistics and activity background ratios, the two factors most affecting the images. A summary of the gated and gatedcombined images is shown in Figure 6-3.

The percent standard deviation in the first background region (in relative units) was 1.0 for the ungated image, 2.82 for the gated image, and 1.87 for the combined-gate image. These results are summarized in Table 6-1. The edges in the combined image appear well preserved with the distance between 10% and 90% activity along the profile of the lesion (with no background) having a value of 8 pixels for the ungated image, 3 pixels for the gated image, and 3 pixels for the combined image.

Discussion We have developed a fully automated algorithm which can be applied to gated image data, which utilizes the information from the phase-activity curves of every voxel. The algorithm is summarized in simple equations, and can be written into software with minimal effort (~20 lines of high level code).

For our implementation, the software was written in the IDL programming language, and took 2 hours to run for a matrix of 128x128x81x8 elements on a p4 pentium processor. We designed our algorithm to be free standing: pre-selected noise and signal amplitude

145

levels, along with other possible features, can be incorporated in the weight factor with the potential of further improvement of the images.

The advantage of using this method, as opposed to the first version of the gate combination methodology is that this method only need the gated data set to execute. The simplicity of this algorithm makes it easier and faster to implement.

The negative side of using this algorithm is that the estimates for the noise contributions, a critical step in our algorithm, is limited. We know that calling higher frequencies noise and lower frequencies true signal, has pitfalls. Both signals and noise reside in all of the frequencies present in the fast Fourier transform of a phase-activity curve. However, this approximation may be suitable, as it is the areas where this approximation is appropriate, that we are most concerned with. Additionally, variations of this methodology may be developed, possibly offering more assurance in accuracy.

146

Figures

Gating candidate

Averaging candidate

Phase

Phase

Figure 6-1: Example of two voxel time activity curves. The left curve represents a voxel which is a good candidate for gating, while the right curve represents a voxel which can be left ungated.

147

Figure 6-2: Ungated, gated, combined gated simulated lesion and human volume comparison. (Top) Image slices displayed from ungated, gated, and combined-gate volumes, for a simulated lesion (left) and simulated human (right). (Bottom) Corresponding standard deviation images derived from 80 resimulations of image.

148

Figure 6-3: Summary of lesion located on an interface with and without gate combination efforts for a lesion with different relative noise levels and different activity ratios. The relative magnitude of counts in the images are shown along the y-axis, and different combinations for relative activities are shown along the x-axis. Figure illustrates differences in lesion detectability and regional noise.

149

Tables

Table 6-1 Summary of statistics for a simulated lesion. (similar to Figure 6-2 but no simulated necrotic center and no background)

Combined Volume

Gated Volume

UnGated volume

Background 1

1.87

2.82

1.00

Background 2 Volume of 70% true value contour

1.85

2.76

1.00

1947

1947

94

92

94

93

Mean of 70% true contour volume

150

7. Radiation Dose Estimates for [F-18]5fluorouracil Derived from PET-based and Tissue-based Methods in Rats Abstract Introduction : Radiation dosimetry assessment often begins with measuring pharmaceutical biodistribution in rodents. The traditional approach to dosimetry in rodents involves a radioassay ex vivo of harvested organs at different time points following administration of the radiopharmaceutical. The emergence of small animal PET presents the opportunity for an alternative method for making radiodosimetry estimates previously employed only in humans and large animals. In the current manuscript, normal-tissue absorbed dose estimates for the 18F- labeled chemotherapy agent [18F]5-fluorouracil ([18F]5-FU) were derived by PET imaging- and by tissue harvesting-based methods in rats.

Methods : Small animal PET data were acquired dynamically for up to 2 hours after injection of [18F]5-FU in anesthetized rats (n=16). Combined polynomial and exponential functions were used to model the harvesting-based and imaging-based time activity data. The measured time-activity data were extrapolated to modeled (i.e. Standard Man) human organs and human absorbed doses calculated.

151

Results : Organ activities derived by imaging-based and by harvesting-based methods were highly correlated (r > 0.999) as were the projected human dosimetry estimates across organs (r = 0.998) obtained with each method. The tissues calculated to receive highest radiation dose by both methods were related to routes of excretion (bladder wall, liver, and intestines). The harvesting-based and imaging-based methods yielded effective dose (ED) of 2.94E-2 mSv/MBq and 2.97E-2 mSv/MBq, respectively.

Conclusions : Small-animal PET presents an opportunity for providing radiation dose estimates with statistical and logistical advantages over traditional tissue harvestingbased methods.

152

Introduction

A prerequisite to the use of diagnostic radiotracers in humans is the measurement of organ radiation doses in animals, used to estimate administered activities which will maintain organ doses within an ‘acceptable’ range, while yielding diagnostically useful scintigraphic images. With the introduction of new radiotracers to study the pharmacokinetics and pharmacodynamics of anti-tumor agents and biological processes in vivo with positron emission tomography (PET), there is a need for efficient evaluation of the time-dependent biodistribution of such tracers in animals.

Radiation dose estimates in small animals such as rodents are traditionally derived by sacrificing the animals at various times points after tracer administration, harvesting and weighing their organs, and measuring their respective activities(87-89). As a recent example, the radiation dosimetry of [18F]fluorocholine was calculated on the basis of data from 29 mice sacrificed at 10, 30, 60, or 120 m post-administration, followed by assay of activities in heart, brain, lung, liver, kidneys, and muscle (90). Larger animals such as monkeys are also often studied with non-invasive imaging methods (91, 92). Once sufficient data have been obtained to reasonably establish that a given administered activity will yield organ doses within an ‘acceptable’ range, dose estimates can be refined using imaging data from a small number of human subjects(90, 93, 94).

153

One of the earlier studies which utilized small animal PET for radiodosimetry measurements was presented by Ugur et. al. (95), who showed that the PET imaging agent 66Ga-DOTATOC can be used both diagnostically, and to provide quantitative data to be used for radiodosimetry of a tumor and other organs in mice. For this study, the dosimetry was calculated by assuming complete local absorption of non-penetrating radiation, thus allowing for the direct conversion of image derived activity concentration to radiation dosimetry (using a conversion factor). A limitation of this approach is the simplification that the dosimetry arising from penetrating radiation in non-significant. Thus this methodology does not extend well to assessing other imaging agents/isotopes, which are often selected because of there high penetrating/non-penetrating radiation properties.

In additional radiodosimetry work by Palm and colleagues (96), 86Y-Trastuzumab PET was used to measure pharmacokinetic parameters with which to estimate 90Y-trastuzumab radiodosimetry in mice, for tumor, liver, kidneys and spleen. This work accounted for some cross-organ absorbed dose (organs mentioned above) using methods based upon the MIRD formalism (97), and murine model specific geometry/S-factors.

The study presented here investigates the feasibility and accuracy of estimating human radiodosimetry, based upon imaging parameters derived from small animal dedicated PET, using [18F]5-FU. As described in more detail in the Discussion, our present study extends and integrates previous work in several ways: In our investigation we use a

154

robust description of organ biodistribution (source organs in dosimetry calculations) and whole body activity; we are using a fluorine-based compound in small animal PET; and we are making and testing human dosimetry estimates from extrapolation of biodistribution data across species (from rats to humans) to calculate dosimetry in human-specific geometry.

Fluorouracil (5-FU) was reported to have tumor inhibitory properties by Heidelberger et al in 1957 (98), and has been established over the subsequent decades to have efficacy against several types of cancers (e.g. colorectal, breast, stomach). Shani and Wolf later pioneered the testing of [18F]-labeled 5-FU to predict tumor response to 5-FU (99).

The radio-labeled form of the fluorouracil compound, [18F]5-fluorouracil ([18F]5-FU) has been studied with PET, and examined with respect to its prognostic value for treatment efficacy (100-102). Based on those early promising studies, [18F]5-FU may be investigated more widely in humans, and accurate human radiodosimetry for this tracer will be valuable for selecting doses that provide for both high quality images and patient safety with respect to radiation exposure.

Methods

155

All animal studies were performed under a protocol approved by the Chancellor’s Animal Research Committee of UCLA. Sixteen Sprague-Dawley rats (avg. weight: 250 g) were injected with a 1,000 μL solution containing ~100 MBq (range: 60-172 MBq) of [18F]5FU in saline, which was administered via tail vein injection. Rats were anesthetized by inhalation of 2% isofluorane and images were reconstructed from dynamic acquisitions, with the rat in the prone position, and the long axis of the animal set parallel to the plane of detectors. The acquisitions were executed using a Siemens/CTI Focus 220 small animal PET scanner (CTI Concorde Microsystems LLC, Knoxville, TN). The 3D listmode data was sorted using Fourier rebinning. Images were reconstructed using a filtered back projection algorithm with a ramp filter - cutoff equal to the Nyquist frequency i.e. 1.0 / (2.0 * sampling). Data were reconstructed with a pixel size of 0.4 mm and a plane separation of 0.8 mm in a 128 x 128 x 95 matrix. Dynamic scans consisted of 10 minute time frames, including upper body and lower body bed positions, and all data were decay-corrected to the beginning of the time frame. Image-based organ uptake measurements were derived from organ regions of interest defined on the scan by a single observer (example in Figure 7-1). ROIs were defined by drawing spheres in representative organ data spaces relative to visual anatomic landmarks and corresponding atlas images. The bladder ROI was drawn using a threshold region-growing algorithm for a late time frame (volume large enough to encompass all counts in the bladder). Following scans, rats were sacrificed, by intracardiac administration of Nembutal(at 30, 60, 90 ,and 120 minutes).

156

Following scanning, the rat organs were harvested and weighed and total organ activity content was quantified using a gamma well counter. All the organ activity measurements were normalized to injected activity. In calculating the organ activity content from the image data, the small animal PET scan measurements were calibrated by comparing the mean organ activity concentration observed in the last frame of the PET image, with the time-corrected harvested organ activity concentration, for each organ, for each rat in order to intrinsically adjust for attenuation, partial volume effects, and counting efficiency. The liver calibration factor was used for organs for which harvesting full activities is difficult (bladder, intestines). Image-based organ activity measurements were made by manually defining regions of interest (ROI) for each organ, calculating the mean activity concentration within the organ ROI, and then calculating total organ activity content by multiplying those concentrations by total harvested organ weight. The bladder volumes were calculated from the scan, and the lung volumes were found by dividing the harvested lung weight by density of 0.296 g/cc (103). All other organ total volumes were derived from their harvested weights, assuming those tissue densities to be equal to 1 g/cc. All organ activities were assumed to be uniform within each organ. A remainder disintegration value was derived from all the unaccounted-for activity (the difference between the total disintegrations and the organ localized disintegrations).

For comparison purposes an additional set of image-based biodistribution measurements were also made using a single global calibration factor. Calibration of the small animal PET scanner was derived using a vial containing a known quantity of 18F assayed in a

157

dose calibrator (Biodex Atomlab™ 300 Dose Calibrator). Vial images were reconstructed using the same parameters as used in the animal studies, and a global calibration factor was found from comparing the raw PET counts to the known activity concentration. Attenuation correction factors were estimated on an organ by organ basis. The estimations were derived by inflating the Moby digital mouse phantom (18) to the approximate volume of the rats we used. The emission and attenuation data were forward projected, corrected for attenuation, and then backprojected into image space (projections done in 2D). Organ attenuation factors were derived from comparing preand post- correction activity values.

A flow chart summarizing the main steps in our methods can be seen in Figure 7-2. The three main steps in our methods for calculating dosimetry were as follows. We began by calculating the total number of disintegrations emitted from each particular animal organ. Tracer biodistribution was then extrapolated from rat to human based on relative organ masses. Finally, human dosimetry estimates were generated using dedicated software (104).

Animals were sacrificed at several time points after injection (30, 60, 90, and 120 minutes). Image-based time activity curves were generated for the rats that were sacrificed at 120 minutes, immediately following the final frame of the 120 minute

158

acquisitions, and normalized by the activity injected. Time activity curves were made in a similar manner for the harvested organ activity measurements.

For both harvesting- and imaging-based methods, a second degree polynomial function (f = ax2+bx+c) was fitted (using Microsoft Excel curve fitting algorithms) to represent the time-activity curve between the time of injection and 120 minutes later. Polynomial functions were used for their flexibility for universal (unmodeled) regions of interest. The time-activity curve for the heart was modeled using two polynomial functions, one representing initial blood pool uptake (0-15 min), and the other representing the rest of the scan data. At 120 minutes after injection, biological distribution was considered stable and physical decay of the isotope was modeled as a monoexponential function. By integrating the modeled activity curves from injection time out to infinity, we calculated the disintegrations attributed to each organ from a unit of injected activity (1 MBq), accounting for both biological half-life as well as physical half-life.

This process was repeated for all organs for both the imaging and harvested data. All measurements were also corrected for a mean measured 2% of total activity remaining in the tail. For bladder measurements we used a non-voiding bladder model. These estimates can thus be regarded as conservative for those patients who are capable of significant urine excretion.

159

For the harvested-organ-based dosimetry set, the bladder and intestinal percent activity data used were derived from the imaging based calculations, since some of the radioactive content of urine and feces would otherwise be lost and unaccounted for during the organ-harvesting process.

Several methods have been previously described for extrapolating organ uptake data in animals to equivalent uptake in humans (93) . For this study we used the proposed method of Kirschner and colleagues (105), as this seems to be the most standardized procedure in literature. The method is based upon a percent kilogram dose per gram (% kg dose / gm) unit:

⎛ % ⎞ ⎜⎜ ⎟ organ ⎟⎠animal ⎝ (% kg dose / gm) = ∗ kg animal = g organanimal

⎛ % ⎞ ⎜⎜ ⎟⎟ ⎝ organ ⎠human ∗ kg human g organhuman

Assuming this parameter is constant across species, one can calculate equivalent organ activities across different organisms, after establishing weight of each organ in each of the species (we used the mean weight of the organs harvested from the animals, and standard weights established for the adult male human organs (103)).

Using the imaging- and the harvesting-based organ time-activity data in rats, the respective human-equivalent organ time-activity data were thus derived and cumulated activities calculated. The total body residence time was calculated to be 2.64 hours

160

(100% x 1 / λ). With those data we applied OLINDA/EXM (104) MIRD-based (106) human dose estimation software to obtain human organ dose estimates.

The standard deviation of the dose delivered to individual organs was calculated using the following equation, similar to previously used methods (107-109):

σ Di2 = ∑ S ij2 ∗ σ 2j j

Where σDi is the standard deviation (SD) of the estimated dose for the ith organ, Sij is the S value for the dose to the ith target organ from the jth source organ (mGy/MBq-s), and

σj is the SD of the integrated activity for the jth source organ.

The source organ standard deviation (σj in the above equation) was designed to account for the inter-individual variation of the image-based time course measurements relative to the gold-standard measurements derived from the harvested data. To find its magnitude for each respective organ, polynomial and exponential models were fit to time activity curves representing mean organ measurements ± one SD, with the SD derived from the spread of the activity measurements derived at each particular time point. An example of the ± SD window can be seen for the liver in Figure 7-3. These ±SD ranges were then normalized by the 120 minute calibration time point, and extrapolated to the human

161

biodistribution data set, thus generating an associated σj to use in the above dosimetry error equation.

Results

Time-activity data in rats were acquired both by dynamic PET imaging and by radioassay of harvested organs (Table 7-1). Percent total disintegrations emitted from each organ calculated with imaging-based and harvesting-based data are shown in Figure 7-4 and Table 7-2. The Pearson correlation coefficient between these data sets was > 0.999, with

or without inclusion of the excretory-related organs (bladder, and intestines) in the correlation assessment.

Human activity distribution projected from our rat-based measurements and resultant human radiodosimetry are displayed in Figure 7-5 and Table 7-3 respectively. This figure and table also display the values corresponding to the activity distribution based entirely on image data (i.e. scanner-based calibration rather than harvested organ-based calibration). In addition, Figure 7-6 has been included to help visualize the projected human [18F]5-FU biodistribution and resultant radiodosimetry.

Harvesting-based and imaging-based dosimetry measurements both illustrated that the organs receiving highest radiation dose were the bladder wall, liver, and intestines, but

162

image-based data demonstrated less inter-individual variation than harvest-based data (e.g. see liver time-activity curves in Figure 7-7). When used to calculate dosimetry for [18F]5-FU in humans, both the harvest-based and image-based methods would give comparable estimates of a safe dose, in terms of both effective dose (2.94E-2 mSv/MBq vs to 2.97E-2 mSv/MBq, respectively), as well as of dose to critical organs. The relative contributions to organ dose from beta and gamma emissions were 47% and 53%, respectively, for the harvested-based distribution and were 48% and 52% for the imaging-based distribution.

Discussion

Using imaging (i.e. PET)-based methods, we have determined that for an administered activity of 750 MBq of [18F]5-FU to Standard Man, the doses to critical organs including the bladder wall, liver, small intestines, and lower large intestines would be 93, 65, 54, and 54 mSv, respectively - well within the range considered acceptable for diagnostic Nuclear Medicine procedures(93, 110). We can also compare our radiodosimetry estimates with published dosimetry estimates from a more widely used fluorinated pyrimidine: [18F]fluorothymidine (FLT)(56), Effective dose equivalent measurements (as described in ICRP 53) for FLT and our imaging based [18F]5-FU were 2.8E10-2, and 3.26E10-2 mSv/MBq, respectively. Both the FLT dosimetry set and our [18F]5-FU dose estimates identified the urinary bladder wall, the liver, and the kidneys as being organs

163

receiving high levels of radiation dose (1.79E-01, 4.51E-02, 3.52E-2 mGy/MBq for FLT, and 1.24E-01, 8.69E-02, 3.74E-02 mGy/MBq for 5-FU, respectively).

Determining tissue distribution of 5-FU has been of long standing interest (56, 111-113). Harvesting-based dosimetry for [18F]5-FU was first published in 1982 by Shani et al. (114). They studied a group of 12 rats injected with a dose of [18F]5-FU and sacrificed at 30, 60 and 120-minute time points following injection. For dosimetry calculations in that article, the uptake was considered instantaneous, having a measured effective half-life within the organs. Tracer-biodistribution was considered similar enough to humans to allow for a reasonable estimation. Residence times were calculated for 11 major organs from the rat measurements, and residence times for the bladder were based on urine based data actually collected from one human patient. They projected that the organ receiving the highest radiation dose in humans would be the bladder wall, as was also the case for our study (Table 7-3). If we compare our imaging-based bladder wall radiation dose estimate of 0.124 mGy/MBq to the bladder wall [18F]5-FU dose estimate by Shani et. al. of 0.197 mGy/MBq, it is seen that the dosimetry of the bladder wall projected from imaging rat bladders (as we’ve done), has a value close to that based upon a study of direct measurements of urine from humans.

Qualitative data supporting interspecies extrapolation performed here comes from a human [18F]5-FU PET scan acquired clinically at our facility. Figure 7-8 shows a side by side comparison of rat and human [18F]5-FU PET scans. Visual inspection of the figure

164

illustrates the similarities in the biodistribution of the radiotracer in both species, with activity in both cases predominantly concentrated in tissues reflecting routes of excretion. As another way of providing external validation of the dosimetry estimates obtained here, we also compared them to the results of work that had been performed in monkeys, provided courtesy of Peter Conti and colleagues (from IND application: #57,954). This study projected the whole body effective dose equivalent (for standard man) to be 2.70E2 mSv/MBq (extrapolated from monkeys), which was similar to our image based EDE estimate of 3.26E-2 mSv/MBq (extrapolated from rats).

In the present work we have taken an approach to estimate [18F]5-FU human dosimetry, with methodology that builds upon prior investigations. Previous work utilizing small animal PET - based determinations of biodistribution of 66Ga-DOTATOC (95) relied upon the assumption that the radioactive content of each organ served as the sole source of the radiation exposure for that organ. Fuller modeling of the relationship between PET images and radiopharmaceutical dosimetry followed in work with 86Y-trastuzumab, which utilized PET-based activity concentration measurement for murine dosimetry derived from MIRD-based calculations (96). We have approached radio-dosimetry derived from small animal PET, with the aim to develop a standardized methodology for this application. Our work involved the use of rats, rather than mice, which by comparison in size, allows for greater accuracy in activity measurements. Secondly, we made an effort to more fully account for the distribution of all the injected activity throughout the rats. That included measuring activity concentration for 11 organs, as well

165

as making certain that the total activity in the organs and the whole body remainder accounted for 100% of the expected activity. We followed up upon using rodent data for biodistribution measurements as an appropriate early step for projecting human biodistribution, with human radiodosimetry estimates that were more accurately projected by using a human anatomical model. Our methods include subject specific biodistribution extrapolations from the measured sources to a model phantom from which the dosimetry was derived, with corresponding S-factors (103). Finally, we applied these methods to the [18F]-labeled chemotherapy agent, [18F]5-FU and presented updated radiodosimetry data for this tracer, comparing or externally validating those data in multiple ways (to harvested rat organ measurements, qualitative comparison to whole-body human distribution, and quantitative comparison to human dosimetry projected from largeanimal studies).

There are in principle several advantages to utilizing PET data of small animals for radiation dosimetry purposes. First, only a fraction of the animals currently used for that purpose need to be purchased and maintained, and far fewer animals need to be sacrificed. Further along those lines, the number of hours involved in dissecting and processing tissues, homogenizing or dissolving them in organic solvents, and doing scintillant-based or direct gamma counting, decreased, and the overall time from the start of a dosimetry-determination project to its conclusion can be abbreviated. Third, the time-course of radiotracer distribution in an individual animal, which is intended to model what would occur in an individual human subject receiving the dose, actually

166

measured, rather than having to be inferred from assembling cross-population data as is traditionally done. This not only requires fewer animals because of the multiple timepoints of data that can be obtained per animal, but also increases the statistical stability of the data, because each individual is serving as its own across-time control, avoiding timepoint by point variance caused by inter-individual differences.

Harvesting radioactivity in organs represents a direct method of quantifying activity distribution. While it is not prone to scatter, attenuation, or other sources of image degradation which can affect imaging data, the method does have some disadvantages. Generating time-activity curves from harvested organs adds to the burden of number of overall animals needed, and the labor required to generate dosimetry data. Additionally, measurements based upon harvesting certain organs like the heart, bladder, or intestines, which are often important contributors to dosimetric analysis, are problematic due to loss of blood, urine or feces occurring during the harvesting process, respectively. Imagingbased measurements, in contrast, are relatively quick and easy to make, requiring fewer specimens and person-hours. With imaging, regional volumes can also be measured without invasive procedures.

The largest difference in overall expense, work and time in the methods we have presented follows from the fewer rats needed to be studied using the imaging based methods, as well as substituting organ ROIs in image analysis software for sacrificing animals, dissecting the animals to recover individual organs at each time point, weighing

167

the organs, and assaying the radioactivity levels in the extracted tissue. The first process took about 15 minutes for each animal at one time point after tracer administration, while the latter process required approximately three times as long. Moreover, once the organ ROIs were defined for a particular animal at one time point, they could be used for the measurements made at the other sampled time points (e.g., at ten minute intervals), while for the harvested data the whole measurement process needs to be repeated for each sampled time point. Once all the data are acquired, the data analysis work is comparable using either method.

With the dose of [18F]5-FU used in this study (approximately 0.5 MBq/g), we were able to observe well defined contours on organs that had taken up moderate levels of activity. This dose was comparable to amounts used to study radio-labeled chemotherapy agents in rodents in previously published work (54, 115, 116).

[18F]5-FU like many tracers studied have a majority of tracer concentration, beyond the perfusion stage, located predominantly in organs involved as routes of excretion (e.g. liver, genitourinary structures). It would be valuable to confirm generalizability of this approach for those traces that do have a large percentage of uptake in other organs.

Projecting pharmaceutical biodistribution and radiotracer dosimetry across species from small animals to humans can be useful for accelerating the development of radioactive compounds to be used in clinical settings, and is a common first step, consistent with the

168

recommendations of ICRP 62 (93), prior to moving forward with human measurements from a small number of volunteers. This strategy is grounded upon an initial estimate that “partition of activity among various tissues, organs, and excretory routes is broadly similar in investigational animals and man” [4]. Along those lines we have provided a comparison to human radiodosimetry estimates derived here and those derived from prior work with a primate model and found them to be substantially similar.

To the extent that small-animal imaging-based methods can be used in place of harvesting-based methods, that process may be facilitated by the kind of approach illustrated here.

Conclusions

Small-animal PET imaging-based dosimetry compares well with dosimetry estimates generated by more traditional methods (i.e. measurements derived from harvested organs), appears less prone to the data scatter associated with inter-individual variation which is inherent in cross-population harvested-based methods, and can be carried out utilizing less time and labor, as well as fewer animals. As this approach is validated with other tracers with substantially different types of distributions, it may prove useful for simply and, rapidly obtaining radiodosimetry estimates, as a step in the process of allowing new tracers to be developed for use in humans.

169

Figures

A

B

Figure 7-1: Illustrative example of ROIs used for dosimetry calculations from example scan (A) 2D view of ROIs. (B) 3D rendering of PET and ROI volumes.

170

Figure 7-2 Flow chart illustrating methods used to project human [F-18]5-FU dosimetry, derived from harvesting-based and imaging-based rodent measurements.

171

0.5

Liver

Activity MBq / MBq injected

0.45 0.4 0.35 0.3 0.25 0.2 0.15 0.1 0.05 0 0

2000

4000

6000 8000 Time (s)

10000

12000

14000

Figure 7-3: Illustration of error window (mean ± SD) used to calculate source organ standard deviation for the liver. The solid black points represents the mean activity time curve data, and the hollow points represent the bounds at a distance of ± 1 SD.

172

Projected Biodistribution in Human with Different Methods 50.0%

45.0%

Imaging-Based Measurements

40.0%

Liver SmInt+cnt*

35.0%

Bladder+Cnt* Lw r Colon+cnt*

30.0%

Kidney Lungs

25.0%

Stomach+cnt Brain

20.0%

Heart Wall Spleen

15.0%

Thyroid Remainder

10.0%

5.0%

0.0% 0.0%

10.0%

20.0%

30.0%

40.0%

50.0%

Harvesting-Based Measurements

Figure 7-4: Biodistribution measurements between derived from imaging-based (harvest calibration) measurements vs. harvesting based measurements.

173

30%

Harvesting

Percent Activity

25%

Imaging (Harvest-based calibration)

20%

Imaging (Scanner-based calibration) 15% 10% 5%

Th yr oi d

Sp le en

al l

He ar tW

Br ai n

Lu ng St s om ac h+ cn t

cn t* ad de r+ C Lw nt rC * ol on +c nt * Ki dn ey s Bl

In t+

Sm

Li v

er

0%

Figure 7-5: Percent total organ disintegration measurements (cumulative activity ) of [F-18]5-FU calculated in humans based upon small animal imaging-based and harvesting-based activity data. (*) denotes harvested measurements taken from image based calculations

174

Human Activity Distribution

Harvesting

Imaging

Human Radiodosimetry Imaging

Harvesting

Figure 7-6: Projected human activity distribution and resultant radiodosimetry, for imaging (harvest-based calibration) and harvesting rat data.

175

Liver Time Activity Curves

Activity MBq / MBq injected

0.8 Harvested activity

0.7

Imaging Activity

0.6

Harvesting activity curve

0.5

Imaging activity curve

0.4 0.3 0.2 0.1 0 0

2000

4000

6000

Time (sec)

Figure 7-7: Time-activity curves for the liver, for harvesting- and imaging-based methods

176

8000

Figure 7-8: Small animal (A) and human (B) PET images acquired after administration of [F-18]5-FU. (Rat image acquired on Siemens/CTI Focus 220 small animal PET scanner and human image acquired on a Siemens/CTI Biograph PET/CT scanner). Rat image is a summed image acquired following 0.68Mbq/g injection.

177

Tables Table 7-1: Activity distribution measurements (rat).

HarvestingImaging-Based Mean (n=4) Harvesting-

Based organ

Imaging-based

Based organ activity

cumulative

organ activity

(MBq), t=120 min

activity (MBq-s)

(MBq), t=120 min

(± SD)

(area under

(± SD)

organ cumulative activity (MBq-s) (area under fitted curve) fitted curve)

organ Heart

5.85E+02 ± 1.62E+02

1.22E+01

5.80E+02 ± 1.47E+02

2.09E+01

Liver

1.11E+05 ± 2.14E+04

2.68E+03

1.10E+05 ± 1.97E+04

2.82E+03

Thyroid

4.32E+02 ± 1.60E+02

8.70E+00

4.51E+02 ± 2.49E+02

1.49E+01

Bladder

7.17E+02 ± 2.26E+02

1.18E+01

1.43E+05 ± 4.75E+04

2.49E+03

Colon

2.62E+03 ± 6.06E+02

6.02E+01

8.26E+04 ± 2.00E+04

2.01E+03

Lungs

1.01E+03 ± 2.97E+02

2.13E+01

3.03E+02 ± 1.03E+02

6.33E+00

Spleen

5.90E+02 ± 1.20E+02

1.21E+01

5.95E+02 ± 1.58E+02

2.18E+01

Kidneys

9.59E+03 ± 3.72E+03

2.11E+02

9.45E+03 ± 3.49E+03

3.13E+02

SmInt

6.73E+03 ± 1.57E+03

3.08E+02

8.42E+04 ± 3.46E+04

1.63E+03

Brain

2.35E+02 ± 8.30E+01

5.27E+00

2.37E+02 ± 1.06E+02

4.12E+00

Stomach

9.52E+03 ± 8.98E+03

1.33E+02

8.22E+03 ± 6.84E+03

1.38E+02

178

Table 7-2: Percent injected dose (PID) estimated for humans (cumulative activity). Radiotracer distributions resulting from measurements based on imaging (harvest-based calibration), and harvesting projected to human biodistributions. (Units in percent injected dose per kilogram, cnt = contents). (*) denotes harvested measurements taken from image based calculations

Liver SmInt+cnt* Bladder+Cnt* Lwr Colon+cnt* Kidney Lungs Stomach+cnt Brain Heart Wall Spleen Thyroid Remainder body

Correlation (H and I) Correlation w/o *'s

PID / organ Harvesting Imaging 24.3% 25.6% 11.4% 11.4% 9.1% 9.1%

PID / kilogram Harvesting Imaging 76.9% 80.9% 6.0% 6.0% 440.9% 440.9%

3.7% 1.1% 0.6% 0.5% 0.2% 0.1% 0.1% 0.0% 48.7%

3.7% 1.7% 0.6% 0.6% 0.1% 0.2% 0.2% 0.0% 46.7%

14.2% 3.7% 0.6% 2.9% 0.6% 0.1% 0.1% 0.0% 0.7%

14.2% 5.4% 0.6% 3.0% 0.5% 0.2% 0.2% 0.0% 0.7%

with remainder

without remainder

with remainder

without remainder

0.99906

0.99941

0.99995

0.99995

0.99905

0.99981

0.99980

0.99981

179

Organ Weight (kg) 0.316 1.910 0.021 0.259 0.310 1.000 0.183 0.299 1.100 1.420 0.413 66.470 73.700

Table 7-3: Radiation dosimetry estimates (mSv/MBq). Estimated radio-dosimetry resulting from measurements based on imaging, and harvesting (OLINDA output);.

Harvesting

(Scanner-based calibration)

Urinary Bladder Wall Liver LLI Wall Small Intestine Kidneys Gallbladder Wall ULI Wall Uterus Ovaries Stomach Wall Adrenals Pancreas Total Body Osteogenic Cells Red Marrow Spleen Heart Wall Muscle Lungs Testes Thymus Breasts Skin Thyroid Brain

Imaging

Imaging

(Harvest-based calibration)

1.24E-01 8.27E-02 7.24E-02 7.20E-02 2.84E-02 2.44E-02 2.25E-02 2.24E-02 2.11E-02 1.66E-02 1.51E-02 1.51E-02 1.22E-02 1.20E-02 1.01E-02 1.01E-02 9.94E-03 9.56E-03 9.25E-03 9.20E-03 7.66E-03 6.73E-03 6.38E-03 6.14E-03 2.27E-03

1.09E-01 6.99E-02 5.14E-02 5.00E-02 3.03E-02 2.22E-02 1.90E-02 2.02E-02 1.84E-02 1.57E-02 1.53E-02 1.54E-02 1.22E-02 1.33E-02 1.02E-02 2.21E-02 1.18E-02 1.00E-02 1.28E-02 9.80E-03 8.83E-03 7.56E-03 7.03E-03 7.99E-03 4.41E-03

1.24E-01 8.69E-02 7.22E-02 7.18E-02 3.74E-02 2.50E-02 2.25E-02 2.22E-02 2.09E-02 1.68E-02 1.56E-02 1.55E-02 1.22E-02 1.17E-02 1.00E-02 1.33E-02 1.15E-02 9.43E-03 9.38E-03 8.95E-03 7.52E-03 6.63E-03 6.24E-03 7.54E-03 2.04E-03

2.94E-02

2.54E-02

2.97E-02

0.9850

0.9977

Effective Dose (mSv/MBq)

Correlation with Harvesting

180

SD

Beta/photon ratio

1.18E-02 6.65E-03 7.32E-03 2.43E-02 7.59E-03 2.29E-03 4.27E-03 3.60E-03 3.76E-03 2.34E-03 2.49E-03 2.63E-03 2.19E-03 2.28E-03 3.97E-03 2.42E-03 2.10E-03 2.17E-03 2.08E-03 2.18E-03 1.78E-03 2.29E-03 1.89E-03

1.97 1.32 2.19 2.63 1.23 0.10 0.12 0.12 0.13 0.59 0.18 0.18 0.56 0.75 0.20 0.58 0.33 0.33 0.31 0.35 0.45 0.55 0.60 0.77 0.22

8. Time-course of effects of external beam radiation on [18F]FDG uptake in healthy tissue and bone marrow.

Abstract:

Introduction: The utility of PET for monitoring responses to radiation therapy have been complicated by metabolically active processes in surrounding normal tissues. We examined the time-course of [18F]FDG uptake in normal tissues using small animaldedicated PET during the 2-month period following external beam radiation. Methods: Four mice received 12Gy of external beam radiation, in a single fraction to the left half of the body. Small animal [18F]FDG-PET scans were acquired for each mouse at 0 (preradiation), 1, 2, 3, 4, 5, 8, 12, 19, 24, and 38 days following irradiation. [18F]FDG activity in various tissues was compared between irradiated and non-irradiated body halves before, and at each time-point after, irradiation. Results: Radiation had a significant impact on [18F]FDG uptake in previously healthy tissues, and time-course of effects differed in different types of tissues. For example, liver tissue demonstrated increased uptake, particularly over days 3-12, with the mean left to right uptake ratio increasing 52% over mean baseline values (p<0.0001); in contrast, femoral bone marrow uptake demonstrated decreased uptake, particularly over days 2-8, with the mean left to right uptake ratio decreasing 26% below mean baseline values (p=0.0005). Significant effects

181

were also seen in lungs and brain. Conclusions: Radiation had diverse effects on [18F]FDG uptake in previously healthy tissues. These kinds of data may help lay groundwork for a systematically acquired database of the time-course of effects of radiation on healthy tissues, useful for animal models of cancer therapy imminently, as well as interspecies extrapolations pertinent to clinical application eventually.

182

Introduction:

Positron-emission tomography (PET) is an expanding, noninvasive imaging technique frequently utilized for evaluating oncologic disease (117). It complements more conventional radiologic imaging techniques (i.e. CT, MRI), by looking at the functional or metabolic properties of suspected or confirmed tumor sites. More recently, evidence has also shown that fusion imaging with PET/CT significantly improves staging accuracy when compared to PET or CT alone (13, 81). Of the various radiotracers used for clinical indications, 18F-fluorodeoxyglucose ([18F]FDG) is the most widely employed. [18F]FDG uptake, often quantified as a standardized uptake value (SUV), has been shown to be elevated in many types of cancers relative to normal tissues (118).

In a recent review of the literature, Juweid et al. summarized how monitoring cancer treatment with PET contributed to tailoring an appropriate therapy regimen (117). In many studies, early metabolic changes measurable by [18F]FDG-PET were highly predictive of clinical responses observed weeks to months later. Such findings have been reported for a variety of cancers, including lymphoma, as well as breast, esophageal, gastric, colorectal, head and neck, and non-small-cell lung cancers (5, 40, 46, 57, 119125). Early declines in [18F]FDG uptake generally correlate with longer progression-free and overall survival.

183

The available data suggest that [18F]FDG may be utilized for predicting treatment responses as early as one to three weeks after the first cycle of chemotherapy in a variety of cancer types (5, 40, 57). This can prevent the prolonged exposure of patients to ineffective treatments with undesirable side effects.

Metabolic activity in tumors also often decreases after successful radiation therapy(126, 127). However, the ability of [18F]FDG-PET in monitoring the effects of radiation treatment has not been firmly established. This is in part due to the problem that, although [18F]FDG is an effective tumor-localizing tracer, it is not tumor-specific: benign processes (e.g. surrounding inflammatory changes, bone marrow suppression and hyperplasia) associated with irradiation also can alter [18F]FDG uptake levels. Hautzel et al. provided preliminary evidence of radiation-related inflammatory changes contributing to initial enhancement of [18F]FDG uptake by assessing metabolism of cervical lymph node metastases in a cancer patient during radiotherapy (126), reporting that low-dose irradiation enhanced tumor glucose uptake, while higher doses were associated with subsequent metabolic decline. More recently, Metser et al., in conducting a systematic review of PET/CT studies performed in oncologic patients during a 6-month period, discovered benign nonphysiological uptake of [18F]FDG in more than 25% of the studies. In half of these, [18F]FDG uptake was comparable to that of malignant sites, and most of the benign lesions were inflammatory in nature (128).

184

Differentiation of inflammatory processes from residual or recurrent disease is complicated, leading to imaging pitfalls such as false-positive readings and, consequently, administration of unnecessary therapy. Data from several recent studies suggest that PET can remain relatively nonspecific for up to 6 months following radiation therapy, due to inflammatory changes which may occur in the first few months after treatment (129).

In a field where treatment regimens often have success rates falling below fifty percent, improved methods for accurate, early prediction of treatment failure would be of substantial clinical value. The purpose of this study was to longitudinally characterize and quantify the time-course of [18F]FDG uptake in a variety of healthy tissues, occurring subsequent to irradiation, under experimentally controlled conditions, through use of noninvasive imaging with small animal-dedicated PET.

Methods:

Irradiation

All animal studies were performed under a protocol approved by the Chancellor’s Animal Research Committee of UCLA. Four male mice (strain C57BL/6) underwent microPET/CT imaging (PET acquired on a Siemens microPET® Focus™ 220 and CT

185

acquired on ImTek's MicroCAT™ II) prior to irradiation, to obtain a measure of baseline metabolic activity. Small animal PET and CT scans were acquired 1 hr after intravenous administration of 7.5 MBq (0.2 mCi) [18F]FDG on days 0 (pre-radiation), 1, 2, 3, 4, 5, 8, 12, 19, 24, and 38. Each mouse was irradiated with 12Gy of external beam radiation (max dose), in a single fraction to the left half of the body.

As the mice used in this experiment were small in size (on the order of 2cm in width across the thorax), great care was taken to deliver a dose distribution to provide a sharp dose falloff from the left side of the mouse to the right. A 6MV Novalis dedicated radiosurgery LINAC was used to deliver a posterior/anterior beam with a half-beam block. Additionally, a lead jig was created and placed directly above the mouse to further reduce the beam’s penumbra and subsequent dose received by the right half of the body Figure 8-1. Film dosimetry of the resulting field and the treatment planning system’s calculations were used to assess the dose falloff and determine required monitor units for a maximum point dose of 12 Gy. Additionally, a Monte Carlo simulation, using a model of the Novalis LINAC and a micro CT of one of the mice, was used to assess and quantify the resulting relative dose distribution in the irradiated mice. Resultant dosimetry from the Monte Carlo simulation is depicted in Figure 8-2, for an axial slice of a mouse CT scan. Metabolic activity, assessed with [18F]FDG small animal PET, in various tissues (i.e. lungs, femoral bone marrow, brain, and liver), was compared between the irradiated left and non-irradiated right body halves before, and at each time-point after, external beam radiation.

186

PET Acquisition

In this study, a total of 44 small animal PET and CT scans were acquired from four different mice. MicroPET/CT images were reconstructed using a filtered back projection algorithm (ramp filter, voxel size 0.04 x 0 .04 x 0.0796 cm3), and the biodistribution of [18F]FDG was assessed in regions of interest (ROIs) with use of the Amide software package (freeware available at http://amide.sourceforge.net). Uptake in irradiated tissue was compared with uptake in non-irradiated tissues. ROIs were obtained for left and right portions of each tissue assessed: lungs, femur, brain, and liver (Figure 8-3). Ratios of left to right uptake in ROIs were calculated for each mouse, for all trial days within the twomonth study period, by a single rater, to eliminate inter-observer variability.

Statistical Analysis

Time activity curves were examined for four organs, using the 11 scans acquired for each animal. Time windows used for statistical analysis were chosen by qualitatively selecting periods where a relatively consistent separation in the left-to-right ratios, relative to baseline data, were apparent on visual interpretation of time-course data (as reflected in Figure 8-4 through Figure 8-7, and in the fifth column of Table 8-1).

Relative uptake values in the analyzed time windows, reported as left to right uptake ratios for each area evaluated, were statistically assessed for significance by use of two-

187

tailed Student t-tests. Response patterns of [18F]FDG uptake in the liver, lungs, bone marrow of the femur, and brain were assessed. At baseline, no significant differences in uptake were found between left and right-sided tissues prior to irradiation (left:right ratios were 1.00 +/- 0.10, 1.08 +/- 0.05, 1.00 +/- 0.10, and 0.99 +/- 0.02, mean +/- SE for liver, lungs, bone marrow, and brain, respectively). Significance of changes in left to right ratio from 1 was assessed for times subsequent to administration of 12Gy external beam radiation.

Results:

Observed as early as the first day, irradiation had a significant impact on [18F]FDG uptake in previously healthy tissues (Table 8-1). The time-course of these effects differed dramatically, depending on the type of tissue examined (Figure 8-4 through Figure 8-7), with the percentage differences of left to right ratios relative to baseline increasing or decreasing from 5% to over 50%.

Liver

Irradiation of the left liver resulted in higher [18F]FDG uptake than in the non-irradiated right side. This effect peaked on day 8, when the left to right ratio was 100% greater than at baseline (p<0.0001), and was most apparent on days 3-12, over which time the left to right ratio averaged 52% higher than at baseline (p<0.0001). Figure 8-4 illustrates the

188

time-course of these effects, with each data point representing the mean [18F]FDG uptake in four mice on each scan day.

Lungs

Irradiation also resulted in higher [18F]FDG uptake in the irradiated left lung compared to the non-irradiated right lung. This rise in the mean left to right uptake ratio was observed as early as day 1, and peaked on day 12, before returning to baseline levels. Figure 8-5 illustrates the time-course of these effects within a 2-month period. The change in mean left to right uptake ratio post-irradiation was found to be statistically significant, resulting in a rise of 16% relative to baseline, averaged over days 1-24 (p<0.0001). It is noteworthy that, at baseline, the lungs demonstrated slightly higher uptake in the left lung relative to the right, most likely due to cardiac spillover. Thus, each mouse was also statistically analyzed after being normalized to its own baseline, and results remained significant, resulting in an increase in ratio of 7% (p=0.01).

Femur

Irradiation decreased the mean left to right uptake ratio in the femur, which was most prominent on trial days 2-8. The most significant decrease was observed on day 8, when uptake was 40% below baseline values (p<0.05). Figure 8-6 illustrates the time-course of these effects over a 2-month period. Over days 2-8, the left to right ratio averaged 26% lower than at baseline (p=0.0005).

189

Brain

As observed in the femur, irradiation decreased the mean left to right uptake ratio in the brain, which was observed on all trial days post-irradiation, again most significant on day 8. Figure 8-7 illustrates the change in 18F-[18F]FDG uptake in the irradiated left brain compared to that of the non-irradiated right brain, resulting in a 5% decrease relative to baseline, averaged over post-radiation days 1-24 (p<0.0001).

Discussion:

In the present study, we systematically documented the direction, magnitude, and timecourse of radiation-induced changes occurring in a variety of tissue types. While the irradiated liver and lungs demonstrated increases in [18F]FDG uptake in the days following irradiation, irradiated femoral bone marrow and brain demonstrated decreases in [18F]FDG uptake during that period. Effects ranged from 5% to over 50% changes in uptake relative to the pre-irradiated baseline, and, each tissue type exhibited a distinct time-course of uptake over a two-month trial period.

In the femur and brain, we observed a decrease in the irradiated/non-irradiated tissue uptake ratio following radiation. The declining uptake in the femur is understandable in the context of previously documented responses (130-132) that bone marrow is highly sensitive to radiation, and decreased [18F]FDG uptake may be a result of functional

190

suppression following radiation. In the brain, [18F]FDG uptake is also not surprising, given that the immune system has less access to brain tissue than to the lung and liver and other tissues, due to the blood-brain barrier(133-135), coupled with normally high rate of glucose metabolism which occurs in the brain at baseline (110, 136), and which can be disrupted by the synaptic dysfunction occurring subsequent to irradiation.

In the lungs and the liver we observed an increase in the irradiated/non-irradiated uptake ratio following radiation. This increase most likely results from an inflammatory response in these tissues (137, 138). Specifically, early inflammation in the lung may stem from the immediate expression of the pro-inflammatory cytokines TNF-alpha, IL1alpha, and IL-6 in the bronchiolar epithelium in the first hours after lung irradiation (139). In the liver, the high levels of inflammation may result from high levels of oxidative stress, as reflected in some studies by elevated levels of peroxidative damage, DNA fragmentation, LDH activity, and nitric oxide levels (140).

We have characterized the time-course of effects of radiation in various healthy tissues from which cancer may arise. Although [18F]FDG-PET is commonly employed for monitoring responses to chemotherapy (5, 40, 57), it has been less utilized in monitoring effects following irradiation. While the exact mechanisms and extent of metabolic responses in healthy tissues have not yet been well defined, interpretation of [18F]FDG uptake can be substantially complicated by radiation-induced inflammation and other effects occurring in surrounding tissues. As discussed by Engenhart et al., it is often

191

difficult to distinguish the difference in [18F]FDG uptake before and after irradiation, as it does not reliably differentiate among proliferation, repair, inflammation, and residual viable tumor cells in patients with inoperable recurrent rectal carcinoma (141). Data established in the present study may be placed in the context of other published studies that have investigated irradiation effects. Ohtsuka et al. investigated non-small-cell lung cancer after neoadjuvant chemoradiotherapy, and found positive [18F]FDG uptake in PET scans despite absence of tumor cells found pathologically (142). Such false positives are thought to be due to either inflammatory lesions with invasion of macrophages and lymphocytes resulting in increased uptake of [18F]FDG (143-145), or metaplastic and proliferative epithelial elements caused by chemoradiotherapy leading to [18F]FDG accumulation (146). Similarly, in our study, [18F]FDG PET demonstrated increased metabolic activity in the liver and lungs. However, not all research has found significance in the post irradiation changes in PET in the organs we have looked at. Castellucci et al. investigated the rate of postactinic inflammatory alterations leading to potential falsepositive PET images in lymphoma patients in hopes of determining an optimal time window between radiation therapy and [18F]FDG-PET; they found that the incidence of inflammation shortly after radiation therapy was not as prevalent as they had expected it to be, and they were unable to establish a strong link to the elapsed time since the end of radiation therapy treatment (147). More research is clearly needed in this area.

In summary, results from our present study indicated effects of tissue irradiation ranging from 5% to over 50% changes in uptake relative to the pre-irradiated baseline, with

192

different tissue types exhibiting distinct time-courses of uptake over a two-month observation period. Limitations to our study include the difficulty of administering a uniform radiation dose across mice, as dose depends on size, shape, and composition of the irradiated subject as well as technical parameters of the linear accelerator. To account for this, we classified dose distribution by utilizing a Monte Carlo simulation, which utilizes a computer model to make iterative predictions about how the radiation was able to be delivered, especially for the left/right comparison.. We utilized mice of the same body weight and age for our study, to obtain as homogenous an effect of irradiation as possible.

It is also important to recognize that different doses and forms of irradiation may yield different time-courses of post-radiation effects. What our results may provide is initial insight into the relative magnitudes of biological effects following irradiation. These preliminary findings of the diverse effects of irradiation in healthy tissues could be useful for animal models of cancer therapy (e.g. xenograft models) and provide a point of reference for further studies aimed at trying to delineate and quantify uptake in tumors and their associated tumor to background ratios. Actual rates of metabolism will also need to be established in humans, as it is common for physiological and pathological processes to be accelerated in mice relative to normal reactions in people(93). Translating these processes to the clinic can potentially aid in the differentiation of inflammatory processes from that of residual or recurrent disease. In PET, lesion characterization is often heavily dependant on lesion-background uptake ratio. Recent

193

literature (82, 127, 148) has suggested that a 20% change in this ratio is clinically significant. However, lesion detection can depend on differences ranging within a few percent. Thus the extent to which radiation impacts this ratio can have direct implications on clinical diagnosis. Both PET and radiation are largely utilized clinically, and further study may expand the role for PET to play for radiation treatment monitoring, as is currently starting to be explored in the clinic(149-151). Examining other radiotracers with this experimental design is also of interest, as difference radiotracers may behave differently, during radiotherapy.(136, 152).

Conclusion:

Different tissues have different metabolic profiles with respect to the direction, magnitude, and time-course of changes occurring after irradiation. We saw increased FDG uptake following radiation in the lungs and liver, while we noticed the opposite effect in the brain and femur. Time courses and rates of reactions varied among these tissues, likely reflecting the variety of biological processes encountered when combining radiation treatment with FDG PET imaging. Data from studies such as this one may help in designing animal models of monitoring tumor responses to irradiation imminently, as well as, ultimately, in translating the findings to optimizing clinical therapeutic monitoring.

194

Figures

Figure 8-1: Images illustrating mouse irradiation setup (at linear accelerator)

195

Figure 8-2: Dosimetry visualization. (A) Monte-carlo estimate of dose distribution for a mouse receiving radiation to the left half of the body from a 6 mV linear accelerator. The distribution represents an axial slice of the mouse, just inferior to the lungs. (B) Illustration portraying the location of the dose calculation shown in figure.

196

Figure 8-3: ROI visualization. Display of hand-drawn ROIs (displayed in orange) for lungs (A), femur (B), brain (C), and liver (D). For each area assessed, ROI’s were drawn using the Amide software package, and the uptake in the irradiated left tissue was compared with uptake in non-irradiated contralateral tissue.

197

Left / Right Liver Ratio 3

Mean Ratio

2.5 2 1.5 1 0.5 0 0

10

20 Day

30

40

Figure 8-4: Time-course of mean [18F]FDG uptake ratio in liver. Each data point represents the mean left:right ratio of uptake values calculated for four mice. The +/standard error is indicated with dashed bars. A thick grey line corresponds to the mean for the range of dates indicated in Table 8-1 (days 3-12 for liver). The thin grey line represents a hand drawn curve overlaid on the data points.

198

Left / Right Lung Ratio 1.4

Mean Ratio

1.3

1.2

1.1

1 0

10

20 Day

30

40

Figure 8-5: Time-course of mean [18F]FDG uptake ratio in lungs. Each data point represents the mean left:right ratio of uptake values calculated for four mice. The +/standard error is indicated with dashed bars. A thick grey line corresponds to the mean for the range of dates indicated in Table 8-1 (days 1-24 for lungs). The thin grey line represents a hand drawn curve overlaid on the data points.

199

Mean Ratio

Left / Right Fem ur Ratio

1.5 1.4 1.3 1.2 1.1 1 0.9 0.8 0.7 0.6 0.5 0.4 0

10

20 Day

30

Figure 8-6: Time-course of mean [18F]FDG uptake ratio in femur. Each data point represents the mean left:right ratio of uptake values calculated for four mice. The +/standard error is indicated with dashed bars. A thick grey line corresponds to the mean for the range of dates indicated in Table 8-1 (days 2-8 for femur). The thin grey line represents a hand drawn curve overlaid on the data points.

200

Left / Right Brain Ratio

1.1

Mean Ratio

1.05 1 0.95 0.9 0.85 0

5

10

15

20

25

30

35

40

Day

Figure 8-7: Time-course of mean [18F]FDG uptake ratio in brain. Each data point represents the mean left:right ratio of uptake values calculated for four mice. The +/standard error is indicated with dashed bars. A thick grey line corresponds to the mean for the range of dates indicated in Table 8-1 (days 1-24 for brain). The thin grey line represents a hand drawn curve overlaid on the data points.

201

Tables

Table 8-1: Summary of mean [18F]FDG uptake ratios observed in four different types of tissue. Irradiation had varying effects on [18F]FDG uptake in previously healthy tissues.

Average Period of magnitude Direction

Time to

Magnitude

most

p-value of change

Tissue

of peak

peak

of peak

apparent

(twoduring

change

change

change

effect of

tailed) noted

irradiation period Liver

↑

8 days

100%

Days 3-12

52%

<0.0001

Lungs

↑

12 days

15%

Days 1-24

7%

0.0127

Femur

↓

8 days

40%

Days 2-8

26%

0.0005

Brain

↓

8 days

10%

Days 1-24

5%

<0.0001

202

Project combinations

9. Radiodosimetry extrapolated from semiautomated analysis of small animal FDG PET scans Abstract The main goals of our project were to develop and validate automated methods to estimate rodent radio-dosimetry using dynamic small animal PET images, and to use the data thus generated to accurately project effective organ dose equivalents for radiation exposure in humans. Methods: Using our image fusion software and the calculated FDG biodistributions in 9 mice previously generated (Chapter 1), we extrapolated the rat biodistribution data to humans. Using this data we applied dose calculation software to obtain human organ dose estimates. As a check of accuracy for our methods we compared our organ dose estimates with two sets of previously published FDG human dosimetry sets. Results: The Pearson correlation coefficient for our human organ dose estimates ranged between 0.75 and 0.97 with a median of 0.95 for the first standard set and between 0.61 and 0.92 with a median value of 0.77 for the second. Conclusion: We have shown our methods to both define the bio-distribution of PET tracers in rodents and to estimate human FDG dosimetry can yield results comparable to those obtained with standard methods. Such dose estimation techniques have the potential to be used in new tracer development with the possibility of saving time, effort, money, and the number of specimens needed for experimentation.

203

Introduction The main goals of the work reported here were to develop and validate methods utilizing rodent radio-pharmaceutical bio-distribution derived from small animal PET and acquired in a semi-automated manner (Chapter 1), to generate predicted radiation organ doses and effective dose equivalents in humans (Chapter 2). The methods presented here were calibrated with and tested against previously published determinations of radiodosimetry in humans.

Methods

Achieving our main goals involved extrapolating the activities of the mouse organs to estimate the activity and biological distribution a human would have, given the same radio-tracer, and with an estimated activity distribution within a human, utilize nuclear medicine dose estimation software to find the expected dose for all defined organs of interest.

By fusing a three-dimensional digital mouse phantom with a small animal PET mouse volume we were able to couple known anatomical geometry from the phantom with

204

activity distribution from the scan. This allowed us to quantify activity concentrations for regions of interest. Nine mice were imaged with 18F-FDG (10-15 MBq) as described in chapter 1, and their FDG biodistributions were extrapolated from mice to humans using Equation 6-1.

With those data we applied the OLINDA/EXM software (104) to obtain human organ dose estimates (as described in chapter 6). Our dynamic small animal PET acquisitions lasted for 80-100 minutes following administration of the radionuclide, and any further distribution was considered static. A non voiding bladder model was used and the only clearance considered was the physical decay of 18F (half-life = 109.8 min).

As a check on the accuracy of human dosimetry estimates generated by our program, we compared organ dose estimates for FDG that we obtained from nine different small animal PET mouse scans using our methods with two sets of previously documented FDG human dosimetry sets. The first set was taken from the Radiation Internal Dose Information Center (153) which used the MIRD technique implemented with the MIRDDOSE3 software, and was based on distribution data gathered in dogs (154) and in humans (155). This study used a dynamic bladder model with 4.8-hour voiding interval. The second set (156) used the MIRD technique according to the procedures outlined in MIRD Pamphlet No. 11 (157), and was based on a collection of published organ residence times combined with previously published mathematical models. This bladdervoiding model was based on 120-minute void intervals.

205

Results We compared human FDG organ dose estimates that we extracted from small animal PET scans using the presented methods to previously published FDG human dosimetry estimates.

We examined nine FDG small animal PET scans, acquired with nine non-fasted mice. The resulting dose estimates are shown in Figure 9-1 and Table 9-1, in units of mGy / MBq injected. Among the three sets of dosimetry estimates (our average predictions and the two standard sets) there was no set that had consistent shift towards higher or lower dosimetric estimations, indicating the three sets had similar total body estimates.

The correlation between our average experimental dosimetry estimates and the two standard sets is displayed in Table 9-2. The correlation coefficient (r) for our nine estimates versus the previously established standards ranged from 0.73 to 0.96 with a median of 0.91 for the first standard set, and from 0.61 to 0.81 with a median value of 0.73 for the second standard set. When the average dose to each organ across all nine mice was compared with the literature standard sets the correlation coefficients were 0.939 for the first set and 0.763 for the second set. One should bear in mind that the two standard sets used are also estimations rather than absolute quantities. The Pearson correlation coefficient (r) between the two standard dosimetry sets is 0.765. This value is lower than many of the comparisons we had between our estimates and each of the sets. The correlation differences between these three methods reflect that contemporary

206

nuclear medicine dosimetry has a large number of variables and includes a substantial amount of estimation when dealing with even the simplest human case: “the standard man”.

The ratios of the mean organ dose estimates to those of the standard sets were compared (Table 9-3). We would expect these ratios to be close to unity, as all the prediction sets are of absorbed dose normalized by one mega-Becquerel of injected activity. We see in the table that these ratios peak and dip above and below unity, mostly staying within a factor of 2, with a few points straying by as much as a factor of 4, illustrating the scope of precision available for reproducing individual organ dose estimates, for all the methods presented.

Discussion

To achieve human radiation dosimetry estimates for positron emitters, which must account for the entire period of above-background exposure following radiopharmaceutical administration, it is necessary to measure the time-course of radiopharmaceutical concentrations within each organ of interest, so that the overall spatiotemporal distribution can be projected. Approaches to make these measurements have included the utilization of thermo-luminescent dosimeters (TLDs), which have been demonstrated to yield reasonable activity distribution estimates (158), or alternately the reading of activity concentration levels directly from whole body human PET scans.

207

In our approach to dose calculation we first had to design software to calculate SUV values in dynamic small animal PET mouse scans. We used an approach of fusing a digital mouse phantom with a small animal PET mouse scan, generating predefined ROIs overlaid on the image. Using an approach like this to generate SUV measurements has some advantages: After the image fusion is concluded all organ ROIs (22 organs) can be rapidly defined. Also by fitting a phantom atop a PET scan, we can find ROI activity values and time activity curves for regions or organs that were not manually definable on the original PET image.

We extrapolated activity data from an FDG injection of a mouse to a human. By extrapolating between species using a percent kilogram dose per gram unit we are not accounting for metabolic differences in uptake between the two species in our estimation. There have been studies nevertheless confirming the predictability of radiation risks with interspecies extrapolation (153, 159), indicating the general feasibility of the approach.

Also, we assessed how well our predictions for human FDG dosimetry agreed with previously published estimates. We obtained nine human dosimetric predictions for FDG, which we compared to two sets of previously established dose estimates. The correlation coefficients between our estimations and the standard sets ranged between 0.61 and 0.96. This level of correlation can be compared to the 0.765 correlation between the two sets used as standards.

208

Conclusions The purposes of our work were to use computing power and small animal PET technology to help calculate the activity distribution in small animal PET mouse scans, and to further take this information and use it to extrapolate human organ dose estimates.

We have shown that the application of our methods to determine human organ FDG dosimetry estimates for a standard man can generate reasonable correlation with estimates previously published. Such dose estimation methods can be used as a quick gauge of an experimental radiotracer’s dosimetry, and have the potential to be used in new tracer development with the prospect of saving time, money, reducing the number of animal specimens needed for experimentation, and helping to best guide research directions.

209

Ad To ren ta als lB od y Sk Br in ea Th sts O y st eo M roid ge us ni cl c e C St T ells om hy ac mu h s U Wa R LI ll ed W M al l ar G r al lb T ow la es Sm dd te al er W s lI nt all es tin Li e LL ver IW Lu all O ngs va r ie Br s Ki ain dn e U ys Pa ter nc us re U rin Sp as ar H le y ea en Bl r t ad W de al rW l al l

Dose (MGy/MBq)

Figures FDG Dose Estimates

0.2

0.18 0.16

0.14 0.12 Mean Estimated

Standard #1

Standard # 2

0.1 0.08

0.06 0.04

0.02 0

Organ

Figure 9-1: Estimated human organ activities following FDG injection (mGy / MBq)

210

Tables

Table 9-1: Estimated human organ activities following FDG injection (mGy / MBq) Predictions

Adrenals Brain Breasts Gallbladder Wall LLI Wall Small Intestine Stomach Wall ULI Wall Heart Wall Kidneys Liver Lungs Muscle Ovaries Pancreas Red Marrow Osteogenic Cells Skin Spleen Testes Thymus Thyroid Urinary Bladder Wall Uterus Total Body

Standard #1

Standard # 2

Mean

Median

Std Dev

0.00E+00 1.90E-02 9.20E-03 1.40E-02 1.70E-02 1.40E-02 1.30E-02 1.30E-02 6.00E-02 2.00E-02 1.60E-02 1.70E-02 1.10E-02 1.70E-02 2.60E-02 1.30E-02 1.20E-02 8.40E-03 3.70E-02 1.30E-02 1.20E-02 1.00E-02 1.90E-01 2.30E-02 0.00E+00

0.00E+00 4.60E-02 0.00E+00 0.00E+00 0.00E+00 0.00E+00 0.00E+00 0.00E+00 6.80E-02 2.10E-02 2.40E-02 1.50E-02 0.00E+00 1.10E-02 1.40E-02 1.10E-02 0.00E+00 0.00E+00 1.50E-02 1.10E-02 0.00E+00 0.00E+00 7.30E-02 0.00E+00 1.20E-02

1.30E-02 1.02E-02 9.60E-03 1.32E-02 2.39E-02 2.25E-02 1.56E-02 1.95E-02 4.79E-02 1.76E-02 1.22E-02 3.38E-02 1.10E-02 1.57E-02 1.35E-02 1.07E-02 1.60E-02 8.21E-03 1.25E-02 2.59E-02 1.34E-02 1.05E-02 1.03E-01 1.86E-02 1.18E-02

1.29E-02 1.00E-02 9.46E-03 1.30E-02 2.24E-02 2.17E-02 1.59E-02 1.83E-02 4.94E-02 1.74E-02 1.21E-02 3.20E-02 1.09E-02 1.58E-02 1.35E-02 1.07E-02 1.59E-02 8.16E-03 1.26E-02 2.38E-02 1.36E-02 1.05E-02 1.17E-01 1.89E-02 1.18E-02

3.00E-04 1.70E-03 3.95E-04 4.82E-04 4.42E-03 4.34E-03 1.17E-03 3.40E-03 2.02E-02 2.71E-03 5.51E-04 6.95E-03 1.73E-04 1.12E-03 2.35E-04 1.45E-04 4.03E-04 1.76E-04 8.49E-04 1.02E-02 1.74E-03 2.60E-04 2.39E-02 1.54E-03 7.26E-05

211

Table 9-2: Correlation coefficients for organ dose estimates Standard #1 Standard # 2 r r Mouse Correlation p value Correlation p value Coefficient: Coefficient: 1 0.913 0.000 0.791 0.000 2 0.906 0.000 0.619 0.002 3 0.961 0.000 0.729 0.000 4 0.945 0.000 0.811 0.000 5 0.866 0.000 0.784 0.000 6 0.879 0.000 0.616 0.002 7 0.932 0.000 0.726 0.000 8 0.733 0.000 0.811 0.000 9 0.932 0.000 0.726 0.000 Mean 0.896 0.000 0.735 0.000 correlation Median 0.913 0.000 0.729 0.000 correlation

212

Table 9-3: Mean ratio of experimental / expected dose per organ

Adrenals Brain Breasts Gallbladder Wall LLI Wall Small Intestine Stomach Wall ULI Wall Heart Wall Kidneys Liver Lungs Muscle Ovaries Pancreas Red Marrow Osteogenic Cells Skin Spleen Testes Thymus Thyroid Urinary Bladder Wall Uterus Total Body Average

strd# 1/ exp/str Ratio exp/str Ratio d#1 std dev d#2 std dev strd# 2 53% 9% 22% 3% 41% 104% 4% 94% 140%

4% 31%

-

-

-

161%

38%

-

-

-

120% 150% 80% 88% 76% 199% 100% 93% 52% 82%

11% 31% 37% 11% 4% 48% 2% 7% 1% 1%

70% 84% 51% 225% 143% 96% 97%

28% 12% 2% 44% 10% 2% 1%

88% 95% 67% 113% 155% 186% 118%

133% 98% 34% 199% 112% 105%

4% 3% 2% 53% 13% 3%

83% 235% -

5% 88% -

247% 118% -

54% 81% 105%

12% 6%

141% 99% 112%

31% 1% 19%

260% 135%

15%

213

10. Biodistribution of 5-FU in Rats measured using Semi-automated software and estimated attenuation correction Project Description While developing our methods for the 5-FU paper (chapter 7), we tried an array of different approaches for using our PET data to derive radiodosimetry estimates. ROIs were drawn using different methods (below) and resultant biodistributions (as described in chapter 7) are presented in Figure 10-1.

Image – harvest calibration: Manual ROI definition consisted of a researcher who manually drew ROI on the PET scans using Amide software, and calibrated using harvested data. This was presented as one of the main methods in chapter 7.

Image – image calibration: Manual ROI definition consisted of a researcher who manually drew ROI on the PET scans using Amide software, and calibrated using the scanner calibration. This was presented as one of the main methods in chapter 7

Image – image calibration with attenuation correct definition:

214

These biodistributions were corrected for expected attenuation which was not corrected for in PET data. Correctional factors were estimated to be as high as 180% for some organs. Our PET images did not have associated attenuation maps, so we estimated the attenuation correction factors by scaling a MOBY phantom (18) to the size of a rat (250 grams). The emission image of the phantom was then forward projected with an attenuation volume, and then backprojected (filtered back projection – ramp filter). Attenuation correction factors were calculated by comparing the emission volume and the reconstructed volume. These standard attenuation correction factors are summarized in Table 10-1, and were applied to the PET data to create pseudo attenuation corrected PET biodistributions.

Software image definition: Software ROI definition utilized the semi-automated ROI calculation software presented in chapter 0. This necessitated a newer version of the software to be created, which was able to stitch together bed frames for the multiple-framed rat PET scans. Additionally, the mouse atlas (based on the Moby phantom) was stretched to approximate an atlas for a rat.

Harvested definition

215

Harvested organ uptake measurements consisted of a researcher who manually extracted organs from the subjects at varying time points following injection, and is explained in further detail in chapter 7.

216

Figures 5-FU Activity Distribution in Rats Measured With Different Methods

0.80000 0.70000 Percent Total

0.60000 0.50000 0.40000

Image - harvest calibration Image - image calibration Image - image calibration with AC Software Image Harvesting

0.30000 0.20000 0.10000

Lw

rC ol on He + C ar t W B nt a l rai l+ n Cn Ri t Sm Sp bs al l lI nt S een es p tin in e+ e c Lu nt Sk ng el s St e om K ton a id Bl ch ne ad +C y de N r+ T Cn Li t ve Bo r dy

0.00000

Organ

Figure 10-1: Biodistributions of 5-FU measured using different methods.

217

Tables Table 10-1: Attenuation correction factors at 511 KeV for standard mouse phantom scaled to rat volume. (* indicates values used in measurements)

Heart Liver Thyroid Bladder Colon Lungs Spleen Kidneys SmInt Brain Stomach

250 gm rat* 1.65 1.68 1.50 1.75 1.46 1.36 1.41 1.66 1.46 1.34 1.64

218

350 gm rat 1.76 1.79 1.58 1.87 1.53 1.42 1.48 1.77 1.53 1.39 1.74

11. Fully Automated Respiratory Gating and Recombination of Gates in Clinical Human PET

Project We have developed methods to both extract respiratory signal from raw PET data, as well as methodology to combine gated images to create images superior to ungated or gated images alone. Both methods work in a fully automated manner, and require no additional hardware or user input.

We have described in detail the respiratory gating and gate combination work in their own respective sections. Following the developments of that methodology, we applied our methods to our human scan data. The level of success we found provides both proofof-principal for our methods as well as promise for better images, as we know with further studying, there are several places to tune and develop our methods.

Implementation In PET imaging, there are three main steps for creating respiratory gated images and recombining them: 1. Acquire a respiratory time-amplitude trace corresponding to patients respiratory cycle and the PET data

219

2. Sort the PET data into differently phased bins (relative to analysis of trace), and reconstruct images (choosing algorithm) 3. Recombine the separated gate images (to recover lost statistics)

Our work which we are presenting focused on the first and third steps in the process as defined above. To generate respiratory gated images with our data, we approached the steps as follows:

Respiratory traces were acquired for selected human scans, using our data-based respiratory gating methodology (chapter 4).

The timing of the respiratory gates for which to sort the raw PET data was derived from the respiratory trace. Peaks (triggers for a new cycle) were defined on the respiratory trace (corresponding to an end of the respiratory cycle), by finding points in the trace that were maxima within a ± 1 second window (i.e. local maxima). Once these points were defined, the time bins for each of the 8 gates were linearly interpolated such that data was binned as a percent time passed between triggers (example in Figure 11-1). Once the raw listmode PET data was separated into bins, images were reconstructed using an OSEM algorithm (2 iterations, 8 subsets).

With a gated image set, a recombined gated image set was constructed using our presented image recombination methodology (attached).

220

Results and Discussion

Five human scans were processed and reconstructed as ungated, gated, and combined gated. Sample planes from each patient are seen in Figure 11-2, Figure 11-3, Figure 11-4, Figure 11-5, and Figure 11-6

Statistics characterizing the differences in the images are summarized in Table 11-1, Table 11-2, Table 11-3, Table 11-4, and Table 11-5. Lesion profiles are illustrated in Figure 11-7. The mean lesion full width half max (FWHM) relative to the ungated lesion measurements were 76% for the gated images, and 78% for the gated combined images. The FWHM measurements are summarized in Figure 11-8. The mean signal-to-noise ratio relative to the ungated image was 59% for the gated images, and 80% for the gatedcombined images. We also took measurements of the signal contrast-to-noise in the images, defined as mean lesion uptake divided by the standard deviation in the lungs. A summary of the contrast-to-noise measurements be found in Figure 11-9. Thus, combined-gated images offered a comparable resolution to the gated images with a stronger signal-to-noise.

As a second check that lesion movement was presented in the gated scans, and that gating methods were correctly working, the center of mass was measured for each patient’s

221

respective lesion (as described in gate recombination chapter). A summary of the measured lesion displacements for each gate can be found in Figure 11-10, with sinusoidal motion illustrating the success of the respiratory gating.

Using our methods, we were able to take raw PET data from actual human PET data, and create superior images corrected for respiratory motion, using fully automated algorithms.

When assessing the gated, ungated, and combined gated images (Figure 11-2, Figure 11-3, Figure 11-4, and Figure 11-5), it is not difficult to notice differences in relative noise levels and improvements from the gated images to the combined gated and ungated images. Differences in resolution among the data sets are harder to see qualitatively in the presented images. For the patient scan in Figure 11-2 the resolution of hot spots appears to slightly improve in the combined gated image relative to the ungated image, but has no discernable improvement for the patient scan in Figure 11-3. Figure 11-6 offers a unique example of resolution effects and gating on detectability. In the ungated image we see that the patient appears to have two hot spots in a close proximity to one another. The gated image offers limited improvement for detectability because the increased noise perturbs the contrast. However the combined gated image offers lesion contrast on par with the ungated image, but sharper resolution for all of the hot spots.

222

Improvements in resolution, or lack thereof, may stem from inaccurate gating of the patient data, (discussed in Chapter 4). For the patients with poorly gated scans, the objective for the gate combination algorithm would be to discern the lack of benefit for the gating of the data, and to weight the final image towards the ungated data. We can see in Figure 11-3 that this is what is happening, because the gated scans here offer little benefit, the algorithm is weighting the data towards the ungated value (as is apparent in the C-g weight distribution images).

The C-g weight distribution, shown in the bottom row of Figure 11-2, Figure 11-3, Figure 11-4, and Figure 11-5, offers insight into how the algorithm is individually weighting voxels between gated and ungated voxel values. We can see in the figures, that the regions corresponding to the location of lesions had clusters of voxels, weighted towards the gated image data; regions where we can expect the gating to be advantageous. The presence of these regions indicate that the algorithm determined the gated data set useful enough to utilize to create a improved combined-gate images. It is also interesting to notice the noise in the weighting factors in the regions weighted towards the ungated values. This stems from a limitation in the algorithms ability to precisely determine a voxel’s weighting, and can potentially be made more uniform with an improved weighting factor (Chapter 5).

Characterizing the relative resolution properties for the ungated, gated, and combined gated values qualitatively was not as easy as characterizing the noise. Figure 11-7 and

223

Figure 11-8 illustrate lesion profiles for the patient scans shown in the presented images, and offer insight into some of the resolution characteristics. We can see for all of the patients that lesion profiles for the combined gated scans offered sharper peaks relative to the ungated profiles, although not as sharp as the gated profiles, and this observation is reiterated in the FWHM measurements. It should be noted though that this assessment of resolution is limited as it only accounts for a line profile along one line of evaluation.

Our contrast-to-noise measurements (Figure 11-9) illustrate the same type of behavior as the resolution measurements, combined-gated images offer a balance between the CNR measurements for the gated or ungated data alone. When using these numbers for evaluation there are several things to note and limitations to understand. Both the numerator (mean activity in lesion) and the denominator (standard deviation of lung ROI) changes for ungated, gated, and combined gated scans, and effects the CNR measurements in competing directions. Because the gated images have more noise, the mean value of the lesions (as defined as top 80% of voxels with ROI) was pushed higher, while the standard deviation in the lungs was higher as well (because of lower statistics). It is for this reason the CNR measurements alone are not enough to characterize the signal improvements and contrast A second limitation of this CNR measurement is in the use of the standard deviation of the lungs in the denominator. The lungs were chosen as they were easily recognizable in all of the scans. The drawback of using the lungs is that because the scans were not corrected for attenuation, there is uneven statistics present for the different regions and different patients.

224

Because of the limitations of the FWHM and CNR values to accurately describe quantitative characteristic of the scans, tables of various data are presented (Table 11-1, Table 11-2, Table 11-3, Table 11-4, Table 11-5). We can see in the tables, for example, that the mean lesion over mean lung measurements relative to the ungated data ranged between 1.01-1.12 for the ungated data, and 1.01-1.09 for the combined gated data, a aesthetically small improvement in the images. We can also see that the percent standard deviation was consistently lower for both the lesions and the lungs in the ungated images relative to the gated images, and higher in the combined-gated images relative to the ungated images. Overall the tables are provided to help gauge the effects of using the different data sets .

The processing time for constructing combined-gated images from raw list mode PET data took approximately 16 hours using a single 3.3 GHz PC for a ten minute scan. Faster processing time can be achieved using different reconstruction algorithms, clustering computer power, and method refinement.

The methods we developed compromise the first and last step in the gating process. Thus the overall improvements in the images seen in the images presented here, have been limited by the quick implementation of dividing the respiratory trace (step 2 in “implementation”), a step we did not focus too much on, however there is literature supporting other approaches.

225

The two algorithms we are presenting do compliment each other nicely in that the gate combination (final step in process) we developed protects against image degradation resulting from poor gating technique, low count statistics, or any other factor which would make gating non-beneficial, thus providing assurance against poorly generated images.

Overall the gated and combined-gated images offer increased lesion detectability and characterization, thus potentially benefiting patients. We have shown proof of principle by successfully gating and recombining data from five human scans. The algorithms are fully automated, and in contrast to newly emerging market technology, require no hardware or user input. Both the algorithms, as well as the combination of the algorithms, have yet to be refined and maximized to reach their potential. The specific improvements for the algorithms are discussed more in their respective sections. Additionally, promise comes from the implementation of phased-based attenuation correction, and from the fact that these methods will only get better as sensitivity and resolution of commercial PET scanners improve.

226

Figures

Binning of Trace 8 6 4

Amplitude

2 0 60

70

80

90

100

-2 -4 -6 -8 -10

Time (s)

Bin

Trace

Figure 11-1: Example of corresponding respiratory trace and gate relationship.

227

110

120

Coronal

Transverse Ungated

Gated

Combinedgated

C-g weight distribution

Figure 11-2: Ungated, gated, and combined-gated image from patient #2 PET data.

228

Sagittal

Coronal

Transverse Ungated

Gated

Combinedgated

C-g weight distribution

Figure 11-3: Ungated, gated, and combined-gated image from patient #4 PET data.

229

Sagittal

Coronal

Transverse Ungated

Gated

Combinedgated

C-g weight distribution

Figure 11-4: Ungated, gated, and combined-gated image from patient #5 PET data.

230

Sagittal

Coronal

Transverse Ungated

Gated

Combinedgated

C-g weight distribution

Figure 11-5: Ungated, gated, and combined-gated image from patient #9 PET data.

231

Sagittal

Coronal

Transverse Ungated

Gated

Combinedgated

C-g weight distribution

Figure 11-6: Ungated, gated, and combined-gated image from patient #11 PET data.

232

Sagittal

Patient #4 Lesion Profile

Relative Amplitude

Relative Amplitude

Patient #2 Lesion Profile

0

20

40

60

80

100

120

140 100

mm

110

130

140

150

160

mm

Patient #9 Lesion Profile

Relative Amplitude

R elative A mplitude

Patient #5 Lesion Profile

0

10

20

30

40

50

60

70

80

Relative Amplitude

Patient #11 Lesion Profile

60

70

80

50

70

90

110

mm

mm

50

120

90

100

110

120

mm

Figure 11-7: Profiles of lesions measured in human scans.

233

130

150

170

FWHM Measurements 25.00 Ungated

20.00

Gated Gated-Combined

mm

15.00

10.00 5.00

e

Ph an to m

n m ul

at io

Sp he r

Scan

Si

nt #1 1 Pa tie

nt #9 Pa tie

nt #5 Pa tie

nt #4 Pa tie

Pa tie

nt #2

0.00

Figure 11-8: Summary of FWHM measurements taken from profiles of lesions in patients. Simulation and phantom data are also included (on right). (FWHM data not included for patient #9 as it is not a suitable measure in this instance).

234

CNR Measurements 90.00 80.00

Ungated

70.00

Gated

CNR

60.00

Gated-Combined

50.00 40.00 30.00 20.00 10.00

e

Ph an to m

n at io

Sp he r

m ul Si

nt #1 1 Pa tie

nt #9 Pa tie

nt #5 Pa tie

nt #4 Pa tie

Pa tie

nt #2

0.00

Scan

Figure 11-9: Contrast-to-noise measurements from patients and simulations, for ungated, gated, and combined-gated measurements.

235

Center of Mass Displacement Displacement [mm]

14

12

Patient #2

10

Patient #4 Patient #5

8

Patient #9 Patient #11

6

4

2

0 0

1

2

3

4

5

6

7

8

Gate Figure 11-10: Displacement of lesion center of mass measurements measured from 5 patients studied.

236

9

Tables Table 11-1: Human #2 Gating Improvement Statistics

mean

mean

lesion

lesion /

/SD

mean

lungs

lungs

%SD of Mean

Std Dev

% SD

Gated Value

Ungated Lesion

109.36

48.55

0.44

0.44

28.81

3.42

Gated Lesion

111.25

56.06

0.50

0.50

16.35

3.56

Combined-Gated Lesion

110.72

53.17

0.48

0.48

23.21

3.50

Ungated lungs

32.00

3.80

0.12

0.12

Gated lungs

31.26

6.80

0.22

0.22

Combined-Gated lungs

31.60

4.77

0.15

0.15

mean

mean

lesion

lesion /

/SD

mean

lungs

lungs

Normalized to ungated value:

%SD of Mean

Std Dev

% SD

Gated Value

Ungated Lesion

1

1

1

1

1

1

Gated Lesion

1.02

1.15

1.14

1.15

0.57

1.04

Combined-Gated Lesion

1.01

1.10

1.08

1.10

0.81

1.03

1

1

1

1

Gated lungs

0.98

1.79

1.84

1.79

Combined-Gated lungs

0.99

1.26

1.27

1.26

Ungated lungs

237

Table 11-2: Human #4 Gating Improvement Statistics

mean

mean

lesion

lesion /

/SD

mean

lungs

lungs

%SD of Mean

Std Dev

% SD

Gated Value

Ungated Lesion

285.65

101.10

0.35

0.35

84.69

6.01

Gated Lesion

290.88

108.88

0.37

0.37

53.48

6.25

Combined-Gated Lesion

286.54

105.27

0.37

0.36

71.33

6.10

Ungated lungs

47.52

3.37

0.07

0.07

Gated lungs

46.54

5.44

0.12

0.12

Combined-Gated lungs

46.95

4.02

0.09

0.09

mean

mean

lesion

lesion /

/SD

mean

lungs

lungs

Normalized to ungated value:

%SD of Mean

Std Dev

% SD

Gated Value

Ungated Lesion

1

1

1

1

1

1

Gated Lesion

1.02

1.08

1.06

1.08

0.63

1.04

Combined-Gated Lesion

1.00

1.04

1.04

1.04

0.84

1.02

1

1

1

1

Gated lungs

0.98

1.61

1.65

1.61

Combined-Gated lungs

0.99

1.19

1.21

1.19

Ungated lungs

238

Table 11-3: Human #5 Gating Improvement Statistics

mean

mean

lesion

lesion /

/SD

mean

lungs

lungs

%SD of Mean

Std Dev

% SD

Gated Value

Ungated Lesion

165.09

59.49

0.36

0.33

46.67

3.76

Gated Lesion

181.30

75.76

0.42

0.42

27.25

4.23

Combined-Gated Lesion

177.62

71.00

0.40

0.39

38.28

4.09

Ungated lungs

43.87

3.54

0.08

0.08

Gated lungs

42.87

6.65

0.16

0.16

Combined-Gated lungs

43.39

4.64

0.11

0.11

mean

mean

lesion

lesion /

/SD

mean

lungs

lungs

Normalized to ungated value:

%SD of Mean

Std Dev

% SD

Gated Value

Ungated Lesion

1

1

1

1

1

1

Gated Lesion

1.10

1.27

1.16

1.27

0.58

1.12

Combined-Gated Lesion

1.08

1.19

1.11

1.19

0.82

1.09

1

1

1

1

Gated lungs

0.98

1.88

1.92

1.88

Combined-Gated lungs

0.99

1.31

1.33

1.31

Ungated lungs

239

Table 11-4: Human #9 Gating Improvement Statistics

mean

mean

lesion

lesion /

/SD

mean

lungs

lungs

%SD of Mean

Std Dev

% SD

Gated Value

Ungated Lesion

93.30

42.01

0.45

0.43

23.24

2.78

Gated Lesion

96.89

46.34

0.48

0.48

16.34

2.82

Combined-Gated Lesion

95.52

44.66

0.47

0.46

20.28

2.81

Ungated lungs

33.55

4.02

0.12

0.12

Gated lungs

34.33

5.93

0.17

0.17

Combined-Gated lungs

33.99

4.71

0.14

0.14

mean

mean

lesion

lesion /

/SD

mean

lungs

lungs

Normalized to ungated value:

%SD of Mean

Std Dev

% SD

Gated Value

Ungated Lesion

1

1

1

1

1

1

Gated Lesion

1.04

1.10

1.06

1.10

0.70

1.01

Combined-Gated Lesion

1.02

1.06

1.04

1.06

0.87

1.01

1

1

1

1

Gated lungs

1.02

1.48

1.44

1.48

Combined-Gated lungs

1.01

1.17

1.16

1.17

Ungated lungs

240

Table 11-5: Human #11 Gating Improvement Statistics.

mean

mean

lesion

lesion /

/SD

mean

lungs

lungs

%SD of Mean

Std Dev

% SD

Gated Value

Ungated Lesion

89.16

27.50

0.31

0.30

45.27

2.82

Gated Lesion

93.16

38.94

0.42

0.42

20.56

2.98

Combined-Gated Lesion

91.94

34.65

0.38

0.37

30.69

2.92

Ungated lungs

31.67

1.97

0.06

0.06

Gated lungs

31.28

4.53

0.14

0.14

Combined-Gated lungs

31.46

3.00

0.10

0.10

mean

mean

lesion

lesion /

/SD

mean

lungs

lungs

Normalized to ungated value:

%SD of Mean

Std Dev

% SD

Gated Value

Ungated Lesion

1

1

1

1

1

1

Gated Lesion

1.04

1.42

1.36

1.42

0.45

1.06

Combined-Gated Lesion

1.03

1.26

1.22

1.26

0.68

1.04

1

1

1

1

Gated lungs

0.99

2.30

2.33

2.30

Combined-Gated lungs

0.99

1.52

1.53

1.52

Ungated lungs

241

Dissertation Conclusions This dissertation approaches the issues of noise, blurring, deconvolution, image generation, image analysis, respiratory gating, radiation effects unknown in the clinic, dosimetry methodology, as well as presentation of dosimetry calculations for an anticancer agent. Our methodologies were designed to best address the problems, and effort was put in to developing ideas in a simple and applicable manner.

The field of PET imaging, and analyzing PET data, is dynamic and resides at an intersection of physics, math, biology, medicine, physiology, and computer science. This stems from the breadth of what PET is capable of doing, which is imaging physiologic functions. Because of this a wide understanding is beneficial. When a researcher plans to read a scan, it is beneficial/necessary for them to understand how the scan was made before conclusions can be drawn. Or if a scan is analyzed, it is important to understand how noise and resolution effect the data. The number of steps required in utilizing PET data, offer a potential for pitfalls, as well as many point for potential improvements.

In this dissertation, several projects were presented, all of which were performed as group efforts, with contributions from different areas of expertise and many talented minds. They were undertaken with the aim of establishing and extending confidence in PET measurements and laying groundwork for future improvements of the field.

242

Appendix

12.

Software

Installation Instructions

Welcome to Stretch Mouse 4000 (Program requires licensed IDL or free IDL virtual machine installed on host computer)

To install: -Copy “PROGRAM” folder with all its contents to desired directory -open file “~\program\programFiles\pathfile.txt” enter the file path to find location of program folder (i.e. where the launch icon is)

ex: U:\mvc\PROGRAM -Launch program by clicking “mvh.sav”

Questions or comments can be addressed to programmer:

Adam Kesner [email protected]

243

User Manual

Adam Kesner 1/30/2007

Stretch Mouse 4000

This document is an informal user guide for semi-automated microPET analysis software.

Introduction – Creating this software has been a part of my thesis work at UCLA. This software loads small animal PET scans, and with user some user input, co-registers a mouse atlas to the pet scan. After the coregistration, SUV measurements are calculated. If the scans are dynamic, organ time activity curve will be generated and saved. The program will also output a file, which describes the projected activity distribution for a human receiving the same radioisotope, and this file is readable using OLINDA/EXM dedicated dosimetry software. This document here describes the used interface, while the methods can be found in our JNM publication “Semiautomated Analysis of Small-Animal PET Data”, Journal of Nuclear medicine, July, 2006.

244

This document is only in a first draft phase (2/2007). I have tried to outline the main functions and user processes, though have not fully covered everything. User input or questions are always welcome. [email protected] GUI interface

The main user interface contains 12 graphics windows. Each set displays the a volume, with a coronal, axial, sagitial, and a projection view. The top set is for PET images, the middle set displays the phantom volume, and the lower set has multi functionality.

245

The user can click in these widows to change the displayed coordinates and navigate around the volumes. The user may also set windowing thresholds and color scales by clicking the controls to the right of the volumes. Color scales can be reversed by clicking the color scale representation.

The bottom left of the screen contains an output communications window. Here commands are logged.

The top right of the GUI contains several check boxes to turn on/off Boolean functions.

*Hold projections angle –function freezes the projection views *Hide crosshairs *Automax within 10^3 – if this is checked, when a user clicks on a point in the graphical windows, the program will automatically search in 3

246

dimensions (within 1000 voxels) for the maximal valued voxel, and center the views around this voxel. (A USEFUL FUNCTION) *Scale globally – keeps the color scale scaling to the global volume *Define high-low – the region grow function will have both higher and lower bounds if box is checked. If not checked the region grow function will only limit the lower bounds *Add defined organs –function not fully working *No Reset – information will not be lost if loading a new PET scan to program. *Dose after map – doses will automatically be calculated and displayed if box is checked. Otherwise, user needs to select functions from menu after map is made. *No Auto head def –program will not automatically calculate landmark head points on the scan when the body volume is calculated.

This button loads a new PET file

New PET scans can also be loaded by selecting “Change PET file” in the “File” menu. One can also load RAT pet scans (or multi bed position scans) here. (a negative number in the “how many scans?” window will change the orientation)

247

Software The way the software works is that the user is asked to define as many organs as possible on the PET scan. The user may use threshold and several other methods for doing this. The program uses these defined organs, and matches them with predefined organs in the atlas, and this landmark registration is used for the co-registration process. The more organs the user can define the better.

Organ Definition: The main way the user defines organs is by using a threshold region growing algorithm. To do this, the user must place the crosshairs in the PET window at the seed point in the PET scan. The user can then select threshold and smoothing parameters from the drop lists in the corners on the right. (note-1 make sure the “Method” drop list says “recursion” for region growing definition, note-2 putting the smoothing function on “select” (above “3” chooses NO smoothing)

Once the smoothing and threshold parameters are chosen, and the crosshairs are place in the PET scan, the user may click “Define organ”.

Once a representation of the organ generated, it will be displayed in the lower set of windows, available for navigation and verification from the user. The user may choose to redefine the organ (with a different threshold, smoothing or seed position), save the organ representation to the database. To save the organ to the data base, select the organ you are defining in the organ drop list (in the phantom window)

248

By selecting this organ, the program will know which volume to match the defined volume to. At this point, the user may press the “add” button, to add the match to the data base.

At this point, an “*” will be displayed next to all the organs that have define matches, showing the user which information will be used for creating a transformation matrix. If the user does not like the volume add, the user may select “delete” while the organ is selected in the organ drop list, and this will clear all information of the defined organ from the database.

249

Body Definition: It should noted that defining the body as a volume is an important part of the software process (while it is not necessary). The software uses the body definition to stretch the phantom in the xy directions. Additionally, the body volume is used to define the head landmark points (as long as the “no auto head def” check box is not clicked)

Alternate Methods for Organ Definition: The user may also define organs as spheres, on the PET scan. To do this set the “method” to “single point”

Now the PET scan will be displayed in the lower set of windows. The user may click in the PET scan to decide where to put the sphere. To change the size of the sphere the user may select values in the “smoothing” drop list (in single point definition mode smoothing has no function, and only refers to sphere diameter). (Also note larger spheres take longer to draw, so be patient.) If a sphere is the desired size and in the right location, the user may click “Define Organ”, and then “Add” to add the organ to the database.

Map Creation: Once there are at least one organs mated in the database, the user may create a fused image by clicking “Create Map” button.

250

Now the program will run. The more organs that were defined the longer the processing will take, but the better the map will be. Once the volumes are fused together, the user may navigate through the two volumes in the lower set of windows. Changing the color scales or windows for the PET or Phantom volume will change the display of the fused images in the lower set of windows. The user may look at different organs in the fused image by selecting different organs from the organ drop list. The “default” organ display all of them. If the user looses the fused image in the lower set of windows and wants them back then they can click the “apply map” button.

Data Generation: To generate data, a map must be loaded, thus the images are coregistered. All data generated will be saved to the “Output” folder. The user may generate data by clicking “calculate organ activity”, “extrapolate activity to human”, or “calculate TACS”, all can be found in the functions menu. Alternatively, All of these function will run if the “dose after map” check box (top right of the GUI) is selected. “calculate organ activity” – will calculate suv and total activity information “extrapolate activity to human” – will calculate the activity a human would receive getting the same biodistribution. This function outputs a text file and a file which can directly be read in by OLINDA/EXM software. (NOTE-a dynamic scan must be loaded for this function) “calculate TACS” – will calculate time-activity curves for loaded scans (NOTE-a dynamic scan must be loaded for this function)

Notes About Software Use: If user is working with the full software code, they may choose the “utility button” from the “functions” menu at any time. This will pause the program and allow the programmer to look at any variable. To continue working following this interruption, the programmer should re-compile, and may have to select “-Break freeze lock-“ from the functions menu.

251

13.

Major Pieces of Code

Respiratory Gating Code

;++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ ; NAME: ; RunSeveralListToVol ; ; PURPOSE: ; Makes one or more calls to reconstruct data from list mode to ungated, gated, combined gated vol ; ; AUTHOR: ; Adam Kesner ; ; INSTITUTION: ; UCLA ; email: alkesner@mednet,ucla.edu ; ;---------------------------------------------------------------------pro RunSeveralListToVol ListToVol_German, Ldcm="F:\Scans from germany\RAW\Singles\02032006\B_02032006.L", Afile = Afile,ndcm="F:\Scans from germany\RAW\Default_Norm.n" end

;++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ ; NAME: ; ListToVol_German ; ; PURPOSE: ; In this program I am consolidating all the functions to take listmote data through the ; steps to sinogram and to images, gated, ungated, and combined gated ; ; ; CATEGORY: ; Respiratory gating ; ; CALLING SEQUENCE: ; ListToVol_German, Ndcm=Ndcm,Ldcm=Ldcm, Afile = Afile, anzFile = anzFile,inSheader = inSheader,inLheader = inLheader ; ; INPUTS: ; Ndcm - normalization file ; Ldcm - listMode file ; afile - atenuation file ; anzFile = anzaibelt trace ; insheader - sinogram header file ; inlheader - listmode header file ; ; Best way to run: place listmode file, anzai file, and attenuation file in own folder, then run, subfolders will be created ; Function need ~50 GB free space to generate sinogram and images, will leave only ~5gb at end (can delete more files if choose to) ; MUST RUN SIEMENS RECONSTRUCTION SOFTWARE FIRST!!! command line: @c:\cti\do_recon ; program set up for "german" type scans : sinograms intarr(192,192,175), images: fltarr(128,128,47), output images (80,80,47) ; ; ; KEYWORD PARAMETERS: ; None ; ; OUTPUTS: ; image: many file and folders are created as outputs ; ;Start with : ;C:\BaseFolder\scan.L ;C:\BaseFolder\scan.daf ;C:\BaseFolder\scan.dat ; & pointer to sinogram and listmode header files

252

; ;Call function and end with: ;C:\BaseFolder\scan.L.hdr ;C:\BaseFolder\GateOutput\C_GATE_dir\vol ;C:\BaseFolder\GateOutput\C_GATE_dir\wvol ;C:\BaseFolder\GateOutput\C_GATE_dir\cvol ;C:\BaseFolder\GateOutput\C_GATE_dir\gvol

<>ungated volume (German scans 1,80,80,47 fltarr) <>weight distribution volume <>Combinded-gated volume (German scans 8,80,80,47 fltarr) <>gated volume (German scans 8,80,80,47 fltarr)

;C:\BaseFolder\GateOutput ;C:\BaseFolder\GateOutput\C_GATE_dir\ImageBoot_

<>SubDirectory (for respiratory gating <>10.s BootStrapped images _N (N = number of resimulations) ;C:\BaseFolder\GateOutput\C_GATE_dir <>SubDirectory (for gate combination) ;C:\BaseFolder\GateOutput\C_GATE_dir\info.txt <>Info file displayes score and traces (image\belt) ;C:\BaseFolder\GateOutput\C_GATE_dir\nGateSino.s <>BootStrapped sinograms (N boots * num gates) ;C:\BaseFolder\GateOutput\C_GATE_dir\nGateSino.ssummed.s <>summedSinograms (ungated) ;C:\BaseFolder\GateOutput\C_GATE_dir\nRandomSino.s <>randoms sinograms (n many cycles) ;C:\BaseFolder\GateOutput\C_GATE_dir\nRandomSino.sRsummed.s,<>summed randoms sinograms (n many cycles) ;C:\BaseFolder\GateOutput\C_GATE_dir\sinoBoot_ 10.s <>Bootstrapped Sinograms _N (N = number of sinograms * gates) ;C:\BaseFolder\GateOutput\C_GATE_dir\trace025.txt <>txt file of gates trace at 0.25 msec ;C:\BaseFolder\GateOutput\image.img <>image reconstructed at 0.5 second frames (german 5min = fltarr(600,80,80,47) ;C:\BaseFolder\GateOutput\image.img.hdr <>header information for single image ;C:\BaseFolder\GateOutput\image.img_all.img <>summed image reconstructed with attenuation ;C:\BaseFolder\GateOutput\norm.n <>normalization file ;C:\BaseFolder\GateOutput\norm.n.hdr <>normalization file header ;C:\BaseFolder\GateOutput\sino.s.hdr <>single sinogram header ;C:\BaseFolder\GateOutput\sino.s.hdr_All.s.hdr <>singel summed sinogram header ;C:\BaseFolder\GateOutput\sino.s_All.s <>Single sinogram ;C:\BaseFolder\GateOutput\sino.slog.txt <>sinogram info ;C:\BaseFolder\GateOutput\tempImageFile.i <>sample single image file (not deleted automatically ;C:\BaseFolder\GateOutput\tempImageFile.i.hdr <>temp image header ;C:\BaseFolder\GateOutput\tempSinoFile.s <>temporary sinogram file ;C:\BaseFolder\GateOutput\tempSinoFile.s.hdr <>temporary sinogram file header ; ; AUTHOR: ; Adam Kesner 4/2007 ; ; INSTITUTION: ; UCLA ; email: [email protected] ; ; MODIFICATION HISTORY: ; ;------------------------------------------------------------------------------pro ListToVol_German, Ndcm=Ndcm,Ldcm=Ldcm, Afile = Afile, anzFile = anzFile,inSheader = inSheader,inLheader = inLheader

;INITIALIZE starttime = gettime() if n_elements(ldcm) eq 0 then begin print, "error, must pass a listmode file to this function!!!!!!!" endif list_mode_file = Ldcm case n_elements(ndcm) of 0:Norm_file = "G:\Scans from germany\RAW\Default_Norm.n" else: Norm_file = ndcm endcase indir = file_dirname(list_mode_file) outdir = indir + "\GateOutput\" FILE_MKDIR, outdir ;LOAD ANZAI TRACE - check in same directory afileSearch = file_search(indir,"*.daf") if afileSearch ne "" and n_elements(anzFile) eq 0 then anzFile = afileSearch(0) case n_elements(anzFile) of 0:anzaifile_file = "G:\Scans from germany\RAW\Singles\20072006\D_20072006.daf" else: anzaifile_file = anzFile endcase ;LOAD Attenuation - TRACE - check in same directory atfileSearch = file_search(indir,"*.dat") if atfileSearch ne "" and n_elements(adcm) eq 0 then adcm = atfileSearch(0) case n_elements(adcm) of 0:attFile = "G:\Scans from germany\RAW\Singles\20072006\D_20072006.daf" else: attFile = adcm endcase Adcm = attFile Ndcm = Norm_file

253

Ldcm =list_mode_file outpath = outdir timeWindow = 500 ;length of time for each frame from the list mode [ms], #sinograms = totaltime/timewindow ;----------------------------------;AUTO FILE NAMES outAdcm = outpath + "Att.a" outNdcm = outpath + "norm.n" outLdcm = outpath + "list.lm"

;location of output .a file ;location of output .a file ;location of output .a file

outputImageFile = outpath+"image.img" inlist = outLdcm sinoFile = outPath + 'sino.s' if if if if

n_elements(insheader) n_elements(insheader) n_elements(inlheader) n_elements(inlheader)

lt ge lt ge

1 1 1 1

;inpust listmode file ;location of the output sinogram file then then then then

inSinoHeaderFile inSinoHeaderFile inListHeaderFile inListHeaderFile

='U:\FolderForRecon\sino.s.hdr' =insheader = "U:\FolderForRecon\list.lm.hdr" =inlheader

outSinoHeader = outpath+"sino.s.hdr" corrHlistmodeFileName = outLdcm+'.hdr' corrHsinoPointer = sinoFile tempImageFile = outpath + "tempImageFile.i" tempSinoFile = outpath + "tempSinoFile.s" ;---------------------------------------------

;PROGRAMS TO RUN THROUGH FOR GATING ;(1) ; *readlmex ;LTVreadlmex_German, ;LTVreadlmex_German, ;LTVreadlmex_German, ;outAdcm = Adcm ;outNdcm = Ndcm ;outLdcm = Ldcm

DICOM->LISTMODE Adcm,outAdcm Ndcm,outNdcm Ldcm,outLdcm

NOT USING IN THIS VERSION

;correct norm header file_copy,/OVERWRITE , ndcm,outNdcm file_copy,/OVERWRITE , ndcm+".hdr",outNdcm+".hdr ;point norm file to new normfile LTVcorrectNheader_German, outNdcm+".hdr", outNdcm ;CREATE SINGROM HEADER ;LTVcreateSheader_German,inSinoHeaderFile,outSinoHeader+'Temp' file_copy,/OVERWRITE , inSinoHeaderFile,outSinoHeader file_copy,/OVERWRITE , inSinoHeaderFile,outSinoHeader+"_All.s.hdr" ;file_delete,outSinoHeader+'Temp' ;ADD IN LATER ;LTVcorrectSheader_German, corrHlistmodeFileName, outSinoHeader+'Temp', corrHsinoPointer ;copy lheader to correct location file_copy,/OVERWRITE , inListHeaderFile,Ldcm+".hdr"

;(2) ;Sortlms LISTMODE->SINOGRAM 192,192,175 write_counter = 0 outLdcm=list_mode_file LTVsortlmS_german, outLdcm, sinofile, timeWindow = timewindow, write_counter = write_counter;, oneframe = "true" ;(3)

RECONSTRICT SINOGRAMS ---> IMAGES

;make temp files FILE_COPY,/overwrite, outSinoHeader, tempSinoFile + ".hdr" openw,8,tempImageFile close,8 openw,8,tempSinoFile close,8

254

fileinfo = FILE_INFO(sinoFile) numberToRecon = fileinfo.size /

12902400.

LTVrecMany_german, sinoFile, outputImageFile, numberToRecon, tempSinoFile, tempImageFile, outNdcm, outAdcm LTVrecMany_german, sinoFile+"_All.s", outputImageFile+"_all.img", 1, tempSinoFile, tempImageFile, outNdcm, outAdcm, boolnorm = "true" FILE_COPY,/overwrite, tempImageFile + ".hdr", outputImageFile + ".hdr" ;FILE_DELETE, tempImageFile ;FILE_DELETE, tempSinoFile ;FILE_DELETE, tempSinoFile + ".hdr" ;FILE_DELETE, tempImageFile + ".hdr"

;---------------------------------------------------------------;---------------------------------------------------------------;---------------------------------------------------------------;-------------------GATE COMBINATION RUITINES-------------------;---------------------------------------------------------------;---------------------------------------------------------------;---------------------------------------------------------------;create combined gates cgate_dir = outdir+"C_GATE_dir\" FILE_MKDIR,cgate_dir infofile = cgate_dir+"info.txt" gateTrace = cgate_dir+"trace025.txt" outsino = cgate_dir+"nGateSino.s" outrandom= cgate_dir+"nRandomSino.s" nboots = 10 outsinoBoot = cgate_dir+"sinoBoot_"+string(nboots)+".s" outputImageBoot = cgate_dir+"ImageBoot_"+string(nboots)+".s" allimageFile = outputImageFile+"_all.img" ;1) make trace FGATE, filename= outputimagefile, anzaifile = anzaifile_file, infoPath = cgate_dir ;FGATE outputs two files: ;2) sort listmode into 8 frame LTVsortlmS_GATE_german, list_mode_file, gateTrace, outsino,outrandom ;pro LTVsortlmS_GATE_german, listmode_file, trace_file,outsino,outrandom ;3) bootstrap gates to more gates makeBootImages_german, sfile=outsino,rfile =outrandom,outsfile=outsinoBoot, matrixSize = [192,192,175,8],bdepth = nboots ;4) reconstruct Booted images LTVrecMany_german, outsinoBoot, outputImageBoot, nboots*8, tempSinoFile, tempImageFile, outNdcm, outAdcm, boolnorm = "true" ;5) make ungated, gated, and combined gated volumes simCombineBootMulti_german, outputImageBoot, cgate_dir,svol = [80,80,47,80], ungatedVol = allimageFile ;simCombineBootMulti_german, name, outpath,svol = svol, ungatedVol = ungatedvol

;NOW WRAP UP endtime = gettime() totaltime = (endtime - starttime)*1.0/3600 print, "COMPLETED ALL WORK #FRAMES:",totaltime ;erase large sinogram files - no room :( sfiles = file_search(outdir,"*.s") for i=0, n_elements(sfiles)-1 do begin info = file_info(sfiles(i)) print, info.size if info.size gt 2000000000. then file_delete,sfiles(i) endfor end

;------------------------------------------------------------------------------; funciton spitz out dicom information from header file ; ; inname: dicom file ; outname: info file to be written ;-----------------------------------------------------------------------------pro LTVcreateSheader_german,inName,outName obj = obj_new('IDLffDICOM')

255

err = obj->read(inName) hdr = obj->GetValue('0029'x,'1010'x,/NO_COPY) openw,11,outName writeu,11,*hdr[0] close,11 end

;--------------------------------------------------------------------------; ; pro LTVcorrectSheader_German ; ; function opens temporary sinogram header and makes changes ; ; listmodeFileName: listmodeFileName ; outname: new file to be written ; sinoPointer: location of corresponding sinogram ; ;---------------------------------------------------------------------------pro LTVcorrectSheader_German, listmodeFileName, outname, sinoPointer line = "" close,1 close,2 openr,1,outname SHname = str_sep(outname,"Temp") openw,2,SHname(0);+".hdr" tagArray tagArray tagArray tagArray tagArray tagArray tagArray tagArray tagArray

= = = = = = = = =

["study date"] [tagarray(*),"study time"] [tagarray(*),"isotope name"] [tagarray(*),"halflife"] [tagarray(*),"branching factor"] [tagarray(*),"radiopharmaceutical"] [tagarray(*),"start horizontal bed position"] [tagarray(*),"start vertical bed position"] [tagarray(*),"image duration"]

;read every line and process while(not eof(1)) do begin readf,1,line nline = line ;change data file pointer to file that was just created if strpos(line,"data file") ge 0 then nline = "!name of data file := "+sinoPointer +" !!! NOTE ONLY MAIN STRINGS COPIED TO THIS HEADER FILE !!!!" ;loop through tags and find new tags if neeeded for i=0,n_elements(tagArray)-1 do if strpos(line,tagArray(i)) ge 0 then nline = ltvgetstring(listModeFileName,tagArray(i)) ;if have new line then replace right side of line if (nline ne line) then begin lvar = str_sep(line,"=") nlvar = str_sep(nline,"=") line = lvar(0)+"="+nlvar(1) if strpos(nline,"timing tags") gt 0 then nline = line endif ;write data to file printf,2,line endwhile close,1 close,2

;DELETE TEMP FILE FILE_DELETE,outname end

256

;don't write second "duration" instance

;--------------------------------------------------------------------------; ; pro LTVcorrectSheader_German ; ; function opens temporary normalization header and makes changes ; ; nFilehdr: noralizationfilehdr ; nfile: new file to be written ; ;--------------------------------------------------------------------------pro LTVcorrectNheader_german, nFilehdr, nfile line = "" close,1 close,2 openr,1,nFilehdr openw,2,nFilehdr+"TEMP" ;read every line and process while(not eof(1)) do begin readf,1,line nline = line ;change data file pointer to file that was just created if strpos(line,"data file") ge 0 then nline = "!name of data file := "+ nFile printf,2,nline endwhile close,1 close,2 FILE_COPY, /OVERWRITE,nFilehdr+"TEMP",nFilehdr FILE_DELETE, nFilehdr+"TEMP" end

;----------------------------------------------------------------------------; pro LTVreadlmex_German sorts DICOM->readable file ;----------------------------------------------------------------------------pro LTVreadlmex_German, inname, outName obj = obj_new('IDLffDICOMEX',inname) hdr = obj->GetValue('0029,1010') if(keyword_set(hdr)) then begin lmlength = obj->GetValueLength('7FE1,1010') openw,lun,outname+'.hdr',/GET_LUN writeu,lun,hdr close,lun free_lun,lun openr,inLun,inname,/GET_LUN fs = fstat(inLun) point_lun,lun,fs.size-lmlength nWordsLeft = lmlength/4L nWords = 1000000L lmstream = lonarr(nWords) openw,lun,outname,/GET_LUN i = 0 while(not eof(inLun)) do begin if(nWords gt nWordsLeft) then begin lmstream = lonarr(nWordsLeft) nWords = nWordsLeft endif readu,inLun,lmstream writeu,lun,lmstream nWordsLeft = nWordsLeft - nWords i = i + 1 print,i,nWords,nWordsLeft wait, .01 endwhile close,lun free_lun,lun endif end

257

;+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ ; NAME: ; LTVsortlmS_German ; ; PURPOSE: ; sorts list more to sinogram file - this version handles 1-many frames ; ; CATEGORY: ; Reconstruction ; ; CALLING SEQUENCE: ; ListToVol_German,inname,outname, timewindow= timeWindow, write_counter = write_counter, oneFrame = oneFramer ; ; ; INPUTS: ; inname: name of listmodefile ; outname: name of output sinogram file ; timewindow: msec for each sinogram ; write_counter: counter of iterations ; oneFrame: set 1 if only want to contract summed image, 0 if want many ; ; AUTHOR: ; Adam Kesner 4/2007 ; ; INSTITUTION: ; UCLA ; email: [email protected] ; ;--------------------------------------------------------------------------------pro LTVsortlmS_German, inname,outname, timewindow= timeWindow, write_counter = write_counter, oneFrame = oneFrame if n_elementS(timewindow) lt 1 then timewindow = 1000 ; miliseconds if n_elements(oneframe) lt 1 then oneframe = "false" name = inname close,1 openr,1,name window,1,xsize=336,ysize=336 write_counter = 0 ;counter how many sinograms were written close,2 openw,2,outname+"log.txt" close,5 openw,5,"U:\Fourier Gating\log.txt" ;erase outfile openw,3,outname close,3 sino = intarr(192,192) sino3d = intarr(192,192,175) sino3dt = intarr(192,192,175) fs = fstat(1) nw = 10000000L stream = lonarr(nw) nread = 0L totPrompts = 0L totRandoms = 0L iLoop = 1L totalTime = 0L while(not eof(1)) do begin if((fs.size - fs.cur_ptr)/4L ge nw) then begin readu,1,stream nread = nread + nw print, 111111111111111111111111111. endif else begin stream = lonarr((fs.size - fs.cur_ptr)/4L) readu,1,stream

258

nread = nread + nw print, 222222222222222222222222222222. endelse idx = where(stream lt 0l,nTags) tags = stream(idx) tagTypes = ishft(ishft(tags,1),-1) idxTime = where(tagTypes gt 0,nTimes) totalTime = totalTime + nTimes gantryTags = ishft(tags,-29) gantryIdx = where((gantryTags and 2L) eq 2L,nGantry) gatingTags =ishft(tags,-28) gatingIdx = where((gatingTags and 6L) eq 6L, nGating) acqIdx = where((gatingTags and 7L) eq 7L, nAcq) print,iLoop,nTimes,nGantry,nGating,nAcq ;-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.;FIX STREAM FOR TIMES ... ; TLA = where(stream lt 0) ;TIME LOCATION ARRAY timearray = ishft(ishft(stream(TLA),1),-1); TA = (timearray*1.0/timewindow) w = where(timearray gt 100000.*timewindow, num)

;chop of numbers gt 10,000 time windows (they are interfeering!)

if num gt 0 then TA(w) = -1 ta = fix(ta,type=3) ;filter out bad data w = where(ta lt 0) if w(0) ne -1 then ta(w) = -1 ta = [0,ta] hist = histogram(ta) wgood = where(hist)+min(ta); ge min([(timewindow*.5),n_elements(ta)*.5] )) + min(ta) ;either theirs more than %75 time (timewindow lt readin) or mare thant %80 of the time while wgood(0) lt 0 do if wgood(0) lt 0 then wgood = wgood(1:*) ;get rid of and negative flagged frames if wgood(0) eq 0 and wgood(1) ne 1 then wgood = wgood(1:*) ;get rid of 0 unless first sino ;dont use if only 1 instance for di=0, n_elements(wgood)-1 do begin w = where(ta eq wgood(di),num) if num lt 3 then wgood(di) = -1 endfor w = where(wgood ne -1) wgood = wgood(w) backupStream = Stream if oneframe eq "true" then begin wta = where(ta gt 0) ta(wta)=0 wgood = 0 endif ;setup counter to stop looping if sinograms appear not to be consequtive badcounter = 0 ;- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - for Isino=min(wgood), max(wgood) do begin stream = backupStream wt = where(TA eq iSino) if wt(0) eq -1 then badcounter

= badcounter +1

if badcounter gt 3 and min(wgood) gt 50 then begin isino = max(wgood)+1 ;break loop - sinograms not consequitive - program will not handle point_lun,1,9999999999999999. ;endsinorecons endif if wt(0) ne -1 then begin myev=stream(TLA[min(wt)]:TLA(max(wt)-1)) stream = myEv idx = where(stream ge 0l, nEv) if idx(0) gt 0 then begin events = stream(idx) eventType = ishft(events,-30) promptIdx = where(eventType eq 1,nPrompts) randomIdx = where(eventType eq 0,nRandoms)

259

totPrompts = totPrompts + nPrompts totRandoms = totRandoms + nRandoms if nPrompts gt 0 then prompts = ishft(ishft(events(promptIdx),4),-4); mod (336l*336l) if nrandoms gt 0 then randoms = ishft(ishft(events(randomIdx),4),-4); mod (336l*336l) ;found one in 835021 is an error out of range, quick fix: wOK = where(prompts lt n_elements(sino3d)-1,nprompts) if wok(0) ne -1 then prompts = prompts(wok) wOK = where(randoms lt n_elements(sino3d)-1,nrandoms) if wok(0) ne -1 then randoms = randoms(wok) for i = 0L,nPrompts-1 do sino3d[prompts[i]] = sino3d[prompts[i]] + 1 for i = 0L,nRandoms-1 do sino3d[randoms[i]] = sino3d[randoms[i]] - 1 print,"Iloop =",iLoop,"

sinoGram =",isino, " ",nprompts;nEv,nTags,totPrompts,totRandoms, totalTime

wait, .001 tvscl, total(sino3d,3) endif endif if isino ne max(wgood) or n_elements(where(ta eq isino)) ge timewindow and oneFrame ne "true" then begin ;write sinogram to file and start new one for every isino, except for last close,3 openw,3,outname, /append writeu,3,sino3d close,3 sino3dt = sino3dt + sino3d sino3d(*)=0 write_counter = write_counter+1 printf,2,"Iloop =",iLoop," sinoGram =",isino, " printf,5,"Iloop =",iLoop,"

",nprompts,totalTime,n_elements(where(ta eq isino)) sinoGram =",isino, " ",nprompts,totalTime,n_elements(where(ta eq isino))

endif endfor stream = backupstream fs = fstat(1) iLoop = iLoop + 1L endwhile print, isino ;write final 3dsino if not many sinograms have been written if write_counter lt 10 then begin close,3 openw,3,outname, /append writeu,3,sino3d close,3 sino3d(*)=0 write_counter = write_counter+1 printf,2,"FINAL Iloop =",iLoop," sinoGram =",isino, " ",nprompts,totalTime,n_elements(where(ta eq isino)) printf,5,"FINAL Iloop =",iLoop," sinoGram =",isino, " ",nprompts,totalTime,n_elements(where(ta eq isino)) endif ;writetotalsinogram to file if oneframe ne "true" then begin close,1 openw,1,outname+"_All.s" writeu,1,sino3dt close,1 endif close,2 close,5 end

;+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ ; NAME: ; LTVrecMany_German ; ; PURPOSE: ; program takes sinogram file and reconstructs many volumes

260

; ; CATEGORY: ; Reconstruction ; ; CALLING SEQUENCE: ; LTVrecMany_German, FILENAME, outputImageFile, numberToRecon, tempSinoFile, tempImageFile, Nfile, Afile, boolNorm = boolnorm ; ; INPUTS: ; FILENAME: filename of short time sinograms place back to back ; outputImageFile: name of images to be place back to back ; numberToRecon: number of iterations for reconstructions (should be number os sinograms ; tempSinoFile: name/place for temporary sinogram file ; tempImageFile : name/place for temporary image file ; Nfile: normalization file ; Afile: attenuation file ; boolNorm: "true" or "false" to reconstruct with normalization file ; ; AUTHOR: ; Adam Kesner 4/2007 ; ; INSTITUTION: ; UCLA ; email: [email protected] ; ;------------------------------------------------------------------------------pro LTVrecMany_German, FILENAME, outputImageFile, numberToRecon, tempSinoFile, tempImageFile, Nfile, Afile, boolNorm = boolnorm ;sinogram size xsize = 192 ysize = 192 zsize = 175 barr = intarr(xsize,ysize,zsize) s = size(arr) close,4 openw,4,outputImageFile close,4 close,1 openr,1,filename if n_elements(boolnorm) eq 0 then boolnorm = "false" for i=1, numberToRecon do begin ;getvolume and put in singogram file readu,1,barr close,2 openw,2,tempSinoFile writeu,2,barr close,2 if boolnorm ne "true" then a = reconSingleGermanyNoNorm(tempSinoFile, acf = acf,outimage = tempImageFile) if boolnorm eq "true" then a = reconSingleGermanyNorm(tempSinoFile, Nfile, acf = acf,outimage = tempImageFile) ;crop image openw,4,outputImageFile,/append aa = a(24:103,24:103,*)

;size aa

3

80 518400

writeu,4,aa tvscl, congrid(total(aa,3),400,400), i mod 4 close,4 ;print update for qw=0,20 do print, i,i,i,i,i,i,i,i,i,i,i,i,i,i,i,i,i,i,i,i,i,i,i endfor end

261

80

47

4

;++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ ; NAME: ; LTVsortlmS_GATE_german ; ; PURPOSE: ; sorts list more to sinogram file for gated sinograms ; ; CATEGORY: ; Reconstruction ; ; CALLING SEQUENCE: ; LTVsortlmS_GATE_german, listmode_file, trace_file,outsino,outrandom ; ; ; INPUTS: ; ; listmode_file: name of input listmode file ; trace_file: name of trace file (set to correspond to 0.025 sec frames) ; outsino: name of output sinogram file (PROMPTS) ; outrandom: name of output sinogram file (RANDOMS) ; ; AUTHOR: ; Adam Kesner 4/2007 ; ; INSTITUTION: ; UCLA ; email: [email protected] ; ;------------------------------------------------------------------------pro LTVsortlmS_GATE_german, listmode_file, trace_file,outsino,outrandom ;READ IN TRACE ;tracefile shoud be in 1-n form gateArray = [0] a="" close,1 openr,1,trace_file while eof(1) eq 0 do begin readf,1,a gateArray=[gateArray,a*1] endwhile gateArray=gateArray(1:*) close,1 tracer = gateArray windowarray = indgen(n_elements(tracer))*1.0*.025*1000

;SORT LIST MODE DATA ---------------------------------------------------if n_elements(oneframe) lt 1 then oneframe = "false" name = listmode_file close,1 openr,1,name window,1;,xsize=336,ysize=336

262

write_counter = 0 ;counter how many sinograms were written close,2 openw,2,outsino+"log.txt" ;erase outfile openw,3,outsino close,3 ;initialize structures sino = intarr(192,192) sino3d = intarr(192,192,175) sino3dGATE = intarr(192,192,175, (max(gatearray)+1)) sino3dT = intarr(192,192,175) Rsino = intarr(192,192) Rsino3d = intarr(192,192,175) Rsino3dGATE = intarr(192,192,175, (max(gatearray)+1)) Rsino3dT = intarr(192,192,175) fs = fstat(1) nw = 10000000L stream = lonarr(nw) nread = 0L totPrompts = 0L totRandoms = 0L iLoop = 1L totalTime = 0L

;Start processing ------------------------------------------while(not eof(1)) do begin if((fs.size - fs.cur_ptr)/4L ge nw) then begin readu,1,stream backupStream = Stream nread = nread + nw print, 111111111111111111111111111. endif else begin stream = lonarr((fs.size - fs.cur_ptr)/4L) readu,1,stream backupStream = Stream nread = nread + nw print, 222222222222222222222222222222. endelse

idx = where(stream lt 0l,nTags) tags = stream(idx) tagTypes = ishft(ishft(tags,1),-1) idxTime = where(tagTypes gt 0,nTimes) totalTime = totalTime + nTimes gantryTags = ishft(tags,-29) gantryIdx = where((gantryTags and 2L) eq 2L,nGantry) gatingTags =ishft(tags,-28) gatingIdx = where((gatingTags and 6L) eq 6L, nGating) acqIdx = where((gatingTags and 7L) eq 7L, nAcq) print,iLoop,nTimes,nGantry,nGating,nAcq ;-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.;FIX STREAM FOR TIMES ... ; stream = backupStream TLA = where(stream lt 0) ;TIME LOCATION ARRAY timearray = ishft(ishft(stream(TLA),1),-1); TA = timearray TA(*)=-1 ntimearray = nwindowarray ntimearray = ntimearray = ntimearray =

timearray = fix(windowarray/25, type = 15) fix((ntimearray+12) /25., type = 15)*25 ntimearray/25 double(ntimearray)

limit = (2.5*median(ntimearray)) w = where(ntimearray gt limit,num) if num gt 0 then ntimearray(w) = double(-1.0) nta = ta

263

for i=0.0, n_elements(ntimearray)-1 do begin w = where(nwindowarray eq ntimearray(i)) if w(0) gt 0 then nta(i) = gatearray(w(0)) if w(0) le 0 then nta(i) = -1 endfor ta = nta backupStream = Stream if oneframe eq "true" then begin wta = where(ta gt 0) ta(wta)=0 wgood = 0 endif

;- - - - - - - - - - - - - - - - - - - iloopstart = min(gatearray) if iloopstart lt 0 then iloopstart = 0 for Isino=iloopstart, max(gatearray) do begin stream = backupStream for ii =0.0, n_elements(TA)-2 do begin if ta(ii) ne isino then stream(tla(ii):tla(ii+1))=-1 endfor idx = where(stream ge 0l, nEv) if idx(0) ge 0 then begin events = stream(idx) eventType = ishft(events,-30) promptIdx = where(eventType eq 1,nPrompts) randomIdx = where(eventType eq 0,nRandoms) totPrompts = totPrompts + nPrompts totRandoms = totRandoms + nRandoms if nPrompts gt 0 then prompts = ishft(ishft(events(promptIdx),4),-4); mod (336l*336l) if nrandoms gt 0 then randoms = ishft(ishft(events(randomIdx),4),-4); mod (336l*336l) ;found one in 835021 is an error out of range, quick fix: wOK = where(prompts lt n_elements(sino3d)-1,nprompts) prompts = prompts(wok) wOK = where(randoms lt n_elements(sino3d)-1,nrandoms) randoms = randoms(wok) for i = 0L,nPrompts-1 do sino3d[prompts[i]] = sino3d[prompts[i]] + 1 for i = 0L,nRandoms-1 do Rsino3d[randoms[i]] = Rsino3d[randoms[i]] + 1 print,"Iloop =",iLoop," sinoGram =",isino, " ",nprompts;nEv,nTags,totPrompts,totRandoms, totalTime wait, .01 tvscl, sino3d(*,*,155),Isino empty sino3dGate(*,*,*,Isino) = sino3dGate(*,*,*,Isino) + sino3d Rsino3dGate(*,*,*,Isino) = Rsino3dGate(*,*,*,Isino) + Rsino3d sino3d(*)=0 Rsino3d(*)=0 endif endfor stream = backupstream fs = fstat(1) iLoop = iLoop + 1L endwhile close,1 close,2 openw,1,outsino openw,2,outrandom for i=0,n_elements(sino3dGate(0,0,0,*))-1 do begin writeu,1,reform(sino3Dgate(*,*,*,i)) writeu,2,reform(Rsino3Dgate(*,*,*,i)) print, "writing 3d volumes to file, ", i wait, .01 sino3dt = sino3dt + sino3DGATE(*,*,*,i) Rsino3dt = Rsino3dt + Rsino3DGATE(*,*,*,i)

264

endfor close,1 close,2 openw,1,outsino+"summed.s" openw,2,outrandom+"Rsummed.s" writeu,1,sino3dt writeu,2,Rsino3dt close,1 close,2 print, "COMPLETED CREATING German SINOGRAMS" print, "SIZE" help, sino3d close,2 end

;+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ ; NAME: ; makeBootImages_german ; ; PURPOSE: ; makes bootstrapped sinograms from grated sino ; ; CATEGORY: ; Reconstruction ; ; CALLING SEQUENCE: ; makeBootImages_german,sfile=sfile,rfile =rfile,outsfile=outsfile, matrixSize = matrixSize,Bdepth = Bdepth ; ; ; INPUTS: ; ; sfile= input sinogram file (prompts) ; rfile= input sinogram file (randoms) ; outsfile= output sinogram file (# sinograms = number of gates * number of bootstraps) ; matrixSize= size of sinograms (x,y,z,gates) ; Bdepth= number of bootstraps ; ; ; AUTHOR: ; Adam Kesner 4/2007 ; ; INSTITUTION: ; UCLA ; email: [email protected] ; ;-------------------------------------------------------------------------pro makeBootImages_german,sfile=sfile,rfile =rfile,outsfile=outsfile, matrixSize = matrixSize,Bdepth = Bdepth if if if if

n_elements(sfile) lt 1 then stop n_elements(outSfile) lt 1 then stop n_elements(matrixSize) lt 1 then matrixSize = [192,192,175,8] n_elements(Bdepth) lt 1 then Bdepth = 10

xsize= ysize= zsize= depth=

matrixSize(0) matrixSize(1) matrixSize(2) matrixSize(3)

arr = intarr(xsize,ysize,zsize) Rarr= intarr(xsize,ysize,zsize) ;erase outfile close,2 openw,2,outSfile close,2 ;make new bootstrapped frames and write to file loc=[0] s = size(arr) close,1 openr,1,sfile for i=0,Bdepth*depth-1 do begin

265

print, i ii = i mod depth wait, .1 close,1 openr,1,sfile frame = arr point_lun,1,ii*s(5)*s(4) readu,1,frame close,1 close,1 openr,1,rfile rframe = Rarr point_lun,1,ii*s(5)*s(4) readu,1,Rframe close,1 w = where(frame lt 0) if w(0) gt 0 then frame(w)=0 ;~************ Create Strap for the TRUES Sinogram totF = total(frame) loc=[0]

*********~~~~~~~~~

for j=1.,max(frame) do begin w = where(frame eq j) for k=1,j do loc = [loc,w] print, "loading loc arr", j,i endfor loc = loc(1:*) n = long64(n_elements(loc)) nframe = frame nframe(*)=0 for k=LONG64(0.),n-1 do begin rnd = randomu(seed,/DOUBLE) place = rnd * (n-1) ;

place = place mod n if place lt n then p = (loc(place)) if p gt 0 then nframe(p) = nframe(p)+1 if k mod 300000 eq 0 then begin print, "writing",double(k)/n, " wait, .1 empty endif endfor

",i

;~******* Create Strap for the RANDOMS sinogram totF = total(Rframe) loc=[0] for j=1.,max(rframe) do begin w = where(rframe eq j) for k=1,j do loc = [loc,w] print, "loading loc arr", j,i endfor loc = loc(1:*) n = long64(n_elements(loc)) nrframe = rframe nrframe(*)=0 for k=LONG64(0.),n-1 do begin rnd = randomu(seed,/DOUBLE) place = rnd * (n-1) place = place mod n p = (loc(place)) if p gt 0 then nrframe(p) = nrframe(p)+1 if k mod 300000 eq 0 then begin print, "writing",double(k)/n, " wait, .1 empty endif

",i

endfor ;FINISH FRAME

266

*************~~~~~~~~~~~~

;subtract randoms from trues outframe = nframe - nrframe ;get rid of negative numbers w = where(outframe eq 0) if w(0) ne -1 then outframe(w)=0 openw,2,outsfile,/append writeu,2,outframe close,2 print, "FINISHED ROUND", i wait, 1 empty endfor end

;+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ ; NAME: ; simCombineBootMulti_german ; ; PURPOSE: ; Combines gated PET data using boostrapped data sets, multi in title because version computes combination for all gates ; ; CATEGORY: ; Gate Combination ; ; CALLING SEQUENCE: ; simCombineBootMulti_german, name, outpath,svol = svol, ungatedVol = ungatedvol, forcedepth = forcedepth ; ; INPUTS: ; name: vname of image file (images are back to back ; outpath: directory for output ; svol: size of image volume [x,y,z,depth] ; ungatedVol: optional volume to use as the ungated volume (want because normalized) ; forcedepth = optional foce for depth ; ; ; AUTHOR: ; Adam Kesner 4/2007 ; ; INSTITUTION: ; UCLA ; email: [email protected] ; ;--------------------------------------------------------pro simCombineBootMulti_german, name, outpath,svol = svol, ungatedVol = ungatedvol, forcedepth = forcedepth infile = name if n_elements(svol) lt 1 then svol = [80,80,47,80] xsize ysize zsize depth

= = = =

svol(0) svol(1) svol(2) svol(3)

;depth = 24 gates = 8 gate = 1 if n_elements(forcedepth) gt 0 then depth = forcedepth arr = fltarr(xsize,ysize,zsize) barr = fltarr(xsize,ysize,zsize) ;load ungated volume

267

if n_elements(ungatedVol) lt 1 then begin close,1 openr,1,infile for i=0, depth-1 do begin readu,1,barr arr = arr + barr endfor close,1 endif else begin close,1 openr,1,ungatedvol readu,1,arr close,1 endelse ungatedvol = arr hitarr = arr slice =54; 30;64 ;hitarr(*,60:*,*) = -1 ;hitarr(*)=-1 ;hitarr(*,slice,*) = arr(*,slice,*) s = size(arr) w0 = where(hitarr eq 0) if n_elements(w0) gt 1 then hitarr(w0) = -1 flag = "false" counter = 0.0 gvol = fltarr(xsize,ysize,zsize,8) vol = fltarr(xsize,ysize,zsize) cvol = fltarr(xsize,ysize,zsize,8) ccvol = fltarr(xsize,ysize,zsize) wvol = fltarr(xsize,ysize,zsize) nvol = fltarr(xsize,ysize,zsize) starttime = gettime() while flag eq "false" do begin w = where(hitarr eq max(hitarr)) w=w(0) if hitarr(w) eq -1 then flag = "true"; stop in this case hitarr(w) = -1 counter = counter + 1 line = gettimearray(w,s,infile,depth) fline = fft(line) afline = abs(fline) afline(0)=0 true_frequencies =depth/gates *[1,2,3,4,5,6,7] ;----------------------------------ENGINE-------------------total_true = (total(afline(true_frequencies)));*1.0/ntrue total_un_true = (total(afline)-total_true );*1.0/nuntrue weight = total_true/(total_un_true+total_true) gatedline = fltarr(8) for di = 0, n_elements(line)-1 do gatedline(di mod gates) =gatedline(di mod gates) + line(di)*1.0/gates gvARR = gatedline;(line(0:7)+line(8:15)+line(16:23)) / 3 ; mean(line([gate,gate+gates,gate+2*gates])) ;gated value ugv = ungatedvol(w)/gates;mean(line) ;ungated value ugvarr = replicate(ugv,n_elements(gvARR)) ;****** ugvarr(*)= mean(gvarr) ;if don't have this then two lines have different DC and causes trouble = could think about more! cv = (weight*gvARR) + ((1-weight)*ugvarr) xyz = array_indices(hitarr,w) gvol(xyz(0),xyz(1),xyz(2),*)=gvARR vol(w)=ugv cvol(xyz(0),xyz(1),xyz(2),*)=cv wvol(w)= weight;total_true/(total_un_true+total_true) if counter mod 100 eq 0 then begin psize = 500 tvscl, congrid(reform(gvol(*,slice,*,0)),psize,psize),0 tvscl, congrid(reform(vol(*,slice,*)),psize,psize),1

268

tvscl, congrid(reform(cvol(*,slice,*,0)),psize,psize),2 tvscl, congrid(reform(wvol(*,slice,*)),psize,psize),3 print, counter, "Percent = ", fix(counter/(n_elements(where(hitarr ne -1))+counter)*100) pcnt = (counter/(n_elements(where(hitarr ne -1))+counter)*100) timepassed = gettime()-starttime print, "Timeleft = :", (timepassed/(pcnt/100)-timepassed)/60 wait, .01 endif endwhile endtime = gettime() print, "time passed", abs(starttime-endtime)/60," minutes" if n_elements(w0) gt 1 then hitarr(w0) = arr(w0) ;Write volumes to file: s= strsplit(name,"\",/extrac) preamble = s(n_elements(s)-4) preamble = "_"+preamble+"_" openw,1,outpath +preamble+"vol" writeu,1,vol close,1 openw,1,outpath +preamble+"gvol" writeu,1,gvol close,1 openw,1,outpath +preamble+"cvol" writeu,1,cvol close,1 openw,1,outpath +preamble+"wvol" writeu,1,wvol close,1 end

269

;+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ ; NAME: ; FGATEforClean ; ; PURPOSE: ; creates a respiratory trace from images reconstructed in short time frames ; ; CATEGORY: ; Respiratory gating ; ; CALLING SEQUENCE: ; FGATE, filename= filename, anzaifile = anzaifile, infoPath, ttlvolFile = ttlvolFile ; ; INPUTS: ; filename = name of image file (backToback images) ; anzaifile = name of anzia belt trace ; infopath : directory to place output files ; ttlvolFile: optional summed image file ; ; OUTPUTS ; info.txt: text file with scoring information, and belt trace and image trace ; trace025.txt: text file with gates from belt trace ; ; AUTHOR: ; Adam Kesner 4/2007 ; ; INSTITUTION: ; UCLA ; email: [email protected] ; ;----------------------------------------------------PRO FGATE, filename= filename, anzaifile = anzaifile, infoPath, ttlvolFile = ttlvolFile ;set output filenames infofile = infoPath+"info.txt" gateTrace = infoPath+"trace025.txt" device, decomposed=0 loadct,3 xsize ysize zsize depth

= = = =

80 80 81 1200

zsize = 47 ; for german scan arr = fltarr(xsize,ysize,zsize) s = size(arr) fileinfo = FILE_INFO(filename) possibleDepth = fileinfo.size / (s(4)*s(5)) depth = possibleDepth bb = ReadAnzai(anzaiFile) ;get rid of 0's wherebb = fltarr(n_elements(bb)) for i=0, n_elements(bb)-1 do if total(extrac(bb,i,10)) eq 0 then wherebb(i) = 1 w = where(wherebb eq 0) bb = bb(w) bbb=bb depth = min([depth,n_elements(bb)-1]) b = bytscl(bb(0:depth))*1.0/256. arr = fltarr(xsize,ysize,zsize) barr = fltarr(xsize,ysize,zsize) donearray = bytscl(arr) emptyarr = arr ga = fltarr(8,xsize,ysize,zsize) garr = barr garr(*)=0 bc = 0 close,1 openr,1,filename

270

;read in values if n_elements(ttlvolFile) gt 0 then begin close,5 openr,5,ttlvolFile readu,5,arr close,5 endif else begin for i=0, 200 do begin readu,1,barr arr = arr + barr print, "reading sum arr", i wait,.01 print, "reading first 10 arrays: wait, .01 view = (i mod 3) + 1 tvscl, total(arr,view),i endfor endelse

", i

point_lun,1,0 readu,1,barr for i=0, 46 do tvscl, arr(*,*,i),i s=size(arr) slopeArr = arr backuporiginal = arr slopeArr(*)=0 sarr = smooth(arr,3,/EDGE_TRUNCATE) if 7 gt 6 then begin for i=0, s(1)-1 do begin for j=0, s(2)-1 do begin for k=3, s(3)-3 do begin slopearr(i,j,k) = sarr(i,j,k+1) - sarr(i,j,k-1) endfor endfor endfor endif slopearr = -slopearr arr = slopearr arr= abs(arr) carr=arr backuparrr=arr space = 1 farr = 0. wmaxFarr=0 wtf = 0 ;initialize variables tline = fltarr(depth) tline2 = fltarr(depth) tline3 = fltarr(depth) tline4 = fltarr(depth) ftline = fft(tline) tfline = fft(fltarr(depth)) tfline2 = fft(fltarr(depth)) ntfline =tfline tphase = fft(tfline) final = fltarr(depth,15) final2 = fltarr(depth,15) finalline=fltarr(depth,15) ;frequencies 2-9 sec (0.5 second frames) low = (depth/18) hi = (depth/4) if n_elements(hilo) gt 0 then begin hi = depth/hilo(0) low = depth/hilo(1) endif

271

phaseline = fltarr(depth, 15) phasearr = fltarr(depth,15) phasearr2 = fltarr(depth,15)

wfb=15 bexists = 1 if n_elements(bexists) gt 0 then begin b = b(0:depth-1) fb = fft(b) fb(0)=0 fb(0:low) = 0 fb(hi:*)=0 wfb = where(fb eq max(fb)) wfb = wfb(0) endif ;---------------------------------------*_*_*_*_*_*_*_*_*_*_** ;- 1) find trigger array for all peaks (inspiration and expiration) if 5 gt 4 then begin erase top = 80000. for i=0.,top do begin;,80000. do begin

;

;* w = where(arr eq max(arr)) w = w(0) arr(w)=-1 emptyarr(w) = max(arr)*20 print, i, " ", slopearr(w) lt 0, wtf, wfb;, ptf wait, .01 empty ;* line = gettimearray(w,s,filename,depth) fline = fft(line) fline(0)=0 fline(0:low) = 0 fline(hi:depth-hi-1)=0 fline(depth-1-low:*)=0 nline = fft(fline,-1,/inverse) nline(0:2) =0 nline(depth-3:*)=0 print, i, "

", slopearr(w) lt 0, wtf, wfb, "____ ",atan(fline(wfb),/phase)," "+filename;, ptf

SD0 = stddev(tline4^0.5) SD1= stddev(tline4^0.5+nline^0.5) SD2= stddev(tline4^0.5-nline^0.5) warr = [sd0,sd1,sd2] ws = where(warr eq max(warr)) case ws(0) of 0: l=5 1: tline4= tline4+nline 2: tline4= tline4-nline endcase SD0 = stddev(tline3) SD1= stddev(tline3+nline) SD2= stddev(tline3-nline) warr = [sd0,sd1,sd2] ws = where(warr eq max(warr)) case ws(0) of 0: l=5 1: tline3= tline3+nline 2: tline3= tline3-nline endcase case ws(0) of 0: l=5 1: tline2 = tline2 + line 2: tline2 = tline2 - line endcase

oldshow = "true" if oldshow eq "true" then begin

272

; show * if i mod 50 eq 0 then begin !p.multi=[0,1,2] plot, tline3(10:*),title = "TIME ACTIVITY OF HISTOGRAM xtitle = charsize xrange = oplot, b(13:*)*(max(tline3(10:*))-mean(tline3(10:*))),

: N = "+STRCOMPRESS(String(fix(i))), "TIME", ytitle ="RELATIVE AMPLITUDE" , = 2 ,xcharsize = 1,ycharsize = 1 , [50,250] color = 200

plot, tline2(10:*),title = "TIME ACTIVITY OF HISTOGRAM (LOG): N = "+STRCOMPRESS(String(fix(i))), xtitle = "TIME", ytitle ="RELATIVE AMPLITUDE" , charsize = 2 ,xcharsize = 1,ycharsize = 1 , xrange = [50,250] oplot, b(10:*)*(max(tline4(10:*))-mean(tline4(10:*))), color = 200 if i mod 500 eq 0 and i ne 0

then begin

;stats! ; DONT NEED SCTION ANY MORE BECAUSE ADDID SECTION ABOVE corr = fltarr(21) for dis=8,12 do corr(dis) = correlate(returnSlope(tline3(10:*)), returnSlope(extrac(b,dis,590))) plot, corr dis = where(corr eq max(corr)) dis = dis(0); ;try #4 st = size(tline3) sb = size(bbb) sbtdiff = abs(sb(1)-st(1)) sbb=size(bbb) fitarray = fltarr(sbtdiff+24) for fiti = -10,sbtdiff+10 do fitarray(fiti+10) =calcStatsSq(tline3, extrac(bbb,fiti,st(1)), fwindow = [low,hi]) fitarray = abs(fitarray) fitarray = fitarray(0:sbtdiff+20) wf = where(fitarray eq max(fitarray)) outt = real_part(tline3(0:*)) outb = extrac(bbb,wf(0)-10,n_elements(outt)) details= abs(calcStatsSq(outt,outb, fwindow = [low,hi])) offset = wf(0)-10 b = outb if details(0) lt 0.4 then begin ;try #3 st = size(tline3) sb = size(bbb) sbtdiff = abs(sb(1)-st(1))+10 sbb=size(bbb) sbtdiff = 600 fitarray = fltarr(2*sbtdiff+3); for fiti = -sbtdiff,sbtdiff-1 do fitarray(fiti+sbtdiff) =calcStatsSq(tline3, extrac(bbb,fiti,st(1)), fwindow = [low,hi]) fitarray = abs(fitarray) fitarray = fitarray(0:2*sbtdiff) wf = where(fitarray eq max(fitarray)) outt = real_part(tline3(0:*)) outb = extrac(bbb,wf(0)-sbtdiff,n_elements(outt)) details= abs(calcStatsSq(outt,outb, fwindow = [low,hi])) offset = wf(0)-sbtdiff b = outb endif ; if ending here and writing info to file if n_elements(infoPath) gt 0 then begin close,2 openw,2,infofile printf,2,filename printf,2,details printf,2,"number of elements"+string( n_elements(outt))+", offset: "+string(offset) printf,2,"*******" printf,2,"tline3 , b" for di=0,n_elements(outt)-1 do printf,2,string(outt(di))+","+string(outb(di)) close,2

273

binTrace = TraceToGate_ExtendedTime(outt,timeperBin=0.5, outtime = 0.025) close,2 openw,2,gateTrace for di =double(0.0) , n_elements(binTrace)-1 do printf,2,bintrace(di) close,2 oldi = i i = top+1 endif

fb = abs(fft(extrac(b,dis,590))) fb(0)=0 fb(590/2:*)=0 wfb = where(fb eq max(fb)) pfb = 590./wfb plot, tline3(10:*),title = "TIME ACTIVITY OF HISTOGRAM : N = "+STRCOMPRESS(String(fix(i))), xtitle = "TIME", ytitle ="RELATIVE AMPLITUDE" , charsize = 2 ,xcharsize = 1,ycharsize = 1 , xrange = [50,450] oplot, b(dis:*)*(max(tline3(10:*))-mean(tline3(10:*))), color = 200 endif tvscl, total(emptyarr+carr,1),0 tvscl, total(emptyarr+carr,2),1 tvscl, total(emptyarr+carr,3),2 endif endif endfor endif end

;++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ ; NAME: ; getTimeArray ; ; PURPOSE: ; Function returns array with values corresponding to single voxel over time ; taken from back to back images. Function used for volume which are very lage ; ; CATEGORY: ; Respiratory gating ; ; CALLING SEQUENCE: ; getTimeArray, loc, sizeframe, filename, depth ; ; INPUTS: ; loc: 1D location of voxel of interest in array ; sizeframe = fltarr(4) corresping to size of volume x,y,z = s(1,2,3) ; ; RETURNS: ; 1D array of corresponding values ; ; AUTHOR: ; Adam Kesner 4/2007 ; ; INSTITUTION: ; UCLA ; email: [email protected] ; ;--------------------------------------------------------------function getTimeArray, loc, sizeframe, filename, depth close,1 openr,1,filename sizeframe = double(sizeframe) array = fltarr(depth) a = 1.0 ;init float if sizeframe(4) eq 2 then a = fix(a) arr = fltarr(sizeframe(1),sizeframe(2),sizeframe(3)) arr(loc) = 100 for i=0.0, depth-1 do begin

274

w = (loc+ i* sizeframe(5))*sizeframe(4)*1.0; * sizeframe(4));elements*bytes/element ;* sizeframe(5)*sizeframe(4) point_lun,1,w readu,1,a array(i) = a endfor close,1 return,array end

;------------------------------------------------------------------------; Function reads through anai belt trace file and returns data array, binned in 0.5 second frames ;------------------------------------------------------------------------function readanzai, inname,limit = limit if n_elements(limit) lt 1 then limit = 10.^9 timeTaken = 0.025 timeReturn = 0.5 str = "" close,9 openr,9,inname flag =0 dataArray = [0.] while eof(9) eq 0 do begin readf,9,str s = STRSPLIT(str,",",/extrac) if n_elements(s) eq 10 and flag eq 1 then dataArray = [dataArray,s(1)] if n_elements(s) eq 10 and flag eq 0 then flag = 1 endwhile close,9 if n_elements(dataarray) eq 1 then stop; return, -1 dataArray = dataArray(1:*) dataArray = congrid(dataArray,n_elements(dataArray)*timeTaken/timeReturn) if n_elements(dataarray) gt limit then data = data(0:limit) return,dataArray end

;---------------------------------------------------------------------; function calculates corralative statistics for two traces, t,b ; fwindow is for optional frequency filtering of t (belt derived trace) ;---------------------------------------------------------------------function calcStatsSq, t,b, Fwindow = Fwindow flow = fwindow(0) Fhi = fwindow(1) n = n_elements(t) low = -1 high = 9999999. i=0 while (t(i) eq 0 and i lt n) do i=i+1 low = i i=n-1 while (t(i) eq 0 and i lt n) do i=i-1 high = i t = t(low:high) b = b(low:high) n = n_elements(t)

275

fb = fft(b) fb(0:flow)= 0 fb(fhi:n-fhi-1)=0 fb(n-1-flow)=0 b2 = fft(fb,-1,/inverse)

st = returnslope(t) sb = returnslope(b) sb2 = returnslope(b2) ntb = total(sb*st)/total(abs(sb*st)) ntb2 = total(sb2*st)/total(abs(sb2*st)) cor = correlate(sb,st) return, [ntb,ntb2,cor]

;frequency filtered

end

;++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ ; NAME: ; TraceToGate_ExtendedTime ; ; PURPOSE: ; Function takes a respiratory trace, defines respiratory cycles by its peaks, seperates number of bins, ; and returns time-gate 1D array ; ; CATEGORY: ; Respiratory gating ; ; CALLING SEQUENCE: ; TraceToGate_ExtendedTime, trace,timeperBin=timeperBin,outtime=outtime ; INPUTS: ; trace: respiratory trace ; timeperBin: timep per bin of respiratroy trace ; outtime: time per bin of output ; ; RETURNS: ; 1D array of corresponding values ; ; AUTHOR: ; Adam Kesner 4/2007 ; ; INSTITUTION: ; UCLA ; email: [email protected] ; ;--------------------------------------------------------------function TraceToGate_ExtendedTime, trace,timeperBin=timeperBin,outtime=outtime if n_elements(timeperBin)lt 1 then timeperBin = .025 ;sec s = size(trace) n = n_elements(trace) peakDef = 1;sec max +-sec nbins=8 if n_elements(outtime) lt 1 then outtime = 0.025 ;seconds

; MAP TERRRAIN amp = trace minArray = fltarr(n) dis = peakdef/timeperBin for i=0.,s(1)-1 do if amp(i) eq min(EXTRAC(amp,i-dis,2*dis)) then minArray(i) = 1 ;for i=50.,s(1)-50 do if amp(i) eq min(EXTRAC(amp,i-dis,i+dis)) then minArray(i) = 1 m = minarray nminarray= congrid(minarray,timeperBin*1.0/outtime*n_elements(minarray),/interp) ;get rid of doubles w = where(nminarray,num) wcheck = abs([w,0]-[0,w]) wbad = where(wcheck eq 1) if wbad(0) ne -1 then nminarray(w(wbad))=0

276

by the

;---------------- MAKE TIME FILE ----------------------ns = size(nminarray) percentarray = fltarr(ns(1)) percentarray(*)=-1 w = where(nminarray,num) nminarray(w)=1.0 for i=w(0),w(num-1) do begin distlow=-1 disthi=-1 percent=-1 ;hi place = i while nminarray(place) ne 1 do place = place + 1 disthi = abs(place - i) ;low place = i while nminarray(place) ne 1 do place = place - 1 distlow = abs(place - i) if (distlow eq 0 and disthi eq 0) then distlow = 0 percent = distlow*1.0 / (distlow+disthi) percentarray(i) = percent endfor percentarray =percentarray(0:w(num-1)) percentarray(0:w(0))=-1 binarray = fix(percentarray*nbins) return, binarray end

;+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ ; NAME: ; getFreqPowerSpace ; ; PURPOSE: ; Function takes respiratory trace and returns a short term fourier transform spectrogram ; ; CATEGORY: ; Respiratory gating ; ; CALLING SEQUENCE: ; getFreqPowerSpace, line ; ; INPUTS: ; line: respiratory trace ; ; RETURNS: ; 2D spectrogram ; ; AUTHOR: ; Adam Kesner 4/2007 ; ; INSTITUTION: ; UCLA ; email: [email protected] ; ;--------------------------------------------------------------function getFreqPowerSpace, line n = n_elements(line) pic = fltarr(n-120,120) for i=0, n-121 do begin nline = extrac(line,i,120) fn = fft(nline) afn = abs(fn) afn(0)=0 pic(i,*) = afn endfor

277

return,pic end

;+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ ; NAME: ; reconSingleGermanyNoNorm ; ; PURPOSE: ; program calls Siemans reconstruction software with correct input files set for germany scans ; (with no normalization) ; ; CATEGORY: ; image reconstruction ; ; CALLING SEQUENCE: ; reconSingleGermanyNoNorm, sfile, acf = acf,tempimage = tempimage ; ; INPUTS: ; sfile: sinogram file ; acf: attenuation file ; tempimage: temporary image file ; ; ; ; RETURNS: ; reconstructed image ; ; AUTHOR: ; Adam Kesner 4/2007 ; ; INSTITUTION: ; UCLA ; email: [email protected] ; ;-------------------------------------------------------------function reconSingleGermanyNoNorm, sfile, acf = acf,tempimage = tempimage imgSize = 128 zoom = 1.0 fovtrim = 1.0 scanfile = sfile;"F:\Scans from germany\working folder\sinoShort.s" tempImagefile = "F:\Scans from germany\working folder\test.dat" if n_elements(tempimage) gt 0 then tempImagefile= tempimage nFile = nfile; "F:\Scans from germany\working folder\Default_Norm.n" Afile = "" scan = [scanFile+".hdr",scanFile] image =[tempImagefile+".hdr",tempImagefile] ;norm = [nFile + ".hdr",nFile] acf = [Afile+".hdr",Afile] ;!!! attenuation taken out ; RECON OPTIONS: FBP, DIFT, OSEM2D, OSEM3D status = syngo_generic_recon_3d( scan image ,$ ZOOM = zoom ,$ ; NORMFILE = norm ,$ NBINS_REBINNED=imgSize NANGLE_REBINNED=168 ,$ /OSEM3D ,$ ; /FBP ,$ ITERATIONS = 2, $ SUBSETS = 8, $ X_OFFSET=0 ,$ Y_OFFSET=0 ,$ TRIM=fovtrim, $ MODELNO = 1024, $ ; ACFFILE=acf, $ ; /SCATTER,$

; ;

,$

,$

XY_FILTER = 'ALLPASS', $ Z_FILTER = 'ALLPASS' ) XY_FILTER='ALLPASS',Z_FILTER='ALLPASS' XY_FILTER='ALLPASS', $

print,'Zoom after: ',zoom openr,lun,image[1],/GET_LUN vol = fltarr(imgSize,imgSize,47) readu,lun,vol close,lun

278

!!!

)

tvscl, total(vol,3) for i=0,46 do tvscl, vol(*,*,i),i free_lun,lun return, vol end

;+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ ; NAME: ; reconSingleGermanyNorm ; ; PURPOSE: ; program calls Siemans reconstruction software with correct input files set for germany scans ; (with normalization) ; ; CATEGORY: ; image reconstruction ; ; CALLING SEQUENCE: ; reconSingleGermanyNoNorm, sfile, acf = acf,tempimage = tempimage ; ; INPUTS: ; sfile: sinogram file ; acf: attenuation file ; tempimage: temporary image file ; ; ; ; RETURNS: ; reconstructed image ; ; AUTHOR: ; Adam Kesner 4/2007 ; ; INSTITUTION: ; UCLA ; email: [email protected] ; ;-----------------------------------------------------------------function reconSingleGermanyNorm, sfile, nfile, acf = acf, outimage = outimage imgSize = 128 zoom = 1.0 fovtrim = 1.0 scanfile = sfile;"F:\Scans from germany\working folder\sinoShort.s" tempImagefile = "F:\Scans from germany\working folder\test.dat" if n_elements(outimage) gt 0 then tempImagefile = outimage nFile = nfile; "F:\Scans from germany\working folder\Default_Norm.n" Afile = "" scan = [scanFile+".hdr",scanFile] image =[tempImagefile+".hdr",tempImagefile] norm = [nFile + ".hdr",nFile] acf = [Afile+".hdr",Afile] ;!!! attenuation taken out

!!!

; RECON OPTIONS: FBP, DIFT, OSEM2D, OSEM3D

; ;

; ;

status = syngo_generic_recon_3d( scan image ,$ ZOOM = zoom ,$ NORMFILE = norm ,$ NBINS_REBINNED=imgSize NANGLE_REBINNED=168 /OSEM3D ,$ ; /FBP ,$ ITERATIONS = 2, $ SUBSETS = 8, $ X_OFFSET=0 ,$ Y_OFFSET=0 ,$ TRIM=fovtrim, $ MODELNO = 1024, $ ACFFILE=acf, $ /SCATTER,$

,$

,$ ,$

XY_FILTER = 'ALLPASS', $ Z_FILTER = 'ALLPASS' ) XY_FILTER='ALLPASS',Z_FILTER='ALLPASS' XY_FILTER='ALLPASS', $

print,'Zoom after: ',zoom openr,lun,image[1],/GET_LUN vol = fltarr(imgSize,imgSize,47)

279

)

readu,lun,vol close,lun tvscl, total(vol,3) for i=0,46 do tvscl, vol(*,*,i),i free_lun,lun return, vol end

Scan Simulation

;------------------------------------------------------------------------;;; ; simScan INCLUDES Z AXIS BLURRING ; ; This function takes an array farward projects it to sinogram space, blurs it in 2D,adds noise, and backprojects it ; ; call: simScan(emmVol, blurkernel, attvol = attvol,boolnoise=boolnoise) ; emmvol - emmision volume ; blurkernel - 3d blurring kernel (representative of 3d space) ; attvol = include attenuation volume if attenuating (make sure correct units) ; boolnoise - set "true" or "false" if want poissan noise added (true by default) ; ; ; ; Adam Kesner ; 2/2008 ; UCLA ;----------- --------------------------------------------------------------function simScan, emmVol, blurkernel, attvol = attvol,boolnoise=boolnoise common rand, seed

if n_elements(attvol) lt 1 then attenuation = "false" if n_elements(boolnoise) lt 1 then boolnoise = "true" blur = "true" activityCalibration = 1.0; s = size(emmvol) k=blurkernel ;DEFINE THE BLURRING KERNEL ;make 2D backupK = k KNL=K sk = size(knl) knl = reform(knl(sk(1)/2,*,*)) knl = knl*1.0 / total(knl) nsk = size(knl)

;make 2d

;DEFINE THE SINOGRAM PARAMETERS mult = 10. nTheta = 144 ;nTheta is the number of projection views nrho = s(1)*mult drho = s(1)*1.0/nrho newvol = emmvol newvol(*) = 0 sinoarray = fltarr(s(1),144,s(3)) sinogramAttvol = fltarr(s(1),144,s(3)) ;------------------------------;LOOP THROUGH SLICES for j=0, s(3)-1 do begin

LOOP 1 - CREATE SINO------------------------

280

image = reform(emmVol(*,*,j))

;

mumap = reform(dataAtn(i,j,*,*));reform(atten[j,*,*]) ;selects the jth plane of the map that has been created sinogramb = transpose(RADON(image, NRHO=nrho, nTHETA=ntheta, RHO=rho, THETA=theta, drho = drho)) ;atnsino = transpose(RADON(mumap, NRHO=nrho, nTHETA=ntheta, RHO=rho, THETA=theta,drho=drho)) ; attenuation sinogram sinogram = rebin(sinogramb,nrho/mult,ntheta) ; combining the oversampled data if attenuation eq "true" then begin imageAtt = reform(attvol(*,*,j))

;

mumap = reform(dataAtn(i,j,*,*));reform(atten[j,*,*]) ;selects the jth plane of the map that has been created sinogrambAtt = transpose(RADON(imageAtt, NRHO=nrho, nTHETA=ntheta, RHO=rho, THETA=theta, drho = drho)) ;atnsino = transpose(RADON(mumap, NRHO=nrho, nTHETA=ntheta, RHO=rho, THETA=theta,drho=drho)) ; attenuation sinogram sinogrambAtt = rebin(sinogrambAtt,nrho/mult,ntheta) sinogramAtt = sinogrambAtt;congrid(sinogrambAtt,nrho/1.,ntheta) ; tossing the oversampled data sinogramAttvol(*,*,j) = sinogramAtt sinogram = sinogram*exp(-sinogramAtt) ; attenuation endif sinoarray(*,*,j) = sinogram print, j, " wait, .01 endfor

phase 1/3"

;------------------------------;blur in 2d for i=0, 144-1 do begin

;

LOOP 2 - BLUR 2Ds ----------------------------

oslice = reform(sinoarray(*,i,*)) slice = oslice;extrac(oslice,-nsk(1)-1,-nsk(2)-1,s(1)+2*nsk(1)+1,s(3)+2*nsk(2)+1) onslice = convol(slice,knl) nslice = onslice;extrac(onslice,nsk(1)+1,nsk(2)+1,s(1),s(3)) nslice = convol(slice,knl,/edge_wrap) could go that way.....? sinoarray(*,i,*) = nslice;nslice

print, i," wait, .01 endfor

phase 2/3"

;------------------------------- LOOP 3 - ADD NOISE AND RECON rho = rebin(rho,n_elements(rho)/mult) for j=0, s(3)-1 do begin

-----------------------

sinogram = sinoarray(*,*,j)

;------BRING IN DA NOISE-----------------idx = where(sinogram gt 0,npts) noisysino = sinogram if (boolnoise eq "true") then begin for k = 0,npts-1 do begin value = sinogram[idx[k]] newValue = randomu(seed,poisson=value,/double) noisysino[idx[k]] = newvalue endfor;k endif sinogram = noisySino

;correct for att sinogramAtt =reform(sinogramAttvol(*,*,j)) if attenuation eq "true" then sinogram = sinogram*exp(sinogramAtt) ; attenuation ;----------------------------------------------------------------------

281

img = FBPK(transpose(sinogram), RHO, THETA,[3,0.5], outsize= [s(1),s(2)]) ;function FBPK, sinogram, oRHO, oTHETA,params, outSize = outSize;, DX = dx, ROT=ROT,ZOOM=zoom, DEG360=deg360,offsets=offsets ;old call: ;oimg = fltarr(s(1),s(2)) ;fbrecons,sinogram,img,[3,0.5],zoom=1.

;applies Big D's filtered backprojection

newvol(*,*,j) = img ;* total(emmVol(*,*,j))/total(img) print, j," WAIT, .01

phase 3/3"

endfor ;j return, newvol end

;++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ ; NAME: ; FBPK - Filtered back projection Kesner version ; child of FBRECONS code ; ; PURPOSE: ; Performs a filtered backprojection of a sinogram ; ; CALL ; FBPK(sinogram, oRHO, oTHETA,params, outSize = outSize) ; ; SINOGRAM- sinogram to be reconsructed (make sure x axis is angle, yaxis is dx) ; oRHO - rho array (should use data returned from farward radon function) ; oTHETA - oTHETA array (should use data returned from farward radon function) ; params - Recon parameters passed as a 2 element vector [filter,cutoff] ; outSize - size of output image array ; ; ; returns reconstructed image ; ;modified (only ramp filter!!!!) ;Adam Kesner ;2/2008 ;----------------------------------------------------------------function FBPK, sinogram, oRHO, oTHETA,params, outSize = outSize;, DX = dx, ROT=ROT,ZOOM=zoom, DEG360=deg360,offsets=offsets

if (NOT keyword_set(outSize)) then begin ss = size(sinogram) outsize=[ss(1),ss(2)] endif if (NOT keyword_set(ROT)) then rot = 0. if (keyword_set(DEG360)) then ANG = 2*!PI else ANG = !PI if (NOT keyword_set(ZOOM)) then zoom = 1. if (NOT keyword_set(DX)) then dx = 1. if (max(sinogram) LE 0.) then return,-1 Idims = size(image) Sdims = size(sinogram) xdim = Idims(1) numProjs = Sdims(2) numViews = Sdims(1) angoff = rot/180.*!PI binsize = float(xdim)/numProjs*zoom if(keyword_set(offsets)) then offsets = [xdim/2,xdim/2] + [-offsets(0),offsets(1)] / (dx) $ else offsets = [xdim/2,xdim/2] twopower = fix(alog(numProjs) / alog(2)+0.5) nfft = 2^(twopower+2) FilterNum = Params(0) TxFilterF = params(1) i = indgen(fix(nfft/2*TxFilterF)+1) TxFilter = fltarr(1024) TxFilter(i) = (1+cos(!PI*i/max(i)))/2 i = indgen(nfft/2-1)+1 TxFilter(nfft-i) = TxFilter(i) TxFilter[*] = 1. Nyquist = 1./(2.*dx) Ramp = findgen(nfft) Ramp(nfft-i) = ramp(i)

282

Ramp = Ramp/max(Ramp) Ramp = Ramp * Nyquist Projection = fltarr(nfft) ProjIdx = indgen(NumProjs) TxFilter = Ramp*0+1 ;TxFilter[*] = 1. filtSino = SINOGRAM filtSino(*)=0 for angle = 0,Numviews-1 do begin theta = angle * ANG/NumViews - rot * !pi/180. Projection = Projection * 0 Projection(0) = reform(sinogram(angle,*)) Projection = (fft(fft(Projection,1)*Ramp*TxFilter,-1)) ; riemann,Projection(ProjIdx),Image,theta,/backproject,/bilinear,d=binsize,cor=offsets filtSino(angle,*)= Projection(ProjIdx) endfor ;theta = !pi*findgen(NumViews)/NumViews ;rho = (findgen(NumProjs)-NumProjs/2.) ;image = radon(transpose(sinogram),theta=theta,rho=rho,/backproject,/linear) a = RADON(filtSino,/backproject, RHO=orho, THETA=otheta,/linear,nx=outSize(0),ny=outSize(1)) return, a;*total(sinogram)/numviews

;a * ANG/NumViews

END

283

References 1. Sweet WH, Mealey J, Jr., Aronow S, Brownell GL. Localization of focal intracranial lesions by scanning of rays from position-emitting isotopes. Clin Neurosurg. 1959;7:159-199. 2. Aronow S, Brownell GL. An Apparatus for Brain Tumor Localization Using Positron Emitting Radioactive Isotopes. IRE National Convention, 1956. 1956. 3.

Anger H, Rosenthal D. Scintillation camera and positron camera. Medical Radioisotope Scanning. 1959:59-75.

4. Phelps ME, Hoffman EJ, Huang SC, Kuhl DE. ECAT: a new computerized tomographic imaging system for positronemitting radiopharmaceuticals. J Nucl Med. Jun 1978;19(6):635-647. 5. Herbertson RA, Lee ST, Tebbutt N, Scott AM. The expanding role of PET technology in the management of patients with colorectal cancer. Ann Oncol. Nov 2007;18(11):1774-1781. 6. Rajendran JG, Greer BE. Expanding role of positron emission tomography in cancer of the uterine cervix. J Natl Compr Canc Netw. May 2006;4(5):463-469. 7.

Hicks RJ. The role of PET in monitoring therapy. Cancer Imaging. 2005;5(1):51-57.

8. Giannopoulou C. The role of SPET and PET in monitoring tumour response to therapy. Eur J Nucl Med Mol Imaging. Aug 2003;30(8):1173-1200. 9. Hoh CK, Schiepers C, Seltzer MA, et al. PET in oncology: will it replace the other modalities? Semin Nucl Med. Apr 1997;27(2):94-106. 10. 45S.

Slomka PJ. Software approach to merging molecular with anatomic information. J Nucl Med. Jan 2004;45 Suppl 1:36S-

11. Hutton BF, Braun M. Software for image registration: algorithms, accuracy, efficacy. Semin Nucl Med. Jul 2003;33(3):180-192. 12. Lee JS, Park KS, Lee DS, Lee CW, Chung JK, Lee MC. Development and applications of a software for Functional Image Registration (FIRE). Comput Methods Programs Biomed. May 2005;78(2):157-164. 13.

Townsend DW, Carney JP, Yap JT, Hall NC. PET/CT today and tomorrow. J Nucl Med. Jan 2004;45 Suppl 1:4S-14S.

14. Coakley FV, Qayyum A, Kurhanewicz J. Magnetic resonance imaging and spectroscopic imaging of prostate cancer. J Urol. Dec 2003;170(6 Pt 2):S69-75; discussion S75-66. 15. Rousset OG, Ma Y, Evans AC. Correction for partial volume effects in PET: principle and validation. J Nucl Med. May 1998;39(5):904-911. 16. D. Louis Collins CJHTMPACE. Automatic 3-D model-based neuroanatomical segmentation. Human Brain Mapping. 1995;3(3):190-208. 17. Shepard D. A two-dimensional interpolation function for irregularly-spaced data. Proc 23rd National Conference ACM. 1968:517-524. 18. Kinahan PE, Alessio AM, Segars WP, et al. Quantitative Evaluation of Respiratory Gated Whole-Body PET/CT Imaging Incorporating Respiration Variability. Paper presented at: IEEE Medical Imaging Conference, 2007; Honolulu, HI. 19. Gangloff A, Hsueh WA, Kesner AL, et al. Estimation of paclitaxel biodistribution and uptake in human-derived xenografts in vivo with [18F]-fluoropaclitaxel. (In Press) J Nuc Med. 2005;46:11:. 20. Lacan G, Kesner AL, Gangloff A, et al. Synthesis of 2-[(2-chloro-2-[18F]fluoroethyl)amino]-2H-1,3,2oxazaphosphorinane-2-oxide18F-fluorocyclophosphamide), a potential tracer for breast tumor prognostic imaging with PET. Journal of Labelled Compounds and Radiopharmaceuticals. 2005;48(9):635-643.

284

21. Antonelli PL. “Transformations of quadrilaterals, tensor fields, and morphogenesis,” in Mathematical Essays on Growth and the Emergence of Form: University of Alberta; 1985. 22. Fitzpatrick JM, Leuze MR. A class of one-to-one two-dimensional transformations Vol 39 Academic Press Professional, Inc.; 1987. 23.

Sibson R, Stone G. Computation of thin-plate splines. Vol 12: Society for Industrial and Applied Mathematics; 1991.

24. Bookstein FL. Principal Warps: Thin-Plate Splines and the Decomposition of Deformations. Vol 11: IEEE Computer Society; 1989. 25. Bookstein FL. Landmark methods for forms without landmarks: morphometrics of group differences in outline shape. Med Image Anal. Apr 1997;1(3):225-243. 26. Tai Y-C, Lin KP, Hoh CK, Huang SC, Hoffman EJ. Utilization of 3-D elastic transformation in the registration ofchest Xray CT and whole body PET. IEEE Trans Nucl Sci. 1997;44(4):1606-1612. 27.

Franke R. Scattered Data Interpolation: Tests of Some Method. Mathematics of Computation. 1982;38(157):181-200.

28. Yang C-S, Kao S-P, Lee F-B, Hung P-S. TWELVE DIFFERENT INTERPOLATION METHODS: A CASE STUDY OF SURFER 8.0. XXth ISPRS Congress, 12-23 July, IstanbuI, Turkey. 2004:778 ff. 29. Fang S, Raghavan R, Richtsmeier J. Volume Morphing Methods for Landmark Based 3D Image Deformation. SPIE Int Symp on Medical Imaging, CA, 1996. 1996. 30. 43.

Ruprecht D, Muller H. Image warping with scattered data interpolation. Computer Graphics and Applications, (1995).37-

31.

Cherry, Sorenson, Phelps. Physics in Nuclear Medicine. USA: Elsivar Science; 2003.

32. Zhang J, Hall NC, Kale M, Hurley D, Knopp MV. Quantification of the accuracy and reproducibility of standardized methods for determining tumor metabolic volume and standardized uptake values Paper presented at: SNM, 2006; San Diego. 33. Sigg D, Iagaru A, Levin C, Quon A. Partial volume correction of PET/CT derived SUV values in lung nodules. Paper presented at: SNM, 2006; Sand Diego. 34. 413.

Hutton BF, Osiecki A. Correction of partial volume effects in myocardial SPECT. J Nucl Cardiol. Jul-Aug 1998;5(4):402-

35. Aston JA, Cunningham VJ, Asselin MC, Hammers A, Evans AC, Gunn RN. Positron emission tomography partial volume correction: estimation and algorithms. J Cereb Blood Flow Metab. Aug 2002;22(8):1019-1034. 36. Antich P. Application of expectation maximization algorithms for image resolution improvement in a small animal PET system. IEEE Trans Nucl Sci. 2005;52:684-690. 37. Dewaraja YK, Ljungberg M, Fessler JA. Anatomical information based partial volume compensation for I-131 SPECT imaging in radioimmunotherapy Paper presented at: SNM, 2006; San Diego. 38. Gerth HU, Juergens KU, Dirksen U, Gerss J, Schober O, Franzius C. Significant Benefit of Multimodal Imaging: PET/CT Compared with PET Alone in Staging and Follow-up of Patients with Ewing Tumors. J Nucl Med. Dec 2007;48(12):1932-1939. 39. Teo BK, Seo Y, Bacharach SL, et al. Partial-volume correction in PET: validation of an iterative postreconstruction method with phantom and patient data. J Nucl Med. May 2007;48(5):802-810. 40. Burri RJ, Rangaswamy B, Kostakoglu L, et al. Correlation of Positron Emission Tomography Standard Uptake Value and Pathologic Specimen Size in Cancer of the Head and Neck. Int J Radiat Oncol Biol Phys. Feb 5 2008. 41. Yamamoto Y, Nishiyama Y, Kimura N, et al. Comparison of (18)F-FLT PET and (18)F-FDG PET for preoperative staging in non-small cell lung cancer. Eur J Nucl Med Mol Imaging. Feb 2008;35(2):236-245. 42. Melek H, Gunluoglu MZ, Demir A, Akin H, Olcmen A, Dincer SI. Role of positron emission tomography in mediastinal lymphatic staging of non-small cell lung cancer. Eur J Cardiothorac Surg. Feb 2008;33(2):294-299. 43. Kim SY, Roh JL, Kim JS, et al. Utility of FDG PET in patients with squamous cell carcinomas of the oral cavity. Eur J Surg Oncol. Feb 2008;34(2):208-215.

285

44. Anderson C, Koshy M, Staley C, et al. PET-CT fusion in radiation management of patients with anorectal tumors. Int J Radiat Oncol Biol Phys. Sep 1 2007;69(1):155-162. 45. Dahele MR, Ung YC. 18F-FDG PET in planning radiation treatment of non-small cell lung cancer: where exactly is the tumor? J Nucl Med. Aug 2007;48(8):1402; author reply 1403. 46. Saad A, Smith B, Hamarneh G, Moller T. Simultaneous segmentation, kinetic parameter estimation, and uncertainty visualization of dynamic PET images. Med Image Comput Comput Assist Interv Int Conf Med Image Comput Comput Assist Interv. 2007;10(Pt 2):726-733. 47. Schinagl DA, Vogel WV, Hoffmann AL, van Dalen JA, Oyen WJ, Kaanders JH. Comparison of five segmentation tools for 18F-fluoro-deoxy-glucose-positron emission tomography-based target volume definition in head and neck cancer. Int J Radiat Oncol Biol Phys. Nov 15 2007;69(4):1282-1289. 48. Drever LA, Roa W, McEwan A, Robinson D. Comparison of three image segmentation techniques for target volume delineation in positron emission tomography. J Appl Clin Med Phys. 2007;8(2):93-109. 49. Chen Y, Yan Z, Shi J. Application of lattice boltzmann method to image segmentation. Conf Proc IEEE Eng Med Biol Soc. 2007;1:6561-6564. 50. 5648.

Wieclawek W, Pietka E. Live-Wire-Based 3D Segmentation Method. Conf Proc IEEE Eng Med Biol Soc. 2007;1:5645-

51. Krol A, Unlu MZ, Baum KG, et al. MRI/PET nonrigid breast-image registration using skin fiducial markers. Phys Med. 2006;21 Suppl 1:39-43. 52. Lucignani G. Time-of-flight PET and PET/MRI: recurrent dreams or actual realities? Eur J Nucl Med Mol Imaging. Aug 2006;33(8):969-971. 53.

Gaa J, Rummeny EJ, Seemann MD. Whole-body imaging with PET/MRI. Eur J Med Res. Jun 30 2004;9(6):309-312.

54. Kesner AL, Dahlbom M, Huang SC, et al. Semiautomated analysis of small-animal PET data. J Nucl Med. Jul 2006;47(7):1181-1186. 55. Boellaard R, Krak NC, Hoekstra OS, Lammertsma AA. Effects of noise, image resolution, and ROI definition on the accuracy of standard uptake values: a simulation study. J Nucl Med. Sep 2004;45(9):1519-1527. 56. Westerterp M, Pruim J, Oyen W, et al. Quantification of FDG PET studies using standardised uptake values in multi-centre trials: effects of image reconstruction, resolution and ROI definition parameters. Eur J Nucl Med Mol Imaging. Mar 2007;34(3):392404. 57. Weber WA, Schwaiger M, Avril N. Quantitative assessment of tumor metabolism using FDG-PET imaging. Nucl Med Biol. Oct 2000;27(7):683-687. 58. Press WH, Flannery BP, Teukolsky SA, Vetterling WT. Numerical Recipes in C: The Art of Scientific Computing 2nd ed: Cambridge University Press; 1992. 59. Boucher L, Rodrigue S, Lecomte R, Benard F. Respiratory Gating for 3-Dimensional PET of the Thorax: Feasibility and Initial Results. J Nucl Med. February 1, 2004 2004;45(2):214-219. 60. Nehmeh SA, Erdi YE, Rosenzweig KE, et al. Reduction of Respiratory Motion Artifacts in PET Imaging of Lung Cancer by Respiratory Correlated Dynamic PET: Methodology and Comparison with Respiratory Gated PET. J Nucl Med. October 1, 2003 2003;44(10):1644-1648. 61. Nehmeh SA, Erdi YE, Ling CC, et al. Effect of Respiratory Gating on Quantifying PET Images of Lung Cancer. J Nucl Med. July 1, 2002 2002;43(7):876-881. 62. Pitman AG, Kalff V, Van Every B, Risa B, Barnden LR, Kelly MJ. Effect of Mechanically Simulated Diaphragmatic Respiratory Motion on Myocardial SPECT Processed With and Without Attenuation Correction. J Nucl Med. September 1, 2002 2002;43(9):1259-1267. 63. Erdi YE, Nehmeh SA, Pan T, et al. The CT Motion Quantitation of Lung Lesions and Its Impact on PET-Measured SUVs. J Nucl Med. August 1, 2004 2004;45(8):1287-1292.

286

64. Ruf J, Lopez Hanninen E, Bohmig M, et al. Impact of FDG-PET/MRI image fusion on the detection of pancreatic cancer. Pancreatology. 2006;6(6):512-519. 65. Livieratos L, Rajappan K, Stegger L, Schafers K, Bailey DL, Camici PG. Respiratory gating of cardiac PET data in listmode acquisition. Eur J Nucl Med Mol Imaging. May 2006;33(5):584-588. 66. Livieratos L, Stegger L, Bloomfield PM, Schafers K, Bailey DL, Camici PG. Rigid-body transformation of list-mode projection data for respiratory motion correction in cardiac PET. Phys Med Biol. Jul 21 2005;50(14):3313-3322. 67. Mijailovich SM, Treppo S, Venegas JG. Effects of lung motion and tracer kinetics corrections on PET imaging of pulmonary function. J Appl Physiol. Apr 1997;82(4):1154-1162. 68. Larson SM, Nehmeh SA, Erdi YE, Humm JL. PET/CT in non-small-cell lung cancer: value of respiratory-gated PET. Chang Gung Med J. May 2005;28(5):306-314. 69. Wurm RE, Gum F, Erbel S, et al. Image guided respiratory gated hypofractionated Stereotactic Body Radiation Therapy (H-SBRT) for liver and lung tumors: Initial experience. Acta Oncol. 2006;45(7):881-889. 70. Holloway CL, Robinson D, Murray B, et al. Results of a phase I study to dose escalate using intensity modulated radiotherapy guided by combined PET/CT imaging with induction chemotherapy for patients with non-small cell lung cancer. Radiother Oncol. Dec 2004;73(3):285-287. 71. Feng B, Bruyant PP, Pretorius PH, et al. Estimation of the rigid-body motion from images using a generalized center-ofmass points approach, 2005. 72. Bundschuh RA, Martinez-Moeller A, Essler M, et al. Postacquisition detection of tumor motion in the lung and upper abdomen using list-mode PET data: a feasibility study. J Nucl Med. May 2007;48(5):758-763. 73. Dawood M, Buther F, Lang N, Schober O, Schafers KP. Respiratory gating in positron emission tomography: a quantitative comparison of different gating schemes. Med Phys. Jul 2007;34(7):3067-3076. 74. Detorie N, Dahlbom M. Rigid vs. non-rigid motion correction for respiratory gated PET. J NUCL MED MEETING ABSTRACTS. May 1, 2007 2007;48(MeetingAbstracts_2):410P-. 75. Wolz G, Nömayr A, Hothorn T, et al. Comparison of performance between rigid and non-rigid software registering CT to FDG-PET. International Journal of Computer Assisted Radiology and Surgery. 2007-12-15 2007;2(3):183-190. 76. Reyes M, Malandain G, Koulibaly PM, Gonzalez Ballester MA, Darcourt J. Respiratory motion correction in emission tomography image reconstruction. Med Image Comput Comput Assist Interv Int Conf Med Image Comput Comput Assist Interv. 2005;8(Pt 2):369-376. 77. Klein GJ, Reutter BW, Ho MH, Reed JH, Huesman RH. Real-time system for respiratory-cardiac gating in positron tomography. IEEE Transactions on Nuclear Science. Aug 1998 1998;45(4):2139-2143. 78. Dawood M, Lang N, Jiang X, Schafers KP. Lung motion correction on respiratory gated 3-D PET/CT images. IEEE Trans Med Imaging. Apr 2006;25(4):476-485. 79. Thorndyke B, Schreibmann E, Koong A, Xing L. Reducing respiratory motion artifacts in positron emission tomography through retrospective stacking. Med Phys. Jul 2006;33(7):2632-2641. 80.

Dahlbom M. Estimation of image noise in PET using the bootstrap method. IEEE Trans Nucl Sci. 2002;49:2062-2066.

81. Kesner AL, Detorie N, Dahlbom M, Czernin J, Silverman DH. Respiratory Gated PET Derived From Raw PET Data. Paper presented at: IEEE annual medical imaging conference, 2007; Honolulu, HW. 82. Weber WA, Ziegler SI, Thodtmann R, Hanauske AR, Schwaiger M. Reproducibility of metabolic measurements in malignant tumors using FDG PET. J Nucl Med. Nov 1999;40(11):1771-1777. 83. Nagel CC, Bosmans G, Dekker AL, et al. Phased attenuation correction in respiration correlated computed tomography/positron emitted tomography. Med Phys. Jun 2006;33(6):1840-1847. 84. Gregory JK, Ronald HH. A 3D Optical Flow Approach to Addition of Deformable PET Volumes. Proceedings of the 1997 IEEE Workshop on Motion of Non-Rigid and Articulated Objects (NAM '97): IEEE Computer Society; 1997. 85.

Mirada medical imaging software. 2007.

287

86. Wolz G, Nomayr A, Hothorn T, et al. Anatomical accuracy of interactive and automated rigid registration between X-ray CT and FDG-PET. Nuklearmedizin. 2007;46(1):43-48. 87. Tang G, Wang M, Tang X, Luo L, Gan M. Pharmacokinetics and radiation dosimetry estimation of O-(2[18F]fluoroethyl)-L-tyrosine as oncologic PET tracer. Appl Radiat Isot. Feb 2003;58(2):219-225. 88. DeGrado TR, Baldwin SW, Wang S, et al. Synthesis and evaluation of (18)F-labeled choline analogs as oncologic PET tracers. J Nucl Med. Dec 2001;42(12):1805-1814. 89. Santens P, De Vos F, Thierens H, et al. Biodistribution and dosimetry of carbon-11-methoxyprogabidic acid, a possible ligand for GABA-receptors in the brain. J Nucl Med. Feb 1998;39(2):307-310. 90. DeGrado TR, Reiman RE, Price DT, Wang S, Coleman RE. Pharmacokinetics and radiation dosimetry of 18Ffluorocholine. J Nucl Med. Jan 2002;43(1):92-96. 91. Deterding TA, Votaw JR, Wang CK, et al. Biodistribution and radiation dosimetry of the dopamine transporter ligand. J Nucl Med. Feb 2001;42(2):376-381. 92. Kurdziel KA, Kiesewetter DO, Carson RE, Eckelman WC, Herscovitch P. Biodistribution, radiation dose estimates, and in vivo Pgp modulation studies of 18F-paclitaxel in nonhuman primates. J Nucl Med. Aug 2003;44(8):1330-1339. 93. ICRP Publication 62: Radiological Protection in Biomedical Research: International Commission on Radiological Protection; 1993. 94. Schmidt D, Langen KJ, Herzog H, et al. Whole-body kinetics and dosimetry of L-3--123I-iodo-alpha-methyltyrosine. Eur J Nucl Med. Sep 1997;24(9):1162-1166. 95. Ugur O, Kothari PJ, Finn RD, et al. Ga-66 labeled somatostatin analogue DOTA-DPhe1-Tyr3-octreotide as a potential agent for positron emission tomography imaging and receptor mediated internal radiotherapy of somatostatin receptor positive tumors. Nucl Med Biol. Feb 2002;29(2):147-157. 96. Palm S, Enmon RM, Jr., Matei C, et al. Pharmacokinetics and Biodistribution of (86)Y-Trastuzumab for (90)Y dosimetry in an ovarian carcinoma model: correlative MicroPET and MRI. J Nucl Med. Jul 2003;44(7):1148-1155. 97. Loevinger R, Budinger T, Watson E. MIRD Primer For Absorbed Dose Calculations. Vol 1-17. New York, NY: Society of Nuclear Medicine; 1988. 98. Heidelberger C, Chaudhuri NK, Danneberg P, et al. Fluorinated pyrimidines, a new class of tumour-inhibitory compounds. Nature. Mar 30 1957;179(4561):663-666. 99. Shani J, Wolf W. A model for prediction of chemotherapy response to 5-fluorouracil based on the differential distribution of 5-[18F]fluorouracil in sensitive versus resistant lymphocytic leukemia in mice. Cancer Res. Jul 1977;37(7 Pt 1):2306-2308. 100. Bellemann ME, Brix G, Haberkorn U, Ostertag HJ, Lorenz WJ. Drug-specific 19F NMR and dynamic 18F PET imaging of the cytostatic agent 5-fluorouracil. IEEE Transactions on Nuclear Science. 1994;41(6):2856-2861. 101. Bading JR, Alauddin MM, Fissekis JD, et al. Blocking catabolism with eniluracil enhances PET studies of 5[18F]fluorouracil pharmacokinetics. J Nucl Med. Oct 2000;41(10):1714-1724. 102. Dimitrakopoulou-Strauss A, Strauss LG, Schlag P, et al. Fluorine-18-fluorouracil to predict therapy response in liver metastases from colorectal carcinoma. J Nucl Med. Jul 1998;39(7):1197-1202. 103. Cristy M, Eckerman K. Specific absorbed fractions of energy at various ages from internal photons sources. Oak Ridge, TN: Oak Ridge National Laboratory; 1987. ORNL/TM-8381 V1-V7. 104. Stabin MG, Sparks RB, Crowe E. OLINDA/EXM: the second-generation personal computer software for internal dose assessment in nuclear medicine. J Nucl Med. Jun 2005;46(6):1023-1027. 105.

Kirschner A, Ice R, Beierwaltes W. Letters to the editor. J Nucl Med. 1975;16:248-249.

106. Siegel JA, Thomas SR, Stubbs JB, et al. MIRD pamphlet no. 16: Techniques for quantitative radiopharmaceutical biodistribution data acquisition and analysis for use in human radiation dose estimates. J Nucl Med. Feb 1999;40(2):37S-61S. 107. Vesselle H, Grierson J, Peterson LM, Muzi M, Mankoff DA, Krohn KA. 18F-Fluorothymidine radiation dosimetry in human PET imaging studies. J Nucl Med. Sep 2003;44(9):1482-1488.

288

108. Mankoff DA, Peterson LM, Tewson TJ, et al. [18F]fluoroestradiol radiation dosimetry in human PET studies. J Nucl Med. Apr 2001;42(4):679-684. 109. Bevington P, Robinson D. Data Reduction and Error Analysis for the Physical Sciences. 2nd ed. ed. New York, NY: McGraw-Hill; 1992. 110. ICRP Publication 53: Radiation Dose to Patients from Radiopharmaceuticals: International Commission on Radiological Protection; 1987. 111. Brix G, Bellemann ME, Haberkorn U, Gerlach L, Bachert P, Lorenz WJ. Mapping the biodistribution and catabolism of 5fluorouracil in tumor-bearing rats by chemical-shift selective 19F MR imaging. Magn Reson Med. Sep 1995;34(3):302-307. 112. Kissel J, Brix G, Bellemann ME, et al. Pharmacokinetic analysis of 5-[18F]fluorouracil tissue concentrations measured with positron emission tomography in patients with liver metastases from colorectal adenocarcinoma. Cancer Res. Aug 15 1997;57(16):3415-3423. 113. Brix G, Bellemann ME, Gerlach L, Haberkorn U. Intra- and extracellular fluorouracil uptake: assessment with contrastenhanced metabolic F-19 MR imaging. Radiology. Oct 1998;209(1):259-267. 114. Shani J, Young D, Schlesinger T, et al. Dosimetry and preliminary human studies of 18F-5-fluorouracil. Int J Nucl Med Biol. 1982;9(1):25-35. 115. Gangloff A, Hsueh WA, Kesner AL, et al. Estimation of paclitaxel biodistribution and uptake in human-derived xenografts in vivo with (18)F-fluoropaclitaxel. J Nucl Med. Nov 2005;46(11):1866-1871. 116. Kesner AL, Hsueh W-A, Htet NL, et al. Biodistribution and Predictive Value of [18F]-fluorocyclophosphamide in mice bearing human breast cancer xenografts. Journal of Nuclear Medicine. Fall 2007 2007. 117. Juweid ME, Cheson BD. Positron-emission tomography and assessment of cancer therapy. The New England journal of medicine. Feb 2 2006;354(5):496-507. 118. Downey RJ, Akhurst T, Gonen M, et al. Preoperative F-18 fluorodeoxyglucose-positron emission tomography maximal standardized uptake value predicts survival after lung cancer resection. J Clin Oncol. Aug 15 2004;22(16):3255-3260. 119. Bender H, Bangard N, Metten N, et al. Possible role of FDG-PET in the early prediction of therapy outcome in liver metastases of colorectal cancer. Hybridoma. Feb 1999;18(1):87-91. 120. Brun E, Kjellen E, Tennvall J, et al. FDG PET studies during treatment: prediction of therapy outcome in head and neck squamous cell carcinoma. Head & neck. Feb 2002;24(2):127-135. 121. Findlay M, Young H, Cunningham D, et al. Noninvasive monitoring of tumor metabolism using fluorodeoxyglucose and positron emission tomography in colorectal cancer liver metastases: correlation with tumor response to fluorouracil. J Clin Oncol. Mar 1996;14(3):700-708. 122. Romer W, Hanauske AR, Ziegler S, et al. Positron emission tomography in non-Hodgkin's lymphoma: assessment of chemotherapy with fluorodeoxyglucose. Blood. Jun 15 1998;91(12):4464-4471. 123. Schelling M, Avril N, Nahrig J, et al. Positron emission tomography using [(18)F]Fluorodeoxyglucose for monitoring primary chemotherapy in breast cancer. J Clin Oncol. Apr 2000;18(8):1689-1695. 124. Spaepen K, Stroobants S, Dupont P, et al. Early restaging positron emission tomography with ( 18)F-fluorodeoxyglucose predicts outcome in patients with aggressive non-Hodgkin's lymphoma. Ann Oncol. Sep 2002;13(9):1356-1363. 125. Wahl RL, Zasadny K, Helvie M, Hutchins GD, Weber B, Cody R. Metabolic monitoring of breast cancer chemohormonotherapy using positron emission tomography: initial evaluation. J Clin Oncol. Nov 1993;11(11):2101-2111. 126. Hautzel H, Muller-Gartner HW. Early changes in fluorine-18-FDG uptake during radiotherapy. J Nucl Med. Sep 1997;38(9):1384-1386. 127. Weber WA, Wieder H. Monitoring chemotherapy and radiotherapy of solid tumors. European journal of nuclear medicine and molecular imaging. Jul 2006;33 Suppl 1:27-37. 128. Metser U, Even-Sapir E. Increased (18)F-fluorodeoxyglucose uptake in benign, nonphysiologic lesions found on wholebody positron emission tomography/computed tomography (PET/CT): accumulated data from four years of experience with PET/CT. Seminars in nuclear medicine. May 2007;37(3):206-222.

289

129. Strauss LG. Fluorine-18 deoxyglucose and false-positive results: a major problem in the diagnostics of oncological patients. European journal of nuclear medicine. Oct 1996;23(10):1409-1415. 130. Higashi T, Fisher SJ, Brown RS, Nakada K, Walter GL, Wahl RL. Evaluation of the early effect of local irradiation on normal rodent bone marrow metabolism using FDG: preclinical PET studies. J Nucl Med. Dec 2000;41(12):2026-2035. 131. Udagawa T, Birsner AE, Wood M, D'Amato RJ. Chronic suppression of angiogenesis following radiation exposure is independent of hematopoietic reconstitution. Cancer Res. Mar 1 2007;67(5):2040-2045. 132. Laing EJ, Fitzpatrick PJ, Norris AM, et al. Half-body radiotherapy. Evaluation of the technique in normal dogs. J Vet Intern Med. Apr-Jun 1989;3(2):96-101. 133. Pachter JS, de Vries HE, Fabry Z. The blood-brain barrier and its role in immune privilege in the central nervous system. J Neuropathol Exp Neurol. Jun 2003;62(6):593-604. 134.

Galea I, Bechmann I, Perry VH. What is immune privilege (not)? Trends Immunol. Jan 2007;28(1):12-18.

135. 280.

Engelhardt B. Regulation of immune cell entry into the central nervous system. Results Probl Cell Differ. 2006;43:259-

136. van Waarde A, Cobben DC, Suurmeijer AJ, et al. Selectivity of 18F-FLT and 18F-FDG for differentiating tumor from inflammation in a rodent model. J Nucl Med. Apr 2004;45(4):695-700. 137.

Molla M, Panes J. Radiation-induced intestinal inflammation. World J Gastroenterol. Jun 14 2007;13(22):3043-3046.

138. Hallahan D, Kuchibhotla J, Wyble C. Cell adhesion molecules mediate radiation-induced leukocyte adhesion to the vascular endothelium. Cancer Res. Nov 15 1996;56(22):5150-5155. 139. Rube CE, Uthe D, Wilfert F, et al. The bronchiolar epithelium as a prominent source of pro-inflammatory cytokines after lung irradiation. International journal of radiation oncology, biology, physics. Apr 1 2005;61(5):1482-1492. 140. Agrawal A, Choudhary D, Upreti M, Rath PC, Kale RK. Radiation induced oxidative stress: I. Studies in Ehrlich solid tumor in mice. Mol Cell Biochem. Jul 2001;223(1-2):71-80. 141. Engenhart R, Kimmig BN, Strauss LG, et al. Therapy monitoring of presacral recurrences after high-dose irradiation: value of PET, CT, CEA and pain score. Strahlenther Onkol. Apr 1992;168(4):203-212. 142. Ohtsuka T, Nomori H, Watanabe K, et al. False-positive findings on [18F]FDG-PET caused by non-neoplastic cellular elements after neoadjuvant chemoradiotherapy for non-small cell lung cancer. Japanese journal of clinical oncology. May 2005;35(5):271-273. 143. Kubota R, Kubota K, Yamada S, Tada M, Ido T, Tamahashi N. Microautoradiographic study for the differentiation of intratumoral macrophages, granulation tissues and cancer cells by the dynamics of fluorine-18-fluorodeoxyglucose uptake. J Nucl Med. Jan 1994;35(1):104-112. 144. Kubota R, Yamada S, Kubota K, Ishiwata K, Tamahashi N, Ido T. Intratumoral distribution of fluorine-18fluorodeoxyglucose in vivo: high accumulation in macrophages and granulation tissues studied by microautoradiography. J Nucl Med. Nov 1992;33(11):1972-1980. 145. Shiraki N, Hara M, Ogino H, et al. False-positive and true-negative hilar and mediastinal lymph nodes on FDG-PET-radiological-pathological correlation. Annals of nuclear medicine. Feb 2004;18(1):23-28. 146. Nishikawa A, Furukawa F, Imazawa T, et al. Cell proliferation in lung fibrosis-associated hyperplastic lesions. Human & experimental toxicology. Sep 1995;14(9):701-705. 147. Castellucci P, Zinzani P, Nanni C, et al. 18F-FDG PET early after radiotherapy in lymphoma patients. Cancer biotherapy & radiopharmaceuticals. Oct 2004;19(5):606-612. 148. Minn H, Zasadny KR, Quint LE, Wahl RL. Lung cancer: reproducibility of quantitative measurements for evaluating 2-[F18]-fluoro-2-deoxy-D-glucose uptake at PET. Radiology. Jul 1995;196(1):167-173. 149. Kong FM, Frey KA, Quint LE, et al. A pilot study of [18F]fluorodeoxyglucose positron emission tomography scans during and after radiation-based therapy in patients with non small-cell lung cancer. J Clin Oncol. Jul 20 2007;25(21):3116-3123.

290

150. van Baardwijk A, Bosmans G, Dekker A, et al. Time trends in the maximal uptake of FDG on PET scan during thoracic radiotherapy. A prospective study in locally advanced non-small cell lung cancer (NSCLC) patients. Radiother Oncol. Feb 2007;82(2):145-152. 151. Wong CY, Thie J, Gaskill M, et al. Addressing glucose sensitivity measured by F-18 FDG PET in lung cancers for radiation treatment planning and monitoring. International journal of radiation oncology, biology, physics. May 1 2006;65(1):132137. 152. Kubota K, Ishiwata K, Kubota R, et al. Tracer feasibility for monitoring tumor radiotherapy: a quadruple tracer study with fluorine-18-fluorodeoxyglucose or fluorine-18-fluorodeoxyuridine, L-[methyl-14C]methionine, [6-3H]thymidine, and gallium-67. J Nucl Med. Nov 1991;32(11):2118-2123. 153. Stabin MG, Stubbs JB, Toohey RE. Radiation Dose Estimates for Radiopharmaceuticals: Oak Ridge Institute for Science adn Education; 1996. 154. Kalff V, Schwaiger M, Nguyen N, McClanahan TB, Gallagher KP. The relationship between myocardial blood flow and glucose uptake in ischemic canine myocardium determined with fluorine-18-deoxyglucose. J Nucl Med. Jul 1992;33(7):1346-1353. 155. Jones SC, Alavi A, Christman D, Montanez I, Wolf AP, Reivich M. The radiation dosimetry of 2 [F-18]fluoro-2-deoxy-Dglucose in man. J Nucl Med. Jul 1982;23(7):613-617. 156. Hays MT, Watson EE, Thomas SR, Stabin M. MIRD dose estimate report no. 19: radiation absorbed dose estimates from (18)F-FDG. J Nucl Med. Feb 2002;43(2):210-214. 157. Loevinger, Berman. MIRD Pamphlet No. 11 - A revised schema for calculating the absorbed dose from biologically distributed radiopharmaceuticals. Reston, VA: Society of Nuclear Medicine; 1976. 158. Deloar HM, Fujiwara T, Shidahara M, Nakamura T, Yamadera A, Itoh M. Internal absorbed dose estimation by a TLD method for 18F-FDG and comparison with the dose estimates from whole body PET. Phys Med Biol. Feb 1999;44(2):595-606. 159. Carnes BA, Olshansky SJ, Grahn D. An interspecies prediction of the risk of radiation-induced mortality. Radiat Res. May 1998;149(5):487-492.

291