Systolic Function ISMRM May 7 2016 Alistair Young Department of Anatomy and Medical Imaging Department of Physiology Auckland Bioengineering Institute University of Auckland
Declaration of Financial Interests or Relationships Speaker Name: Alistair Young I have the following financial interest or relationship to disclose with regard to the subject matter of this presentation: Company Name: Siemens Type of Relationship: Consultant
CMR Analysis • Imaging + Analysis = Disease Evaluation •
Cardiac imaging biomarkers are important indicators of disease and prognosis
• Modern CMR provides a wealth of data, including: • • • • • • • • •
LV and RV: mass, volumes, ejection and filling rates (E and A) Regional: wall thickness, strain, T1, T2, scar (LGE), ECV, perfusion, asynchrony LA and RA: areas, volumes, scar Aortic: distensibility, diameter, PWV Aortic valve: regurgitant fraction, area, leaflets, jet Mitral valve: regurgitant fraction, area, leaflets Spectroscopy: proton, sodium Flow: turbulence, shear stress, pressure gradient Mechanics: myocardial compliance, active tension generation, stress, work Li et al. JACC Imaging 2010;3:860-866
CMR analysis standards • Post-processing protocols •
Schulz-Menger JCMR 2013;15:35
• Normal values •
Kwael-Boehm JCMR 2015;17:29
• T1 Mapping •
Kellman JCMR 2014;16:2
• Perfusion •
Jerosch-Herold JCMR 2010;12:57
• Phase Contrast •
Nayak JCMR 2015; 17:71
• 4D Flow •
• …
Dyverfeldt JCMR 2015;17:72
SSFP Cardiac Imaging
Young and Prince. Ann Rev Biomed Eng 15:433-61; 2013
Young et al. MRM in Revision
vi. The LV end-systolic image should be chosen as the image with the smallest LV blood volume. For its identification, the full image stack has to be
LV volumetrics
valve cusps are identified on the basal slice(s) the contour is drawn to include the outflow tract to the level of the aortic valve cusps.
Figure 1 Left ventricular (LV) chamber quantification. For LV chamber quantification, the endocardial (blue) and epicardial (yellow) contours are delineated in diastole (top) and systole (bottom) in a stack of short axis slices that cover the whole left ventricle. a) and b) Illustrates the approach with inclusion of the papillary muscles as part of the LV volume. c) and d) Shows the approach with exclusion of the papillary muscles from the LV volume.
Schulz-Menger et al. JCMR 2013;15:35
Consensus Ground Truth
LVOT
Apex
Base
Fig. 2 Three examples of difficult cases with larger reader disagreement, showing how STAPLE can estimate consensus contours that gives the Suinesiaputra et al. JCMR 2015;17:63
Consensus Ground Truth R7 R6 R5 R4 R3 R2 R1 -40
-20
0
20
40
-40
-20
EDV
0
20
40
ESV
-40
-20
0
20
EPIVOL
Fig. 3 Reader bias and precision against the estimated consensus. Each bar denotes the mean ± one standard deviation
thatet the al. outflow tract 2015;17:63 was contoured as part of the LV cavSuinesiaputra JCMR
Limitations
40
Segmentation Challenge
AO AU DS INR SCR
LV mass (g)
!1.81 (6.42) !2.14 (6.71) 6.76 (10.32) 124.35 (36.08) !1.99 (25.67)
Good Segmentation
4
6
Oversegmentation
[6.64] [7.01] [12.29] [129.42] [25.61]
INR
2
AO AU
DS
SCR
−2
0
Logit (p)
(a) Basal slice
(b) Mid-ventricular slice
Undersegmentation
−4
Poor Segmentation
−4
−2
0
2
4
6
Logit (q)
(c) Apical slice
Fig. 9. Distributions of sensitivity and specificity values after Logit transformation (13). The value of Logit(0) is equal to sensitivity/specificity at 0.5. Therefore regions centered at the origin (dashed lines) define four characteristics of segmentation results. The rater labels are at the peak of each distribution, with the colors are: INR in red, SCR in green, and manual raters (AU, AO and DS) in black. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
Figure 5: Three representative examples of raters and CS* taken from base (a), mid-ventricular (b) and apical (c) slices.
Table 5 Suinesiaputra al.binary Medical Analysis 18 (2014) To measure the similarityet of two images, we Image ejection fraction (EF) and LV mass. Endocardial and50–62 Bhattacharyya distances between each distributions in Fig. 9.
!24.69 (16.8 !15.41 (13.1 !21.47 (20.4 90.23 (32.52) !36.18 (26.2
LV Segmentation Challenge • https://www.kaggle.com/c/second-annual-data-sciencebowl NIH/Booz Allen Hamilton • Andrew Arai, Michael Hansen • 14 March deadline • $200,000 in prizes
Feature Tracking •
Non-rigid registration
Li et al. JACC Imaging 2010;3:860-866
Li et al. JACC Imaging 2010;3:860-866
Shape and Function • Cardiac shape and function are important predictors, and indicate previous exposure to risk factors • ESV, sphericity is predictive of worse outcomes after myocardial infarction
Normal
Concentric
Hinojar et al Individualized cardiovascular risk assessment by cardiovascular magnetic resonance. Future Cardiology Vol. 10, No. 2, Pages 273-289, March 2014.
Eccentric
Shape and Motion • 3D beating model • Interactively customized to patient images • Relatively compact representation • Suitable for biomechanics, and atlasing
From Shape to Pathology
40
SDI = 1.38%
0
20
EF (%)
60
Regional wall motion
0.2
0.4
0.6
0.8
1.0
20
SDI = 21.96%
-20
0
EF (%)
40
60
0.0
0.2
0.4
0.6
0.8
1.0
0
20
40
SDI = 17.09%
-20
EF (%)
60
80
0.0
0.0
0.2
0.4
0.6
0.8
1.0
experiments, IMCA required 8.13 s processing compared with 0.75 s for LDA on a standard desktop (Intel i5 quadprocessor 3.4 GHz, 8 GB RAM). Nine logistic regression models were studied (Table 4; Fig. 4). The baseline model included only the sex, age, height, weight, diastolic blood pressure and history of Table 2 LDA and IMCA Scores for MESA and DETERMINE diabetes. The MASSVOL model include baseline vari(mean ± SD) ables as well as ED volume, ES volume and LV mass ED since these are the standard remodeling indices currently MESA DETERMINE p value used clinically [17]. Also, for comparison with [6], an ED LDA −0.30 ± 0.61 1.99 ± 0.77 <0.0001 ESVI+EDVI model was formulated to include ES volES LDA −0.33 ±10th 0.48 2.18 20th ± 0.80 <0.0001 1st 30th 40th and ED volume 50th index (together 60th 80th 90th 99th ume index with 70th baseline ED&ES LDA −0.34 ± 0.44 2.25 ± 0.73 <0.0001 variables). IMCA and LDA models included the baseED IMCA −0.29 ± 0.66 1.94 ± 0.65 <0.0001 line variables plus the single standardized index derived ES IMCA −0.31 ± 0.58 2.07 ± 0.68 <0.0001 from IMCA or LDA respectively. Both IMCA and LDA −0.32 ± 0.56 2.13 ± 0.57 <0.0001 modes showed very high odds ratio of the disease (all ESED&ES IMCA ORs were over 100). All goodness-of-fit measures (Deviance, AIC, BIC and AUC) of the IMCA and LDA models were smaller than the baseline model and the MASSVOL 0.25 model. ES shape feature models showed better perforDETERMINE Asymptomatic Myocardial Infarction MESA mance than the analogous ED shape feature models for IMCA and LDA. The combination of ED&ES Fig. 3 The representation of disease remodeling. In theshape figure, the corresponding shapes from the 0.2 derived shape indices allow for a continuousboth features also improved agreement over just ES or ED and myocardial infarct group (DETERpercentiles of the IMCA ED&ES index are shown. Mean values (black triangles) for the asymptomatic (MESA) shape features separately. Finally, the combined ED&ES MINE) show 0.15 over 50 percentiles of separation for this index. Percentiles correspond to the histogram shown in Fig. 2 IMCA logistic model achieved the lowest Deviance, AIC, BIC and highest AUC. Considering the AUC as a measure of discrimina0.1 tory power, all LDA and IMCA modes had significantly www.cardiacatlas.org/challenges/ more discrimination than the baseline (p < 0.05) and lv-statistical-shape-modelling-challenge/ Table 4 0.05 Assessment table showing measures of goodness-of-fit for the logistic regression models MASSVOL models (p < 0.05). Botheight the LDA and IMCA ED&ES coupled modes showed better discrimination LR coefficient (β1) than either σ (βthe value(p < 0.05). Deviance AIC BIC AUC (%) 1) ED and ES Pmodes The IMCA 0 -3 -2 -1 0 1 2 3 4 ED&ES and IMCA ED showed better discrimination Standardized IMCA Scores Baseline – – corresponding LDA – modes (p < 0.05), 1500 1518 1569 76.94 than their but the of IMCA scores at ED&ES between MESA and DETERMINE is shown in Fig. 2. The asymptomatic group and the myocardial infarction group were best discriminated with IMCA scores. The Pearson correlation coefficients
Probability
Shape Classification Challenge
Congenital Heart Disease
Step 1: Segment exemplar
Step 2: Align
Step 3: Define Template
Step 4: Subdivide
• ~300 cases contributed • customizable shape templates • biomechanical analysis cardiacatlas.org
Step 5: Customize
Step 6: Modeling
Tagging
Young and Prince. Ann Rev Biomed Eng 15:433-61; 2013
Tagging
Young and Prince. Ann Rev Biomed Eng 15:433-61; 2013
DENSE
Young and Prince. Ann Rev Biomed Eng 15:433-61; 2013
DENSE
Young et al. Magn Reson Med 2012; 67(6):1590-1599
Generalized Strain Analysis
Young et al. Magn Reson Med 2012; 67(6):1590-1599
Motion Tracking Challenge
Tobon-Gomez Med Image Anal, 17(6):632-48; 2013
Feature Tracking
Original
Myo
ExMyo
Contour
Cowan et al. JACC Imaging 2015;12:1465-1466
Strain
• • • •
Change in shape Depends on orientation Principal strain oriented obliquely Torsion = 90-α
Fonseca et al. Am J Cardiol 2004, 94:1391-1395 (2004)
Torsion
• Normalize torsion for heart size by multiplying by radius
θCL =
(φapex −φbase )(rapex + rbase )
rbase φbase D φapex
2D rapex
Young AA JCMR 2012; 14:49
θCL
Strain Rate • Relaxation rate • Depends on peak strain
Fonseca et al. Am J Cardiol 2004, 94:1391-1395 (2004)
Strain • Aging: RC â; RL â; T á; RT â (Fonseca Am J Physiol 2004)
• Type 2 diabetes: C â; L â; T â; RC â; RL â; RT â (Fonseca Am J Cardiol 2004)
• Hypertrophic cardiomyopathy: C â; L â; T á; (Young Circ 1993)
• carriers with normal wall thickness: T á; (Rüssel JCMR 2011)
• Dilated cardiomyopathy: regional (Young Cardiovasc Res 2001)
• Aortic stenosis C â; T á; (Delhaas MRM 2004)
MRE
Anisotropic stiffness
60 Hz
0 mm
0.2 mm
damping
Transverse stiffness (kPa)
Fibre stiffness (kPa)
Poisson’s ratio
Renee Miller, Poster 1959 Li et al. JACC Imaging 2010;3:860-866
Summary • SSFP systolic function • • •
Manual contouring requires extensive training Automated methods becoming available Include/exclude trabeculations
• Strain • • •
DENSE has the best resolution, but longest acquisition time Tagging is the most widely accepted reference Non-tagged SSFP images can be used for global strain • maybe regional strain in patients with regional disease
• Standards and Benchmarks •
needed for validation and training
Li et al. JACC Imaging 2010;3:860-866
Investigators (Auckland)
Alistair Young
Avan Suinesiaputra
Martyn Nash
Brett Cowan
Pau MedranoGracia
Kathleen Gilbert
Investigators (UCSD)
Andrew McCulloch
Jeff Omens
Bioengineering
Bioengineering
James Perry
Lucila Ohno-Machado
MD, Rady Children's Hospital
Bioinformatics