S516

Int J CARS (2012) 7 (Suppl 1):S507–S521 usefulness to diagnostic assistance. This demonstrates that quantification using regional extraction is effective in the assisted diagnosis of at least some cancers. In the future, we would like to raise verification precision for uterine cervix cancer determination by increasing the number of cases. We would also like to conduct similar evaluations for other areas to verify the degree of complexity of case to which this technology is applicable.

Fig. 1 System operation screen. The entire screen (right) and data input portion of the screen (left)

Tumor classification for volumetric tracking of optic pathway gliomas in longitudinal studies: method and case study L. Weizman1, D. Ben Bashat2, L. Joskowicz1, B. Shofty2, S. Constantini2, L. Ben Sira2 1 The Hebrew University of Jerusalem, School of Engineering and Computer Science, Jerusalem, Israel 2 Tel Aviv Sourasky Medical Center, Tel Aviv, Israel Keywords MRI
Brain tumors
Follow up
OPG

Fig. 2 Output result. a Region recognition, b Numeric output

extraction uses three parameter sets to search simultaneously for the cell, the nucleus, and the intranuclear structures, i.e., three differentsized types of structures. Determination of cancer probability In this experiment, we investigated for the presence of squamous cell carcinoma in the uterine cervix. Squamous cell carcinoma has numerous features. We selected nucleus size, the easiest among them to quantify, as the determination criteria. A high probability of cancer is judged for nuclei larger than a specified size that quantify to greater than a specified value. Determination was divided into three levels, high probability, suspect, and not suspect. Display of regional extraction results Regional extraction results were displayed so that the cell, nucleus, and intranuclear structure were clearly distinguished. Physicians were then asked to check the numbers of suspect nucleus and evaluate the state of regional extraction, see Fig. 2. Results Three cases each of squamous cell carcinoma specimens, specimens with no abnormalities, and non-cancerous pathological specimens were prepared. The system’s output results (i.e., the system’s determination) and diagnosis according to physician criteria were compared. Physicians also viewed regional extraction images and evaluated the correctness of regional extraction. This produced the following results for regional extraction. Intranuclear structure and cell regional extraction was characterized by numerous areas of inaccurate judgment. In contrast, nucleus regional extraction and determination based on nucleus size exhibited minimal error and agreed with physician judgment. The system agreed with physician judgment of cancer probability for the nine cases. Conclusion Though regional extraction results were not accurate for some areas, such as intranuclear structures and cell outlines, nucleus extraction, critical to determination, exhibited high precision, showing its

123

Purpose Tumor volume change over time is an important parameter for therapy assessment. Currently, this assessment is mostly based on linear measurements. Manual quantification of brain tumor volume, although possible and considered the ground truth, is time consuming and may suffer from inconsistencies of the delineated tumor boundaries between and within observers over time [1]. This inter/intra observer variability may mislead the evaluation tumor progression and can result in improper treatment decisions. For instance, Weltens et al. [2] show that the interobserver variability of the same tumor, delineated by nine different observers, may differ by up to 30 % of the mean volume. Our goal is to develop a volumetric tumor growth model to quantify the tumor volume and that of its internal components over time. Previous publications on OPG segmentation focused on the segmentation of OPG at a single time point, while in this work we focus on the longitudinal internal changes of the tumor. The aim is to decrease the inter-/intra-observer variability and to provide accurate monitoring of changes in the tumor following treatment. This will provide a new perspective on tumor burden and natural history and enable a better estimation of treatment effectiveness. Methods We have developed and implemented a method for volumetric quantification of brain tumor and its internal components. The method relies on a baseline segmentation and a tumor growth model (Fig. 1). The method consists of four steps. In step 1, the MR pulse sequences of every scan are co-registered to a representative pulse sequence within that scan, followed by co-registration of the follow-up scan to

Fig. 1 Block diagram of the proposed tumor follow-up method

Int J CARS (2012) 7 (Suppl 1):S507–S521

S517 Tumor response prediction to neo-adjuvant chemotherapy for breast cancer: dependence on ultrasonic vascularity morphology D.-R Chen1, W.-T. Cheng2, Y.-L.Huang2 1 Changhua Christian Hospital, Laboratory of Cancer Research, Changhua, Taiwan 2 Department of Computer Science, Tunghai University, Taichung, Taiwan Keywords Neo-adjuvant chemotherapy
Vascularity morphology
Tumor response
Breast cancer

Fig. 2 Automatically generated progression graph

the baseline scan. In step 2, the input tumor boundaries of the baseline scan are overlaid on the follow-up scan to detect boundary segments that have changed. In step 3, internal classification of the overlaid tumor area is computed from the follow-up scan data. In step 4, the detected boundary segments from step 2 are updated using the result of step 3. We have also developed an application that manages the MR scans of the patients and enables proper examination of the tumor development and the natural history of the patient. Results Four patients with Optic Pathway Gliomas (OPGs) participated the study. Each patient was scanned at 5 time points during the period of the study. Experimental results yield a mean surface distance error of 0.28 mm and a mean volume overlap difference of 12.34 % as compared to manual segmentation by an expert radiologist. Our experimental results indicate that our method accurately segments and classifies OPGs in the follow-up scheme. In addition, our application enables to explore the progression of the internal components of the tumor over time, and sheds light on the internal behavior of the tumor. Figure 2 shows the progression of the tumor and its components over time for patient 1. Although there seems to be tumor progression during the period of the study, a deeper look into the components of the tumor reveals that the increase in the gross total volume of the tumor is because of the progression of cyst component, which in many cases can be treated with a cerebral shunt. Conclusion Our method effectively quantifies the volume of the tumor and its components. The volume overlap error is significantly lower than the one provided by methods of similar applications, thanks to the strong prior of the tumor delineation of the baseline image. The heterogeneity of OPGs, the long course, and the young age of the patients are all driving the demand for more efficient and accurate means of tumor follow up. Our follow-up application brings awareness to internal tumor changes over time, which is very important for the decision of the right moment for treatment. References [1] Fiorino C, Reni M, Bolognesi M, Cattaneo G, Calandrino M. (1998) Intra and interobserver variability in contouring prostate and seminal vesicles: implications for conformal treatment planning. Radiotherapy and Oncology. 47:285–292. [2] Weltens C, Menten J, Feron M, Bellon E, Demaerel D, Maes F, den Bogaert WV, van der Schueren, E. (2001) Interobserver variations in gross tumor volume delineation of brain tumors on computed tomography and impact of magnetic resonance imaging. Radiotherapy and Oncology. 60:49–59.

Purpose Tumor vascularity, an important factor correlated with tumor malignancy, would be used to evaluate the effect of the neo-adjuvant chemotherapy prior to surgery. High-definition flow (HDF) Doppler ultrasound was performed to investigate blood flow and solid directional flow information in breast tumors. In this study, vascularity morphology features from HDF power Doppler ultrasound imaging were extracted as early predictors for evaluate chemotherapy effects. Firstly, this study designed an automatic method to extract vascular centre-lines from the tumor area. Then the vascularization indices were estimated from the extracted vascular centre-lines. Finally, a decision tree model with all characteristics was employed as a tumor response predictor to neo-adjuvant chemotherapy for breast cancer. Methods Data acquisition Thirty-two consecutive T2 breast cancer (Tumor size [2 cm and B5 cm) patients, who received neo-adjuvant chemotherapy were recruited for this study. The diagnosis of breast cancer was made by core needle biopsy. Pre-operative intravenous chemotherapy was given for six courses in each patient and 3 weeks per cycle. Epirubicin (Pharmorubicin, Pfizer Pharmaceuticals, New York City, NY, USA) 80–90 mg/m2, cyclophosphamide 500 mg/m2 and 5-Fluorouracil 500 mg/m2 on day 1 every three weeks. Sonographic examinations were done (period N1–N6) by using 3D power Doppler ultrasound with the HDF function (Voluson 730, GE Medical Systems, Zipf, Austria, equipped with RSP 6-12 transducer). The period N0 was the sonographic before the chemotherapy. Vascular feature extraction Figure 1 shows an example of vascular centre-line extraction procedure from the an HDF Doppler ultrasound imaging. This study performed 3D Gaussian low-pass filter to smooth the vascular images. After pre-processing, an automatic extracting method was performed to locate centre-line of each vessel. An efficient parallel thinning algorithm was utilized to extract vascular centre-lines. The proposed method directly extracted vascular centre-lines from elongated 3D binary objects and provided good results and preserved topology. Selected vascularity morphology features were then estimated from the extracted vascular centre-lines: (1) (2)

(3) (4) (5)

Number of branch (denoted NB)—The total number of branch in a 3D HDF image; Shortest distance between vessels and the tumor center (denoted SDVC)—The shortest distance between the vascular center-line and the barycenter of tumor; Number of tree (denoted NT)—The total number of vascular tree; Standard deviation of vascular direction (denoted DE); Entropy of vascular direction (denoted EN);

This study calculated the five features from the tumor area and shell outside thickness 3 mm-surrounding the breast lesion. Evaluation of chemotherapy response The chemotherapy treatment effect of the 32 patients was evaluated by the clinical tumor response. The clinical tumor response was classified as complete response (CR), partial response (PR), stable

123