in conjunction with the Medical Image Computing and Computer‐Assisted Interventions Conference
2017 International MICCAI BraTS Challenge September 14, 2017 Quebec City, Quebec, Canada (pre‐conference proceedings)
Fully Automated Glioma Brain Tumors Segmentation and Patient Overall Survival Prediction with SVMs Learning Algorithms: BraTS’2017 Challenge Alexander F. I. Osman[0000-0002-1286-475X] American University of Beirut Medical Center, Beirut, 1107 2020 Riad El-Solh, Lebanon
[email protected]
Abstract. The aim of this work was to develop a model for accurate auto-segmentation of the glioma brain tumors in multimodal MRIs and prediction of patient overall survival based on SVMs algorithms. BraTS’2017 datasets were used in this study. We developed a model based on an SVMs algorithms for autosegmentation of gliomas. Image intensity features were extracted for this purpose as well as pre- and post-processing on the MRIs were employed. The model was trained using the provided training datasets. Then the trained model was used to produce segmentation labels on the validation datasets. Also, the auto-segmented labels were called in combination the patient age parameter for OS classifications prediction. The OS prediction model was trained using the BraTS’17 OS data. A confusion matrix was plotted to evaluate the OS predictor performance in the three classification categories, i.e. long, short, and mid-survivors. The evaluated segmentation results of the edema sub-region tumor on Flair MRIs is not reported. However, for OS classification prediction the algorithm’s sensitivity and specificity metrics between the automatically predicted and clinically obtained OS were 100.0% for long-survivors (>15 months); short-survivors (15 months), mid-survivors (5 to 15 months), and short-survivors (15 months); short-survivors (