Longitudinal Brain Tumor Tracking, Tumor Grading, and Patient Survival Prediction Using MRI

This work aims to develop novel methods for brain tumor classification, longitudinal brain tumor tracking, and patient survival prediction. Consequently, this dissertation proposes three tasks. First, we develop a framework for brain tumor segmentation prediction in longitudinal multimodal magnetic resonance imaging (mMRI) scans, comprising two methods: feature fusion and joint label fusion (JLF). The first method fuses stochastic multi-resolution texture features with tumor cell density features, in order to obtain tumor segmentation predictions in follow-up scans from a baseline pre-operative timepoint. The second method utilizes JLF to combine segmentation labels obtained from (i) the stochastic texture feature-based and Random Forest (RF)-based tumor segmentation method; and (ii) another state-of-the-art tumor growth and segmentation method known as boosted Glioma Image Segmentation and Registration (GLISTRboost, or GB). With the advantages of feature fusion and label fusion, we achieve state-of-the-art brain tumor segmentation prediction. Second, we propose a deep neural network (DNN) learning-based method for brain tumor type and subtype grading using phenotypic and genotypic data, following the World Health Organization (WHO) criteria. In addition, the classification method integrates a cellularity feature which is derived from the morphology of a pathology image to improve classification performance. The proposed method achieves state-of-the-art performance for tumor grading following the new CNS tumor grading criteria. Finally, we investigate brain tumor volume segmentation, tumor subtype classification, and overall patient survival prediction, and then we propose a new context- aware deep learning method, known as the Context Aware Convolutional Neural Network (CANet). Using the proposed method, we participated in the Multimodal Brain Tumor Segmentation Challenge 2019 (BraTS 2019) for brain tumor volume segmentation and overall survival prediction tasks. In addition, we also participated in the Radiology-Pathology Challenge 2019 (CPM-RadPath 2019) for Brain Tumor Subtype Classification, organized by the Medical Image Computing & Computer Assisted Intervention (MICCAI) Society. The online evaluation results show that the proposed methods offer competitive performance from their use of state-of-the-art methods in tumor volume segmentation, promising performance on overall survival prediction, and state-of-the-art performance on tumor subtype classification. Moreover, our result was ranked second place in the testing phase of the CPM-RadPath 2019