Machine Learning-Based Automated Methods for Brain Tumor Segmentation, Subtype Classification, Tracking and Patient Survival Prediction

Abstract

This chapter discusses novel machine learning and context-aware deep learning methods for automated brain tumor growth modeling, tumor segmentation, tumor subtype classification, tumor tracking and patient survival prediction using multimodal magnetic resonance imaging (mMRI). We first report a machine learning framework for brain tumor growth modeling, tumor segmentation and tracking in longitudinal mMRI scans, comprising of two methods: feature fusion and joint label fusion (JLF). The first method combines 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 fuse segmentation labels obtained from (i) the stochastic texture feature-based and Random Forest (RF)-based tumor segmentation methods; and (ii) another state-of-the-art tumor growth and segmentation method known as boosted Glioma Image Segmentation and Registration (GLISTRboost, or GB). We then discuss a novel deep learning pipeline, known as Context-Aware Convolutional Neural Network (CANet), for tumor segmentation, tumor subtype classification and patient survival prediction. We evaluate the methods using dataset of multimodal Brain Tumor Segmentation Challenge (BraTS) 2015, BraTS 2019, and the Computational Precision Medicine: Radiology-Pathology Challenge on Brain Tumor Classification 2019 (CPM-RadPath), respectively. The evaluation result shows the proposed methods achieve state-of-the-art in longitudinal brain tumor tracking and tumor subtype classification. The performances also suggest that the proposed methods offer promising results in tumor segmentation and overall survival prediction. Moreover, our result has been ranked second for tumor subtype classification in the testing phase of 2019 CPM-RadPath Challenge.