Support vector machine (SVM) active learning for automated Glioblastoma segmentation

Abstract

Accurate segmentation of Glioblastoma multiforme (GBM) from MR images is important for sub-typing in diagnosis, determining tumor margins in surgical planning, and selecting appropriate therapies. However, it is a challenging and time-consuming task because GBM has a variety of imaging characteristics and often deforms nearby tissues in the brain. In this paper, we propose a support vector machine (SVM) active learning approach to address the problem of GBM segmentation from multi-modal MR images. First, a knowledge-based fuzzy clustering algorithm is performed to segment the brain tissues into six classes including white matter (WM), grey matter (GM), cerebrospinal fluid (CSF), T2-hyperintense regions, necrosis and enhanced tumor. Then, the SVM active learning approach is applied to refine the segmentation. Comparative studies with other segmentation methods indicate that the proposed algorithm can segment GBM more accurately.