Segmentation and Recovery of Pathological MR Brain Images Using Transformed Low-Rank and Structured Sparse Decomposition

Abstract

We present a common framework for the simultaneous segmentation and recovery of pathological magnetic resonance (MR) brain images, where low-rank and sparse decomposition (LSD) schemes have been widely used. Conventional LSD methods often produce recovered images with distorted pathological regions, due to the lack of constraint between low-rank and sparse components. To address this issue, we propose a transformed low-rank and structured sparse decomposition (TLS2D) method, which is robust for extracting pathological regions. Moreover, the well recovered images can be obtained using both structured sparse and computed image saliency as the adaptive sparsity constraint. Experimental results on MR brain tumor images demonstrate that our TLS2D can effectively provide satisfactory performance on both image recovery and tumor segmentation.