Local Feature Extraction and Information Bottleneck-Based Segmentation of Brain Magnetic Resonance (MR) Images

Pengcheng Shen,Chunguang Li

doi:10.3390/e15083295

Abstract

Automated tissue segmentation of brain magnetic resonance (MR) images has attracted extensive research attention. Many segmentation algorithms have been proposed for this issue. However, due to the existence of noise and intensity inhomogeneity in brain MR images, the accuracy of the segmentation results is usually unsatisfactory. In this paper, a high-accuracy brain MR image segmentation algorithm based on the information bottleneck (IB) method is presented. In this approach, the MR image is first mapped into a “local-feature space”, then the IB method segments the brain MR image through an information theoretic formulation in this local-feature space. It automatically segments the image into several clusters of voxels, by taking the intensity information and spatial information of voxels into account. Then, after the IB-based clustering, each cluster of voxels is classified into one type of brain tissue by threshold methods. The performance of the algorithm is studied based on both simulated and real T1-weighted 3D brain MR images. Our results show that, compared with other well-known brain image segmentation algorithms, the proposed algorithm can improve the accuracy of the segmentation results substantially.

Highlights

IntroductionBased on the capability of helping brain-disease diagnosis and neuroscience research, segmentation of brain tissue in MR images is a topic of major interest in the field of brain magnetic resonance (MR)
Based on the capability of helping brain-disease diagnosis and neuroscience research, segmentation of brain tissue in MR images is a topic of major interest in the field of brain magnetic resonance (MR).Among various segmentation schemes of brain MR images, manual segmentation performed by medical experts is usually considered to be the ground truth
We focus on the issue of segmenting brain MR images and propose a segmentation algorithm based on the soft-version information bottleneck (IB) method

Summary

Introduction

Based on the capability of helping brain-disease diagnosis and neuroscience research, segmentation of brain tissue in MR images is a topic of major interest in the field of brain magnetic resonance (MR). We focus on the study of this question, and in this paper, we propose an IB-based segmentation algorithm, which automatically segments the brain MR images with high accuracy. In [24], the authors presented an image segmentation algorithm based on a hard version of the IB method and studied its performance on general images. We focus on the issue of segmenting brain MR images and propose a segmentation algorithm based on the soft-version IB method. A soft-version IB method performs a soft clustering in the local-feature space By this process, one original brain MR image is clustered into several clusters of voxels.

IB-Based Segmentation Method

Feature Extraction in Brain MR Images

IB-Based Clustering

Brain Tissue Classification

Results and Discussion

Performance of the Algorithm under Empirical Parameters

Effect of the Free Parameters on the Algorithm’s Performance

Conclusions