Markov multiple feature random fields model for the segmentation of brain MR images

Abstract

In the research of expert and intelligent systems on three-dimensional (3-D) brain magnetic resonance (MR) images, segmentation is the fundamental step of quantitative analysis of brain tissues. Due to the influence of various factors on brain MR images, segmentation is challenging. The Markov random field (MRF) model is promising for segmentation given some recent encouraging results. However, the traditional MRF model usually relies on a single feature. In this paper, we first propose a novel Markov multiple feature random fields (MMFRF) model, which is able to combine various types of features into a unified decision model using the Bayesian framework. Second, we mainly focus our research on two feature random fields in the MMFRF model for the segmentation of brain MR images. The intensity feature random field and the texture feature random field are combined into a unified framework to model the image. In particular, we use patch-based 3-D texture features through gray-level co-occurrence matrix (GLCM) statistics to construct the texture feature random field. The performance of our proposed method is compared with some state-of-the-art approaches on both real and simulated brain MR datasets. The experimental results demonstrate that the performance of the proposed method is superior to the competing approaches. In theory, the traditional MRF model can be treated as a special case of the proposed general MMFRF model where only one feature random field is considered. Furthermore, the results also show the feasibility of employing the proposed method, which provides accurate and efficient brain tissue segmentation, to develop effective expert and intelligent systems for brain MR images.