Spatial spectral Gaussian mixture model approach for automatic segmentation of multispectral MR brain images

The Gaussian Mixture Model (GMM) is one of the most widely used models for statistical segmentation of brain Magnetic Resonance (MR) images. Because the GMM is a histogram-based model, has an intrinsic limitation which spatial information is not included. This problem causes the GMM to make good results only on images with low levels of noise and high level of contrast. In this paper, an automated algorithm for tissue segmentation multispectral magnetic resonance (MR) images of the brain is presented. We introduce a spatial spectral GMM which augment histogram information with spatial data using adaptive Markov random fields and real prior information which is generated form a spectral clustering. We have called this approach “Spatial Spectral Segmentation” (SSS). The Expectation-Maximization (EM) algorithm is utilized to learn the parameter-tied, spatial spectral Gaussian mixture model. Segmentation of the brain image is achieved by the affiliation of each pixel to the component of the model that maximized the a posteriori probability. Also we propose a complete preprocessing to obtain a comprehensive segmentation approach. The presented algorithm is used to segment Multispectral included T1, T2 and PD simulated and real MR images of the brain into three different tissues (WM, GM and CSF) The performance of the SSS based method is compared with that of popular EM segmentation. The experimental results show that the proposed method is robust.