The segmentation of Magnetic Resonance (MR) brain images into cerebrospinal fluid (CSF), gray matter (GM) and white matter (WM) is an intensive research direction in the field of medical image analysis. The accuracy of the tissue segmentation results has a critical impact on the following neurological and image processing applications. The Gaussian mixture model (GMM) is commonly utilized by the study community for the representation of the voxel intensity distribution among the three tissues. Unfortunately, standard GMM-based methods often ignore to take the spatial information within voxel neighborhoods into consideration. Hence, the segmentation quality shows obvious sensitivity to initialization conditions and image noise. In this paper, an improved GMM model is proposed to improve the segmentation accuracy. Firstly, the GMM is established to characterize the intensity distributions of different tissues. Secondly, the spatial information is taken into account to determine the prior distributions of different tissue types, with the neighborhood entropy as the weighted term. Moreover, the Expectation Maximization (EM) algorithm is employed iteratively as the optimizer to estimate the proposed model parameters. The performance of the proposed method is validated by real MR brain images, demonstrating its advantages in accuracy and robustness over other GMM-based approaches.
All Science Journal Classification (ASJC) codes
- Radiology Nuclear Medicine and imaging
- Health Informatics
- Entropy Weighting
- Gaussian Mixture Model
- MR Brain Image Segmentation
- Spatial Information