A New Algorithm for Segmentation of Brain MR Images with Intensity Nonuniformity Using Fuzzy Markov Random Field

It is difficult to segment MR images accurately due to factors such as noise contamination, intensive inhomogeneity and partial volume effect (PVE). In this paper, a fuzzy MRF model based on its conventional version is developed for the segmentation of MR images with intensive inhomogeneity. By solving the mathematic model, we derive a formula to compute the membership values for each voxel with respect to different categories, and a estimation of intensive inhomogeneity. We thus propose an efficient and unsupervised algorithm to implement the accurate segmentation for MR brain images. The simulated brain images and real clinical images are selected to test the proposed algorithm. The experimental results show that the segmentation accuracy is improved significantly in comparison with either conventional model-based algorithms or fuzzy C-mean segmentation algorithms.