A Modified MRF Algorithm Based on Neighborhood Spatial Information for MRI Brain Tissue Segmentation

\ The paper presents a brain tissue segmentation algorithm in magnetic resonance images based on neigh-borhood spatial information combining classical FCM (Fuzzy C-Means) clustering method and classical MRF (Markov Random Field) method. The accurate clustering center has an important effect on the segmentation results with classical MRF method; while the FCM clustering method has the advantage of the accuracy of the clustering center, but it doesn't take full account of spatial information. In this paper, first, classical FCM clustering method added neighborhood spatial information is adopted to obtain more accurate initial parameters for the further segmentation; and subsequently, in order to improve the accuracy of segmentation results, neighborhood spatial information is also added to classical MRF method to conduct twice segmentation. Experiments are carried out on clinical test data and IBSR data sets. Experimental results show that the proposed algorithm has more accurate segmentation of MRI (Magnetic Resonance Imaging) brain tissue comparing with the classical FCM and classical MRF methods, demonstrating that the effective usage of the neighborhood spatial information can obtain more accurate initial parameters and increase the accuracy of the algorithm, and the proposed algorithm is superior to the classical FCM and MRF methods.