Brain MR image segmentation and bias field estimation using coherent local and non-local spatial constraints

Clinical brain MR images usually contain noise and bias field (BF), which make the brain tissue segmentations difficult. Most of the current segmentation methods only focus on one unfavorable factor. The Coherent local intensity clustering algorithm (CLIC) algorithm proposed recently is good at dealing with the BF problem in images, but it has a poor anti-noise ability, for it doesn't consider non-local spatial constraint. In this paper, taking care of all these unfavorable factors simultaneously, we introduce the non-local spatial constraint into CLIC algorithm for brain MR image segmentations. Therefore, the proposed algorithm drives by both the coherent local and non-local spatial constraints. The coherent local information ensures the smoothness of the bias field estimation and the non-local spatial information reduces the noise effect during the segmentation. The proposed method has been successfully applied to brain MR images, and experiment results show that this method has stronger anti-noise property, smoother bias field estimation and higher segmentation precision than other reported fuzzy clustering algorithms.