A recursive framework for bias field estimation and segmentation of noisy brain MR images

Estimation of bias field together with the tissue class of a noisy Magnetic Resonance image has been a challenging task because of the nonlinear nature of bias field. In order to address this issue we have proposed two new schemes. The first one is the recursive framework, where class labels and bias fields have been estimated simultaneously. In one part of the recursion, a variable variance Adaptive Meanshift based Algorithm (VVAMS) with inherent mode seeking ability has been proposed to eliminate noise. In the other part of the recursion, class labels have been estimated using the modified possibilistic fuzzy algorithms of Jeetashree et al. [1]. The proposed scheme has successfully been tested with a large number of noisy slices from Brainweb database. The quantitative evaluation of the segmented image demonstrates the improved performance of the proposed algorithm as compared to other existing methods.