Spatially Constrained Likeliness-based Fuzzy Entropy Clustering Algorithm and its Application to Noisy 3D Brain MR Image Segmentation

Abstract

This paper proposes a spatially constrained likeliness-based fuzzy entropy clustering algorithm for noisy 3D brain MR image segmentation. It introduces a likeliness measure with respect to a voxel under consideration by using intensity distribution surrounding its local neighborhood. We use this measure as an additional membership function and named as fuzzy likeliness measures. We integrate these two fuzzy membership functions into a objective function by means of a regularizing parameter. Further, we introduce a fuzzy entropy using the fuzzifier weighted fuzzy likeliness measures to define the information uncertainty associated with a voxel in order to finding its cluster. By integrating weighted fuzzy membership function and fuzzy likeliness measure we generate the final membership function. The experiments on noisy 3D brain MR image volumes that include simulated and clinical data suggest that the proposed algorithm is superior while comparing with several state-of-the-art algorithms in terms of Dice coefficient, partition coefficient and partition entropy.