A New Entropy Based Fuzzy Clustering Algorithm for Volumetric Noisy Brain MR Image Segmentation

Abstract

In this paper we have proposed a new entropy based fuzzy clustering algorithm for segmentation of volumetric noisy brain MR image data. The algorithm utilizes intensity distribution from spatial cubic local neighborhood characterizing a possibility measure that defines likeliness of a voxel under consideration to belong into a cluster or region. This is realized by judiciously defining a Gaussian density function. We then normalized these likeliness measures to use them as an alternative membership function. In addition to the fuzzy membership function, this normalized likeliness measure is also incorporated into the objective function using a regularizing parameter that resolves the trade-off between these two terms. Finally, a fuzzy entropy defined by Shannon’s function using the normalized likeliness measures is introduced that defines the vagueness and ambiguity uncertainty while classifying a voxel into its possible cluster. Therefore, the cluster prototypes of the proposed algorithm utilize the fuzzy membership functions, likeliness measures and fuzzy entropy. To validate the algorithm, we have performed both qualitative and quantitative analysis on noisy simulated and clinical brain MR image volumes. Its results are found to be superior while comparing with some of the state-of-the-art algorithms.