Adaptive Fuzzy Evidential Reasoning for Automated Brain Tissue Segmentation

This paper presents an adaptive fuzzy evidential reasoning approach, for segmenting multi-modality MR brain images. A novel fuzzy evidence structure model is proposed under the assumption that each information source provides two types of evidence: probabilistic evidence and fuzzy evidence. A new information measure, called hybrid entropy, is employed for evaluating the overall uncertainty contained in a fuzzy evidence structure. For adaptive reasoning, two discounting strategies are included. To handle conflict between the probabilistic evidence and the fuzzy evidence, local discounting takes into account Kullback-Leibler distance between the two types of evidence. Global discounting takes into account source quality, in terms of Shannon entropy and hybrid entropy, for dealing with conflict of sources. To demonstrate its effectiveness, the approach is applied to segmenting multi-modality MR brain images. It is concluded that the proposed approach performs better than Kmean clustering, majority voting, fuzzy set operators, and Bayesian approach.