Brain MR image segmentation using fuzzy entropy-based clustering algorithm

Abstract

In this thesis work, we proposed an image segmentation method that is based on fuzzy c-means algorithm. It has been modified by Shannon entropy by which we can segment a noisy 2D brain MR image which also contains high Intensity Inhomogeneity (IIH). We fail to achieve comparable segmentation accuracy using standard Fuzzy Clustering Algorithm (FCM). As the segmentation is not so perfect, it becomes hard to find the abnormality or the diseases in the tissue areas in brain. To reduce this problem, we considered many algorithms that are based on standard FCM algorithm. Among them the entropy based clustering algorithms reduces the uncertainty of a pixel being in a cluster. Those algorithms perform well in some scenarios but appears to fail in majority of cases. In this thesis, we tried to improve the performance of a particular entropy based clustering algorithm. A very detailed comparative analysis has been made with different cases also including a contrast with previous algorithms. The new segmentation method was applied to BrainWeb brain MR database at varying noise and inhomogeneity. It gives us a better segmentation result when evaluated in terms of Dice Similarity Coefficient, Partition Coefficient, Partition Entropy, Segmentation Accuracy, Tissue Segmentation Accuracy.