3D Brain MR Image Segmentation using a Fuzzy Entropy-Based Fuzzy Clustering Algorithm

This is the authors' abstract. We don't add key points for this paper.

This paper presents a fuzzy clustering algorithm by incorporating fuzzy entropy for segmentation of 3D brain magnetic resonance (MR) images. Intensity inhomogeneity (IIH) and noise strongly affect brain MR images because of improper scanning environment or patient's unconscious movements at the time of image acquisition. The occurrence of this phenomenon makes the soft tissue regions blurry, specifically at the boundary region and imposes uncertainty for clustering the soft tissue regions. The presented algorithm utilizes fuzzy entropy, which associates with the local membership functions to mitigate the underlying uncertainty for voxel classification. It integrates fuzzy entropy with the membership functions of local neighbouring regions of the voxel. The objective function conjugates global membership functions with the distance of the cluster center and the voxel along with a controlling factor as a first term. The multiplicative factor of the local membership function, the mean distance of neighboring voxels from the cluster center, possibility parameter and a complemented controlling factor is added as the second term. Finally, the third term is the fuzzy entropy defined with the local membership function. The suggested algorithm has tested in a qualitative and quantitative manner by utilizing 6 simulated image volumes and 4 real patient volumetric brain MR images. The results of the experiments demonstrate outperforms other competitive algorithms developed recently. The dice score similarity of the proposed method by using simulated image volume 77.26%. The outcome of real patient image volume is measured by partition entropy and partition coefficient not by dice similarity coefficient as for real image ground truth images are not available.