An optimal Kernelized Fuzzy C-Means for Automated Segmentation of Breast MRIs

Medical image segmentation is an indispensable process in screening and determining various structures in the breast Magnetic Resonance Images. Although Fuzzy C-Means method has proven to have high capacity of segmenting the medical images, it yet faces some challenges such as noise sensitivity, computational complexity, and etc. Moreover, random initialization of cluster centers can let the clustering process easily fall onto the local minimum, leading to accuracy degradation in image segmentation. To mitigate the above issues, this paper introduces an optimal Fuzzy C-Means method based on minimal spanning tree. The proposed method adopts a robust initialization which automatically decides the number of clusters and initial cluster centers from the given dataset. This improves the segmentation performance significantly. In addition, by deciding the window size of pixel neighbor and the weights of neighbor memberships, the proposed approach adaptively incorporates spatial information to the clustering process and increases the algorithm robustness to noise pixels. To estimate the performance of the proposed method, experimental work is executed on synthetic image, and real breast MRIs. The proposed method is validated by comparing the results with that of the existing methods in the various cluster validity functions.