Brain MR image segmentation with fuzzy C-means and using additional shape elements

Abstract

Using the intensity of the element in interest, standard FCM generates the membership values to all classes. When used for segmentation of images, this method is not capable of correcting the effects of noise. To overcome that problem, we propose a modification on the standard method. The voxels in the neighborhood are taken into account, forming the shape elements in additon to the intensity of the voxel in interest. The resulting input vector is used with FCM. The proposed method was tested with MR brain image with and without added syntetic noise.