Automatic Segmentation of Three Dimensional Brain Magnetic Resonance Images Using Both Local and Non-Local Neighbors with Considering Inner and Outer Class Data Distances

Abstract

In this article a new combined method from genetic and fuzzy c-means algorithm (FCM) for discovering the correct number of segments and automatic segmentation of human normal and abnormal brain MR images is proposed.
For reducing the effect of the noise in segmentation process, we use the local and non-local neighbors, and also a new method for finding the appropriate neighbors of the voxels and adjusting their weights. In addition, by decreasing the distance between the data and data cluster centers, the distances between the data from other cluster centers are increased and caused a better and more precise detection of segments boundaries.
The proposed method was applied to 10 clinical MRI data set images. Our experimental results shows that the presented method has a significant improvement compare to other similar methods.