A Hybrid Approach of Image Segmentation for Medical MR Images Using SMKFCM and Genetic Algorithm

Abstract

Segmentation of medical image is a major tool for analyzing and solving accurately a broad scope of difficulty in medical magnetic resonance (MR) imaging technique. Unsupervised image segmentation approach based on evolutionary approach of grayscale MR image automatically segments images into different essential parts. The proposed algorithm focuses on to segment MR images precisely by employing intensity statistics with relationships of neighborhood. Spatial multiple kernel FCM (SMKFCM) algorithm assists in population generation of genetic algorithm (GA), that segments automatically MR images. This approach is very powerful for segmentation of MR images that works with spatial statistics of multiple and single feature data. The SMKFCM clustering algorithm is better suitable in absence of noise and it considers only pixel attributes, not its neighbors. In image segmentation problem, accuracy degradation can be achieved by using genetic algorithm. Thus, the global minima can be reached by using the clustering objective function. The proposed algorithm has 93%, 92%, and 91% accuracy at 6% noise level to segment medical MR images for the cluster white matter (WM), gray matter (GM), and cerebrospinal fluid (CSF), respectively, for the MR images.