Brain MR Segmentation using a Fusion of K-Means and Spatial Fuzzy C-Means

Abstract

This study presents an automatic segmentation of the brain tissues in Magnetic Resonance Image using a fusion of Spatial Fuzzy C-Means (sFCM) and K-Means Algorithms (sFCMKA). The segmentation of the standard FCM algorithm have been realized to be highly sensitive to noise and therefore fail in providing accurate results. The K-Means algorithm while having the advantage of simplified clustering results is disadvantaged for merely classifying pixels into just one cluster segment. In overcoming this problem, first, FCM algorithm incorporated with a spatial neighborhood function is presented to eliminate the effect of noise. Also, the K-Means algorithm is embedded to sort pixels in tissues which are not intertwined with another. Since they are nonoverlapping tissues, such pixels are grouped in a singular cluster without having membership to another cluster segment. sFCMKA achieves this by placing a threshold parameter on pixels having contrasting cluster result from both algorithms and then reprocessed. The efficiency of the proposed algorithm is demonstrated by evaluating the results with the input image’s manual segmentation using the Jaccard similarity index, Dice coefficient and accuracy index. The results of the proposed algorithm depicts a far more admirable accuracy quality compared to existing clustering based segmentation methods.