A New Technique for K-Means Cluster Centers Initialization of WMH Segmentation

Abstract

K-means algorithm is the most common clustering algorithm being used in medical image processing application. However, the performance of k-means clustering algorithms which converges to numerous local minima would rely on the best initial cluster centers. Generally initial cluster centers are selected randomly and the results are varying on different runs of the algorithm on the same dataset. In this paper, a new method for selecting the best initial centers of k-means clustering is proposed for grouping brain tissues of MRI images. The selection of initial cluster centers namely as Gray Scale Region Intensities (GSRI), is made based on the average intensity value of grayscale region of the images. The proposed method is compared to other method namely as Gray Scale Division Equality (GSDE) which the initial centers were computed by dividing the gray scale 255 and number of clusters. The results show that GSRI outperformed GSDE method in terms of refined segmented regions and converge to local minima with higher iteration number. As a conclusion, it is observed that the newly proposed method has good performance to obtain the initial cluster centers for the k-means algorithm.