Novel seed selection techniques for MR brain image segmentation using graph cut

Abstract

ABSTRACT Brain tumour segmentation is a challenging task because of high diversity of the tumour for different patients and the ambiguity in its location. The high accuracy of extracting the tumour region from the normal brain solely depends on the seed selection. This prior knowledge of the seed points is also required for the initialisation of the graph cut segmentation. In this paper, two techniques (Centroid-Based Seed Selection [CBSS] and k-Mean Seed Selection [KMSS]) for the seed selection and segmentation of the tumour region using the graph-cut technique are proposed. The intensity distribution in the MRI brain images is utilised for the calculation of these points. In (CBSS), the symmetrical intensity distribution on both half of the brain is exploited. While in (KMSS) different groups of the similar intensity distribution are formed. The seed values obtained initialise the Graph-cut method to perform the segmentation. The performance parameters for the proposed methods are evaluated using MR brain images from standard and real time dataset. The mean Dice Sensitivity Coefficient (DSC) and mean Jaccard Index (JI) values are evaluated for indicating effectiveness and accuracy of the proposed work.