Intuitionistic Gustafson-Kessel algorithm for segmentation of MRI brian image

Abstract

Gustafson-Kessel algorithm is well known clustering technique which provides better clusters in comparison to conventional fuzzy c-means clustering of data consisting of different shape, size and densities. It employs a covariance matrix to adapt automatically the local distance metric to the different shape of the cluster and adapting the distance inducing matrix accordingly. A major challenge posed in the segmentation of MRI brain image is to take into account the uncertainty in the final localization of the feature vectors in the term of pixel value. This arises due to imprecise information and noise present in the image. This may lead to improper assignment of membership value during the membership updating process in clustering. In literature, Intuitionistic fuzzy set is proposed which handles it is in better way that arises from imprecise information in ill-defined data. In this paper we propose intuitionistic Gustafson-Kessel (IGK) algorithm which utilizes the benefits of intuitionistic fuzzy set theory to achieve better segmentation of MRI images. Experiments are performed on T1-weighted MRI brain images which are publicly available. Experimental results show that the intuitionistic Gustafson-Kessel algorithm performs significantly better in comparison to Gustafson-Kessel (GK) algorithm. © 2012 Springer India Pvt. Ltd.