Malaysian Journal of Mathematical Sciences, January 2018, Vol. 12, No. 1
New Clustering Algorithm for Classification of Brain MRI Regions by using m-Universal Metric Technique
Nezhad, A. D., Beizavi, S., and Hekmatimoghaddam, S. H.
Corresponding Email: dehghannezhad@iust.ac.ir
Received date: 2 May 2017
Accepted date: 31 December 2017
Abstract:
In this paper, we introduce an \(m\)-dimensional \((m > 2)\) distance metric over a given space to define a new universal metric space. An algorithm for segmentation is purposed in this paper that uses universal metric space as generalized \(K\)-means clustering technique to segment brain and to separate the brain tumors in MR brain images. This article analyzes the segmentation method and separate brain tumor region based on the proposed algorithm. On the basis of this algorithm, target parameter \(m\) that is representing the similarity measure in \(K\)-means algorithm, is changed from case \(m=2\) to cases \(m=3\) to \(m=5\). Brain tumor is separated for different values of \(m\). Finally, to evaluate the proposed algorithm. The results of this procedure for 50 patients, (the selected from MRI center of Yazd city) are compared with segmentation that was performed by the radiologist. Based on these criteria, the proposed algorithm showed an accuracy of about 98.91 in case \(m=5\). So the use of this method for processing magnetic resonance images can give acceptable results for separating tumor from normal brain tissue.
Keywords: \(U_m\)-metric, generalized \(K\)-means clustering technique, segment brain
Indexing
