Brain tumour grading in different MRI protocols using SVM on statistical features

Abstract

In this paper a feasibility study of brain MRI data set classification, using ROIs which have been segmented either manually or throug h a superpixel based method in conjunction with statistical pattern recognition me thods is presented. In our study, 471 extracted ROIs from 21 Brain MRI datasets are u sed, in order to establish which features distinguish better between three grading c lasses. Thirty-eight statistical measurements were collected from the ROIs. We found by using the Leave-One-Out method that the combination of the features from th e 1 st and 2 nd order statistics, achieved high classification accuracy in pair-wise grading comparisons.