A NOVEL METHOD FOR AUTOMATIC CLASSIFICATION OF MULTIPLE SCLEROSIS LESION SUBTYPES USING DIFFUSION TENSOR MR IMAGES

Abstract

Purpose: An important work in making the treatment plan and following the therapeutic effects of multiple sclerosis (MS) patients is quantification of MS lesion subtypes. This study was designed to propose a new automatic method for classification of MS lesion subtypes based on conventional magnetic resonance (c-MR) and diffusion tensor MR (DT-MR) images. Method: In this paper, fuzzy [Formula: see text]-nearest neighbor (F-KNN) algorithm is used for classifying MS lesions into three subtypes including black holes (chronic), enhancing lesions (acute) and T2-lesions. In order to achieve this classification, F-KNN algorithm is used with features of c-MR images including T1-weighted (T1-w), fluid attenuated inversion recovery (FLAIR), gadolinium enhanced T1-weighted (Gad-E-T1-w) and T2-weighted (T2-w) and then with added features extracted from DT-MR images (diffusion indices) including mean diffusivity (MD) and fractional anisotropy (FA). Results: Evaluation of the proposed method was performed by manual classification of chronic, acute and T2-w lesions using the criterion of sensitivity (Se), specificity (Sp) and accuracy (Acc). The results showed that application of the c-MR with the DT-MR images makes possible the classification of MS lesion subtypes with high sensitivity and allows ability to evaluate MS disease in the treatment process.