Automatic Segmentation of Multiple Sclerosis Lesions in Brain MR Images

Abstract

Magnetic resonance (MR) imaging is one of the most important tools for multiple sclerosis (MS) diagnosis. However, detection and segmentation of MS lesion in MR images is challenging. Variability in lesion location, size, shape, and anatomical variability between subjects are some factors that cause accurate identification of MS lesions in MR images extremely difficult. On the other hand, manual segmentation is time-consuming. Furthermore, it is subject to observer variability. Several methods have been proposed to automatically segment MS lesions. They have been classified as supervised and unsupervised approaches. In this paper, we used both strategies, using combination of hidden Markov random field (HMRF), k-nearest neighbors (KNN) and support vector machine (SVM) algorithms. The performance of proposed approach is quantitatively evaluated on 20 MS patients that have provided by MS lesion segmentation grand challenge dataset (MICCAI 2008). The average value of dice coefficient percentage (80.03%) and Positive Predictive Value (0.7661) are computed by spatially comparing the results of present procedure with expert manual segmentation. The results showed acceptable performance for the proposed approach, compared to those of previous work.