Multiple sclerosis detection from brain MRI images using least square-SVM

Abstract

Multiple sclerosis (MSC) is a disease that impairs brain-to-body communication by causing the immune system to attack the protective sheath that surrounds nerve fibers. Early detection of the disease is very important so that proper medication helps an individual to recover. In this work, a framework based on LS-SVM has been proposed to detect multiple sclerosis from brain Magnetic Resonance Images (MRI). The proposed framework is categorized into four stages: named as pre-processing, extraction of the region of interest, segmentation of multiple sclerosis using LS-SVM, and, evaluation of results. During MRI image procurement the image is contaminated by the noise from scanners which is successfully eliminated in pre-processing stage by contrast enhancement and histogram equalization. The region of interest (ROI) is extracted from its skull by morphological operations followed by Gaussian filtering and adaptive histogram equalization. Multiple sclerosis is detected from ROI images by using LS-SVM-based segmentation. Finally, results are evaluated on various evaluation metrics such as accuracy, sensitivity, specificity, Dice and Jaccard. An overall accuracy of 96.39% has been achieved in multiple sclerosis extraction from brain MRI images.