Segmentation of multiple sclerosis lesions in magnetic resonance images via texture analysis guided by a map of hyperintensities

• Grant number: 19/23198-7
• Support Opportunities: Scholarships in Brazil - Scientific Initiation
• Effective date (Start): February 01, 2020
• Effective date (End): January 31, 2021
• Field of knowledge: Physical Sciences and Mathematics - Computer Science
• Principal Investigator: Ricardo José Ferrari
• Grantee: Fernanda Carolina Ferreira
• Host Institution: Centro de Ciências Exatas e de Tecnologia (CCET). Universidade Federal de São Carlos (UFSCAR). São Carlos , SP, Brazil

Abstract

Multiple sclerosis (MS) is a multifactorial inflammatory disease that affects the central nervous system (CNS) and mainly affects a young adult population. MS is considered an autoimmune disease because, for some unknown reason, the immune system attacks the myelin sheath that covers neurons, compromising CNS functions. Magnetic resonance imaging (MRI) has been used very successfully in the diagnosis and monitoring of MS, as it allows good differentiation between soft tissues. The lesion load volume in MS, determined on T2-weighted (T2-w) or fluid-attenuated inversion recovery (FLAIR) images, is an important quantitative measure used to assess disease progression. The conventional method of measuring lesion load volume is using the manual delineation of MRI lesions made by a specialist with the help of a computer. This research proposal aims to develop an automatic method for the segmentation of MS lesions in MR images that will use texture attributes, extracted from lesions delineated by an expert in the images, to train a one-class Support Vector Machine (SVM) classifier. The estimated classifier will then be used to classify only voxels located in hyperintense regions of FLAIR images, indicated by a hyperintensity map resulting from a work recently developed by our research group. Such a map contains the MS lesions that are generally hyperintense on FLAIR images and other hyperintense structures that are expected to be discarded via texture classification. The results of the developed method will be compared quantitatively with the markings of two specialists and the results of two other software.