Segmentation of multiple sclerosis plaques in 3D magnetic resonance brain images using probabilistic anatomical atlases and one-class support vector machines

• Grant number: 15/06702-2
• Support Opportunities: Scholarships in Brazil - Scientific Initiation
• Effective date (Start): September 01, 2015
• Effective date (End): December 31, 2015
• Field of knowledge: Engineering - Biomedical Engineering - Bioengineering
• Principal Investigator: Ricardo José Ferrari
• Grantee: Fábio Henrique Clug
• 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 an inflammatory disease that affects the central nervous system and causes the destruction of myelin, which is a fundamental protein in nerve impulse transmission. The demyelinated areas get an appearance of plaques, which subsequently become sclerotic (harden). These lesions (or plaques) cause signals and intermittent neurological symptoms that can progressively aggravate with the evolution of the disease. Currently, exams using magnetic resonance imaging (MRI) are used for the confirmation of the diagnostic and monitoring of the disease progression and its therapy. Studies using MRI support the hypothesis that the cognitive disorders in MS patients have correlation with the lesion load. The traditional method to measure the volume of the MS lesions (lesion load) is the manual delineation of lesions in 3D MR images. This procedure, in addition to consuming a long time from the radiologist, is prone to a wide variability within and between observers. In this context, this scientific research project aims to develop an automatic computational technique for segmentation of MS plaques in 3D MR images. The approach to be investigated will use probabilistic anatomical atlas to assist in the selection of samples of Gray Matter (GM) and Cerebral Spinal Fluid (CSF), which will then be used to train a Support Vector Machine to a class (GM + CSF = "positive" class). The voxels belonging to the second class, corresponding to White Matter(WM) and lesions (WM + lesions = "negative" class), will be processed by an outliers detector for the separation of the lesions. The developed technique will be analyzed quantitatively using MICCAI-2015 database images.