Automatic computational scheme for segmentation, volumetric assessment and analysis of multiple sclerosis lesions in magnetic resonance images of the human brain

• Ano de defesa: 2019
• Autor(a) principal: Freire, Paulo Guilherme de Lima
• Orientador(a): Ferrari, Ricardo José
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Universidade Federal de São Carlos; Câmpus São Carlos
• Programa de Pós-Graduação: Programa de Pós-Graduação em Ciência da Computação - PPGCC
• Palavras-chave em Português: Esclerose múltipla; Segmentação; Classificação; Textura; Ddistribuição t-Student
• Palavras-chave em Inglês: Multiple sclerosis; Segmentation; Classification; Texture; Student's t-distribution
• Área do conhecimento CNPq: CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO::METODOLOGIA E TECNICAS DA COMPUTACAO
• Link de acesso: https://repositorio.ufscar.br/handle/20.500.14289/12106

Multiple Sclerosis (MS) is an inflammatory demyelinating (that is, with myelin loss) disease of the Central Nervous System (CNS). It is considered an autoimmune disease in which the immune system wrongly recognizes the myelin sheath of the CNS as an external element and attacks it, resulting in inflammation and scarring (sclerosis) of multiple areas of CNS’s white matter. Multi-contrast magnetic resonance imaging (MRI) has been successfully used in diagnosing and monitoring MS due to its excellent properties such as high resolution and good differentiation between soft tissues. In this context, from a computational standpoint, two important goals stand out: lesion segmentation and lesion classification, the latter being related to the identification of which lesions are under an inflammatory state, also called active or enhancing lesions. Nowadays, the preferred method to segment MS lesions is manual delineation, made by specialists with limited aid of a computer. However, this approach is tiresome, expensive and prone to error due to inter- and intra-variability between observers caused by low contrast on lesion edges. Here, we propose the development of an automatic computational technique based on Student’s t-distribution finite mixture models and probabilistic atlases to segment and measure MS lesions volumes in MR images. Regarding the identification of enhacing lesions, Gadolinium-based contrasts are used to visually highlight them during an MRI procedure. However, recent studies indicate that patients gradually lose their ability to eliminate the contrast substances from their bodies when they undergo many contrast injections throughout their lives, which is the case for MS subjects. In this sense, in this work we used textural features to distinguish enhancing (active) and nonenhancing lesions without the aid of intravenous injection of Gadolinium-based contrast, thus eliminating the risk of accumulation of this substance in one’s body and making the MRI procedure faster and cheaper.