Metodologia computacional para a segmentação da próstata e classificação de lesões em imagens de ressonância magnética utilizando o modelo de Ising

• Ano de defesa: 2019
• Autor(a) principal: REIS, Artur Bernardo Silva
• Orientador(a): SILVA, Aristófanes Corrêa
• Banca de defesa: SILVA, Aristófanes Corrêa , PAIVA, Anselmo Cardoso de , CONCI, Aura , PACIORNIK, Sidnei , CARVALHO FILHO, Antonio Oseas de
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal do Maranhão
• Programa de Pós-Graduação: PROGRAMA DE PÓS-GRADUAÇÃO EM ENGENHARIA DE ELETRICIDADE/CCET
• Departamento: DEPARTAMENTO DE ENGENHARIA DA ELETRICIDADE/CCET
• País: Brasil
• Palavras-chave em Português: Câncer de próstata; Detecção de lesões da próstata; Ressonância magnética; Modelo de Ising; Algoritmo de Wolff.; Máquinas de vetores suporte
• Palavras-chave em Inglês: Prostate cancer; Abnormal tissue detection in prostate; Magnetic resonance imaging; Ising model; Wolff 's algorithm; Suport vector machine
• Área do conhecimento CNPq: Análise de Algoritmos e Complexidade de Computação
• Link de acesso: https://tedebc.ufma.br/jspui/handle/tede/2627

Prostate cancer is the second most prevalent type of cancer in the male population worldwide. The adoption of prostate imaging tests for the prevention, diagnosis, and treatment has grown. It is known that early detection increases the chances of an effective treatment, improving the prognosis of the disease. With this aim, computational tools have been proposed with the purpose of assisting the specialist in the interpretation of imaging tests, especially magnetic resonance imaging (MRI), providing the detection of lesions. The research of this doctoral work has as primary objective the proposition of an automatic methodology for the detection of lesions in the prostate. We divide the proposed methodology into two stages. In the first stage prostate segmentation is performed, for this purpose, the Ising model is used, models of probability, quality threshold and fusion of atlas labels. The second stage consists of the classification of abnormal tissues in the prostate. To this end, we extract lesion candidates through the Wolff algorithm, then texture characteristics are extracted using the Ising model, and finally, the vector machine is used to classify lesion or healthy tissue. The methodology was validated using three bases of T2-weighted MRI images. We used three bases for prostate segmentation. However, we used only one in prostate segmentation and lesion detection. The total number of images used in the validation of prostate segmentation was 108. The experimental results obtained here indicate an excellent perspective, considering that we obtained a mean Dice similarity coefficient (DSC) of 94.03 % in the step of. We validated The lesion detection stage on a set of 28 images with lesion markers. The methodology obtained a sensitivity of 95:92%, specificity of 93:89% and accuracy of 94:16%. These are promising since they were more significant than other methods compared.