Diagnóstico de câncer de mama a partir de imagens de mamografia 2d utilizando descritores de forma 3d

Detalhes bibliográficos
Ano de defesa: 2018
Autor(a) principal: SOUZA, Johnatan Carvalho lattes
Orientador(a): PAIVA, Anselmo Cardoso de lattes
Banca de defesa: PAIVA, Anselmo Cardoso de lattes, SILVA, Aristófanes Corrêa lattes, SANTANA, Ewaldo Eder Carvalho lattes, CARVALHO FILHO, Antonio Oseas de lattes, ROCHA, Simara Vieira da lattes
Tipo de documento: Dissertação
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:
Palavras-chave em Inglês:
Área do conhecimento CNPq:
Link de acesso: https://tedebc.ufma.br/jspui/handle/tede/2140
Resumo: Breast cancer is the second major cause of death by cancer in the female population and the fifth leading cause of death from cancer overall. However, it is known that breast cancer has a better prognosis and higher chances of cure if diagnosed at early stages. Therefore, an early detection is extremely important and the more information is available to the expert, the greater the chances of a correct diagnosis. This scenario justifies the need for the development of computational techniques to support the early detection of breast cancer. Therefore, the purpose of this work is to present a method for breast cancer classification from mammography images using shape analysis and pattern recognition techniques. For that, it is investigated the use of the descriptors Relief Index, Average Slope, Section Area, Section Convolution, D1dist, D2dist and D3dist. These shape descriptors are not traditional in the context of medical images analysis. They are so called \dental" shape descriptors, which have been used in dental ecology as ecometrics, characteristics of organisms that reflect a species’ ecology, to analyze dental shape of mammals and reconstruct past environments. Several experiments with combinations of descriptors are performed, producing several feature vectors. Then, these vectors are submitted to the support vector machine classifier. The proposed method revealed promising results. The best result, on average, was 92.58% accuracy, 92.80% sensitivity and 92.28% specificity