Diferenciação do padrão de malignidade e benignidade de massas em imagens de mamografias usando padrões locais binários, geoestatística e índice de diversidade

• Ano de defesa: 2014
• Autor(a) principal: ROCHA, Simara Vieira da
• Orientador(a): PAIVA, Anselmo Cardoso 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: Reconhecimento de Padrões; Padrões Locais Binários; Geoestatística; Índice de Diversidade; Máquina de Vetores de Suporte; Câncer de Mama
• Palavras-chave em Inglês: Pattern Recognition; Local Binary Pattern; Geostatistics; Diversity Index; Support Vector Machine; Breast Cancer
• Área do conhecimento CNPq: Processamento Gráfico; Ciência da Computação
• Link de acesso: http://tedebc.ufma.br:8080/jspui/handle/tede/1822

Breast cancer is the second most frequent type of cancer in the world, being more common among women, and representing 22% of the new cases every year. A precocious diagnosis improves the chances of a successful treatment. Mammography is one of the best ways to precocious detection of non-palpable tumor that could lead to a breast cancer. However, it is well known that this exam's sensibility may vary a lot. This is due to factors such as: the specialist's experience, patient's age and the quality of the exam image. The use of Image Processing and Machine Learning techniques has becoming a strong contribution to the specialist diagnosis task. Thes thesis proposes a methodology to discriminate patterns of malignancy and benignity of masses in mammographic images using texture analysis and machine learning. For this purpose, the methodology combines structural and statistical approaches for the analysis of texture regions extracted from mammograms. Furthermore, this research extends the concept of Diversity Index through the use of species co-occurrence information in order to increase the efficiency of extraction of texture features. The techniques used are Local Binary Pattern, Ripley's K function and diversity indexes (Shannon, Mcintosh, Simpson, Gleason and Menhinick indexes). The extracted texture is classified using a Support Vector Machine into benign and malignant classes. The best results obrained with Ripley's K function were 92,20% of accuracy, 92,96% of sensibility, 91,26% of specificity, 10.63 of likelihood positive ratio, 0,07 of likelihood negative ratio and an area under ROC curve Az of 0,92.