Classificação de imagens histopatológicas de câncer de pênis usando redes convolucionais e transferência de aprendizagem

• Ano de defesa: 2022
• Autor(a) principal: LAUANDE, Marcos Gabriel Mendes
• Orientador(a): BRAZ JÚNIOR, Geraldo
• Banca de defesa: BRAZ JÚNIOR, Geraldo , ALMEIDA, João Dallyson Sousa de , GOMES JUNIOR, Daniel Lima , OLIVEIRA, Rui Miguel Gil da Costa , MARQUES, Fátima de Lourdes dos Santos Nunes
• 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 CIÊNCIA DA COMPUTAÇÃO/CCET
• Departamento: DEPARTAMENTO DE INFORMÁTICA/CCET
• País: Brasil
• Palavras-chave em Português: câncer de pênis; imagens histopatológicas; redes neurais convolucionais; Densenet; transferência de aprendizado.
• Palavras-chave em Inglês: penile cancer; histopathological images; convolutional neural network; Densenet; transfer learning.
• Área do conhecimento CNPq: Ciência da Computação
• Link de acesso: https://tedebc.ufma.br/jspui/handle/tede/3842

Penile cancer is a rare tumor that accounts for 2% of cancer cases in men in Brazil, with Maranhão being one of the states with the highest incidence. Histopathological analyzes are commonly used in its diagnosis, making it possible to assess the degree of the disease, its evolution, and its nature. For approximately a decade, scientific works in Deep Learning have been developed to help pathologists make decisions quickly and reliably, opening up possibilities for new contributions to improve such a complex and time-consuming activity for these professionals. In this work, the development of a method based on deep learning and transfer of learning is presented. These concepts were applied using a DenseNet-type convolutional network to diagnose penile cancer in a dataset of histopathological with 194 images built by the Legal Amazon Penis Cancer Project. In addition, another method based on deep features was evaluated for comparative purposes. The Grad-CAM algorithm was also used to explain the generated model to discover which regions of the images were relevant during the classification process. In the experiments performed to differentiate normal histopathological images and those with squamous cell carcinoma, an F1-Score of up to 97.39% and a sensitivity of up to 98.33% was achieved.