Aplicação de modelos de deep learning para a detecção do câncer de pulmão

• Ano de defesa: 2024
• Autor(a) principal: Atoche Galarreta, Andrés Iván [UNESP]
• Orientador(a): Lotufo, Anna Diva Plasencia [UNESP]
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Estadual Paulista (Unesp)
• Palavras-chave em Português: Câncer de pulmão; Aprendizado de transferência; Aprendizado profundo; Imagens médicas
• Palavras-chave em Inglês: Deep learning; Lung cancer; Transfer learning; Inception; Xception
• Link de acesso: https://hdl.handle.net/11449/255701

In Latin America, in recent years, lung cancer has emerged as one of the leading causes of mortality. According to estimates from 2023, in Brazil, this type of cancer ranks third in incidence among men and fourth among women. This issue is exacerbated due to limited access to healthcare services, especially among individuals living in poverty, who are disproportionately affected. This landscape presents a significant challenge for our society.In order to address this issue, this study focused on evaluating the performance of five different convolutional network models applied to cancer diagnosis using medical imaging. The aim was to determine which models respond best to lung cancer diagnosis. To achieve this, a dataset consisting of lung computed tomography scans classified into three states—malignant, benign, and normal—was utilized. Employing transfer learning, these five models were trained and evaluated under different scenarios.The results obtained revealed that the InceptionV3 model demonstrated the best performance in terms of accuracy. On the other hand, the Xception model achieved an accuracy exceeding 90% in six out of the eight evaluated scenarios.