Segmentação de Rins, Cistos e Tumores Renais em Imagens de Tomografia Computadorizadas baseada em Módulos de Pirâmide de Pooling e Redes Convolucionais

• Ano de defesa: 2025
• Autor(a) principal: MATOS, Caio Eduardo Falcão
• Orientador(a): BRAZ JUNIOR, Geraldo
• Banca de defesa: BRAZ JUNIOR, Geraldo , PAIVA, Anselmo Cardoso de , GOMES, Herman Martins , FERNANDES, Leandro Augusto Frata , 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 DOUTORADO EM CIÊNCIA DA COMPUTAÇÃO
• Departamento: COORDENAÇÃO DO CURSO DE CIÊNCIAS DA COMPUTAÇÃO/CCET
• País: Brasil
• Palavras-chave em Português: Câncer de rim; Segmentação de rim; Segmentação de cistos e tumores renais; Redes neurais convolucionais; Aprendizado profundo; Módulos de pirâmide de pooling; Tomografia computadorizada; Imagens médicas
• Palavras-chave em Inglês: Kidney cancer; Kidney segmentation; Kidney cyst and tumor segmentation; Convolutional neural networks; Deep learning; Pyramid Pooling Module; Computed tomography; Medical Images
• Área do conhecimento CNPq: Processamento de Sinais Biológicos
• Link de acesso: https://tedebc.ufma.br/jspui/handle/tede/6542

The early detection and diagnosis of renal cancer play a crucial role in patient prognosis and treatment, significantly increasing the chances of cure and survival. In this context, advances in radiological imaging enable medical specialists to analyze and identify suspicious lesions through imaging techniques. Computed Tomography (CT) has become a widely used tool in aiding the diagnosis of renal cancer due to its ability to generate a large volume of detailed images of the body’s internal structures, including renal cysts and tumors. However, the analysis of these images is manually performed by physicians, who must, review each scan individually, making the process susceptible to fatigue and distraction. This highlights the need for computational techniques to automate and enhance the accuracy of suspicious lesion segmentation, contributing to the early diagnosis of renal cancer. In recent years, Convolutional Neural Networks (CNNs) have gained significant attention for automatic segmentation tasks, providing valuable support to specialists in the detection and diagnosis of medical images. In this research, we developed a convolutional model that leverages the multi-resolution concepts of pooling pyramids (Pyramid Pooling Module - PPM and Atrous Spatial Pyramid Pooling - ASPP) combined with the UNet architecture for the segmentation of kidneys, cysts, and tumors in CT images. The proposed methodology produced promising results on the datasets from the Kidney and Kidney Tumor Segmentation Challenge KiTS21 and KiTS23, achieving Dice scores of 92,32% and 93,78% for kidneys and masses, 91,38% and 93,06% for renal masses, and 87,84% and 91,44% for tumors, respectively.