Pulmonary nodule segmentation in computed tomography with deep learning

• Autor(a) principal: Gomes, João Henriques Oliveira
• Data de Publicação: 2017
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10071/15479
• topic: Computer vision; Medical imaging; Pulmonary nodules; Image segmentation; Deep learning; Ciência da computação; Visão computacional; Tecnologia médica; Cancro; Machine learning

Early detection of lung cancer is essential for treating the disease. Lung nodule segmentation systems can be used together with Computer-Aided Detection (CAD) systems, and help doctors diagnose and manage lung cancer. In this work, we create a lung nodule segmentation system based on deep learning. Deep learning is a sub-field of machine learning responsible for state-of-the-art results in several segmentation datasets such as the PASCAL VOC 2012. Our model is a modified 3D U-Net, trained on the LIDC-IDRI dataset, using the intersection over union (IOU) loss function. We show our model works for multiple types of lung nodules. Our model achieves state-of-the-art performance on the LIDC test set, using nodules annotated by at least 3 radiologists and with a consensus truth of 50%.