A Deep Learning Approach for Red Lesions Detection in Video Capsule Endoscopies

• Autor(a) principal: Coelho, Paulo
• Data de Publicação: 2018
• Outros Autores: Pereira, Ana, Leite, Argentina, Salgado, Marta, Cunha, António
• Tipo de documento: Artigo
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10400.8/3645
• topic: Lesion detection; Gastrointestinal bleeding; Machine learning; Capsule endoscopy; Deep learning; U-Net; Computer Science; Image Analysis

The wireless capsule endoscopy has revolutionized early diagnosis of small bowel diseases. However, a single examination has up to 10 h of video and requires between 30–120 min to read. Computational methods are needed to increase both efficiency and accuracy of the diagnosis. In this paper, an evaluation of deep learning U-Net architecture is presented, to detect and segment red lesions in the small bowel. Its results were compared with those obtained from the literature review. To make the evaluation closer to those used in clinical environments, the U-Net was also evaluated in an annotated sequence by using the Suspected Blood Indicator tool (SBI). Results found that detection and segmentation using U-Net outperformed both the algorithms used in the literature review and the SBI tool.