Feature Context Aggregation Network with Edge Enhance for Endoscopic Gastrointestinal Bleeding Images Segmentation

Gastrointestinal (GI) bleeding is one of the most common abnormalities in the GI tract. Accurate segmentation of GI bleeding regions is helpful to identify a variety of GI diseases such as ulcers, polyps, tumors, and Crohns disease, which is of great importance to assist doctors in precise diagnosis. However, the interference factors, such as air bubbles and secretions in the bleeding region may cause the problems of hole and fuzzy edge in the segmentation results. To solve these two problems and improve segmentation accuracy, a new deep learning method is proposed to extract the context information and the edge information of GI bleeding regions by using Context Information Aggregation Module (CIAM), Feature Attention Module (FAM) and Edge Enhancement Module (EEM). The proposed method is evaluated on a public dataset and achieves 86.069% in mean intersection over union (IoU), which shows better performance than the most advanced GI bleeding segmentation methods.