GWVSeg-Net: An Efficient Method for Gastrointestinal Wall Vascular Segmentation

Abstract

Precisely and automatically segment the blood vessels in the gastrointestinal wall and analyze their distribution state, which is of great significance to reduce or even avoid serious complications such as iatrogenic colonic perforation. In this paper, we propose the novel gastrointestinal wall vascular segmentation network (GWVSeg-Net) to capture a wider range of semantic features and improve the ability of inter-class recognition and intra-class aggregation by using the global pyramid attention module (GPA). In addition, in order to improve the ability of the model to accurately distinguish between mucosal folds and vessels, a new loss function is proposed to train the model. Experimental results show that the proposed method is superior to the existing advanced segmentation networks in the performance of gastrointestinal wall vascular segmentation.