Abstract

Info

Title:: Gastric tumor segmentation by U-Net based on reverse attention mechanism

Author(s):: WANG Ping1; XU Kaicheng2; ZHANG Yichi2; WANG Hailing2; CAI Qingping3; WEI Ziran3; HU Zunqi3; 1. School of Continuing Education, Shanghai University of Engineering Science, Shanghai 201620, China 2. School of Electronic and Electrical Engineering, Shanghai University of Engineering Science, Shanghai 201620, China 3. Department of Gastrointestinal Surgery, Shanghai Changzheng Hospital, Shanghai 200003, China

Keywords:: Keywords: gastric tumor segmentation deep learning image processing reverse attention mechanism U-Net

Abstract:: Abstract: The global and local attention mechanisms are used to localize tumor, and a reverse attention mechanism is added to the model to remove the salient features from the original feature map while retaining the edge contour information. In addition, deep supervision is also applied to supervise the training of each deep decoding layer, which effectively suppresses gradient disappearance and enhances segmentation accuracy. The gastric CT data set used in the study is from Shanghai Changzheng Hospital. The performance of U-Net model with reverse attention mechanism in gastric tumor segmentation has been greatly improved when compared with the traditional U-Net networks (U-net, Attention U-net and ET-Net), which proves the effectiveness of the proposed model.