Abstract

Info

Title:: CAMU-Net: an improved model for retinal vessel segmentation based on Attention U-Net

Author(s):: TANG Yunfei1; 2; DAN Zhiping1; 2; HONG Zhengtian1; 2; CHEN Yonglin1; CHENG Peilin3; CHENG Guo4; LIU Fangting5; 1. Hubei Key Laboratory of Intelligent Vision Based Monitoring for Hydroelectric Engineering, College of Computer and Information Technology, China Three Gorges University, Yichang 443002, China 2. Yichang Key Laboratory of Intelligent Medicine, College of Computer and Information Technology, China Three Gorges University, Yichang 430002, China 3.Tongji Hospital, Tongji Medical College, Huazhong University of Science and Technology, Wuhan 430030, China 4. Tsinghua Shenzhen International Graduate School, Shenzhen 518055, China 5. Hongan County Peoples Hospital, Huanggang 438400, China

Keywords:: Keywords: retinal vessel image segmentation deep learning CAMU-Net attention mechanism

Abstract:: An improved U-Net model (channel attention module U-Net, CAMU-Net) is proposed to achieve precise segmentation of retinal vessels. CAMU-Net model enhances its understanding of regional features by employing residual enhancement convolution to extract important information from the regions, improves the global feature acquisition capability by introducing feature refinement module to promote feature extraction, realizes precise segmentation by adding channel attention module to capture image features accurately, and enhances its capability to perceive target boundaries and details through a multi-scale feature fusion structure. The ablation study on the DRIVE dataset validates the role of each module in retinal vessel segmentation. The comparison with other mainstream network models on DRIVE and STARE datasets verify that CAMU-Net model is superior to other models.