RADCU-Net: residual attention and dual-supervision cascaded U-Net for retinal blood vessel segmentation

Yong Yang, Weiguo Wan, Shuying Huang, Xin Zhong, Xiangkai Kong

Abstract: The automated segmentation of retinal blood vessels plays an important role in the computer aided diagnosis of retinal diseases. In this study, we propose a novel retinal vessel segmentation method based on residual attention and dual-supervision cascaded U-Net (RADCU-Net). Specifically, a residual attention U-Net (RAU-Net), including a residual unit and an attention mechanism, is constructed to improve the feature representation ability by explicitly modelling the interdependency among the channels of the convolutional features. To boost the accuracy of retinal blood vessel segmentation, a cascaded RAU-Net framework is constructed by concatenating two RAU-Nets with the proposed residual attention modules. Moreover, a dual-supervision training strategy is designed to improve the supervision of the cascaded RAU-Net parameter learning by adding an additional balanced cross-entropy loss function in the middle of the cascaded RAU-Net. The results of extensive experiments on the DRIVE and STARE datasets demonstrate that the proposed method achieves better performance compared to state-of-the-art methods. Our method provides a meaningful attempt to improve blood vessel segmentation and can further facilitate the diagnosis of ophthalmological diseases.