CRAUNet: A cascaded residual attention U-Net for retinal vessel segmentation

Retinal vessels play an important role in judging many eye-related diseases, so accurate segmentation of retinal vessels has become the key to auxiliary diagnosis. In this paper, we present a Cascaded Residual Attention U-Net (CRAUNet) that can be regarded as a set of U-Nets, that allows coarse-to-fine representations. In the CRAUNet, we introduce a DropBlock regularization similar to the frequently-used dropout, which greatly reduces the overfitting problem. In addition, we propose a multi-scale fusion channel attention (MFCA) module to explore helpful information, and then merge this information instead of using a direct skip-connection. Finally, to prove the effectiveness of our method, we conduct extensive experiments on DRIVE and CHASE_DB1 datasets. The proposed CRAUNet achieves area under the receiver operating characteristic curve (AUC) of 0.9830 and 0.9865, respectively, for the two datasets. Compared to other state-of-the-art methods, the experimental results demonstrate that the performance of the proposed method is superior to that of others.