DASENet:A Detail Aware U-Net with Shuffle Excitation for Retinal Vessel Segmentation

Retinal vessel images serve as valuable tools for early screening and timely diagnosis of ocular disorders. With the advancement of deep learning, a multitude of segmentation models have been utilized for retinal vessel image segmentation tasks. However, the challenge of enhancing feature expression capabilities and refining the accuracy of retinal blood vessel segmentation persists. In this study, we propose an end-to-end retinal vessel segmentation network named DASENet. To capture finer vessel details and enhance feature expression capabilities, a detail aware block (DAB) is proposed for feature extraction and to heighten the network's focus on vessel information. Moreover, shuffle excitation (SE) within skip connections is proposed to capture contextual dependencies between features and reduce differences between feature maps at different depths. Experimental results on the DRIVE, STARE and CHASE_DB1 datasets demonstrate that our DASENet outperforms other related advanced work.