An Application of Multi-Level Global Attention Framework for Retinal Vessel Image Segmentation

Accurate segmentation of retinal vessels is vital for clinical diagnosis, yet challenges like complex vascular structures, noise, and low contrast persist. While deep learning has enhanced segmentation through contextual feature extraction, issues such as information loss and limited detail representation remain. To this end, we introduce an application of Dense Attention U-Net model with a multi-level global attention mechanism for retinal vessel image segmentation. Experiments on three public datasets show our model outperforms existing methods, achieving F1 scores over 82% and demonstrating its effectiveness in retinal vessel segmentation.