Retinal Blood Vessel Image Segmentation Method Based on Improved U-Net

The observation of retinal vessel morphology helps to evaluate both ocular and systemic vascular diseases. Therefore, the task of retinal vessel segmentation holds high clinical value and is an important task in medical diagnostic techniques. This paper proposes an improved U-Net model that integrates the Enhanced Local Attention (ELA) mechanism and dilated convolution. This combination enhances the understanding of regional features and achieves precise segmentation of retinal vessels. Comparative experiments conducted on the DRIVE dataset with mainstream network models have validated that the improved model outperforms other models.