SE-Half-UNet: Accurate and Low-Cost Retinal Vessel Segmentation from Fundus Images

Accurate retinal vessel segmentation from eye fundus images is a major challenge in automated methods. U-Net has been widely used in medical image segmentation, and its effectiveness has led researchers to propose various U-Net variants. However, most of these variants are resource-intensive, suggesting that further development is needed. This study introduces a novel approach called SE-Half-UNet, which derives from the Half-UNet framework by incorporating the Squeeze-and-Excitation (SE) Module to bolster the network’s representative capacity by emphasizing channel relationships within feature maps. Our method achieves superior performance in retinal vessel segmentation, with a high F1-Score of 0.8303 and 0.8372 for DRIVE and CHASE_DB datasets, respectively. This new method facilitates the accurate analysis of eye images and encourages computer research advances.