SFAG-Net: A Retinal Vessel Segmentation Network Based on Synergistic Fusion and Adaptive Guidance

Abstract Automatic segmentation of retinal vessels in fundus images provides a crucial basis for the clinical diagnosis of ocular pathologies. However, fundus image analysis presents significant challenges due to the complex vascular topology, low background contrast, and substantial noise interference. Convolutional Neural Networks possess powerful inductive biases, whereas the Transformer excels at capturing long-range dependencies. Leveraging these attributes, this paper proposes a retinal vessel segmentation network based on synergistic fusion and adaptive guidance. The network employs a dual-branch encoder to extract local detail features and global semantic features, respectively. To enable effective complementarity between the two branches, a Synergy Fusion Integrator is constructed using the cross-attention mechanism. Additionally, the Adaptive Feature Guidance Module is designed to enhance multi-scale detail perception. This module dynamically reinforces critical features via a spatial attention mechanism, significantly improving the recognition accuracy of thin vessels and boundary structures. Experimental results demonstrate that the proposed network outperforms current advanced vessel segmentation methods on three public retinal vessel segmentation datasets. It exhibits robust segmentation performance and strong generalization capability, offering reliable technical support for computer-aided clinical diagnosis.