AI-driven saliency-guided retinal vessel segmentation framework for sustainable digital pathology

Accurate segmentation of retinal blood vessels is essential for the early diagnosis of ophthalmic and systemic diseases such as diabetes, hypertension, and cardiovascular disorders. However, challenges such as low contrast, complex vessel geometry, and the presence of pathological artifacts often degrade segmentation performance, particularly for thin vessels and boundary regions.

To address these challenges, this study proposes an AI-driven saliency-guided boundary refinement framework (SGB-Net). The model integrates a progressive boundary refinement (BR) module to enhance vessel edge representation and a feature-guided encoder-decoder network incorporating scale-adaptive (SA) and attention enhancement (AE) modules. The SA module captures multi-scale contextual features, while the AE module refines feature representations by emphasizing relevant structures and suppressing background noise. The proposed framework was evaluated on three publicly available datasets: DRIVE, STARE, and CHASE_DB1.

Experimental results demonstrate that the proposed method achieves superior segmentation performance, with Dice scores of 98.30%, 78.40%, and 84.60% on the DRIVE, STARE, and CHASE_DB1 datasets, respectively, and AUC values up to 0.9899. The model shows improved capability in preserving thin vessels, enhancing boundary continuity, and reducing false positives under complex imaging conditions compared to existing state-of-the-art methods.

The proposed SGB-Net effectively addresses key limitations in retinal vessel segmentation by combining boundary refinement with multi-scale and attention-based feature learning. Its robustness to noise and pathological variations makes it suitable for large-scale digital pathology applications and supports more reliable automated retinal analysis. Future work may focus on improving sensitivity and extending the framework to other medical imaging modalities.