Enhanced retinal blood vessel segmentation via loss balancing in dense generative adversarial networks with quick attention mechanisms

• Published in: International ophthalmology Vol. 45; no. 1; p. 402
• Main Authors: Sandeep, Daria, Baranitharan, K., Padmavathi, A., Guganathan, Loganathan
• Format: Journal Article
• Language: English
• Published: Dordrecht Springer Netherlands 03.10.2025; Springer Nature B.V
• Subjects: Accuracy; Algorithms; Attention; Automation; Blood vessels; Datasets; Diabetes; Diabetes mellitus; Disease detection; Efficiency; Flexibility; Fluorescein Angiography - methods; Fundus Oculi; Generative Adversarial Networks; Glaucoma; Graph neural networks; Humans; Hypertension; Image Processing, Computer-Assisted - methods; Machine learning; Medicine; Medicine & Public Health; Methods; Neural networks; Neural Networks, Computer; Noise reduction; Occlusion; Ophthalmology; Optimization; Original Paper; Retina; Retinal Diseases - diagnosis; Retinal Vessels - diagnostic imaging; Retinopathy; Robustness; Segmentation; Semantics; Training; Veins & arteries; Retinal vessel segmentation; Quasi-cross bilateral filtering; Medical imaging; Swarm bipolar algorithm; Directed acyclic graph neural network with VGG16; Dense generative adversarial network with a quick attention network
• ISSN: 1573-2630; 0165-5701; 1573-2630
• DOI: 10.1007/s10792-025-03754-5

Purpose Manual segmentation of retinal blood vessels in fundus images has been widely used for detecting vascular occlusion, diabetic retinopathy, and other retinal conditions. However, existing automated methods face challenges in accurately segmenting fine vessels and optimizing loss functions effectively. This study aims to develop an integrated framework that enhances vessel segmentation accuracy and robustness for clinical applications. Methods The proposed pipeline integrates multiple advanced techniques to address the limitations of current approaches. In preprocessing, Quasi-Cross Bilateral Filtering (QCBF) is applied to reduce noise and enhance vessel visibility. Feature extraction is performed using a Directed Acyclic Graph Neural Network with VGG16 (DAGNN-VGG16) for hierarchical and topologically-aware representation learning. Segmentation is achieved using a Dense Generative Adversarial Network with Quick Attention Network (Dense GAN-QAN), which balances loss and emphasizes critical vessel features. To further optimize training convergence, the Swarm Bipolar Algorithm (SBA) is employed for loss minimization. Results The method was evaluated on three benchmark retinal vessel segmentation datasets—CHASE-DB1, STARE, and DRIVE—using sixfold cross-validation. The proposed approach achieved consistently high performance with mean results of accuracy: 99.87%, F1- score: 99.82%, precision: 99.84%, recall: 99.78%, and specificity: 99.87% across all datasets, demonstrating strong generalization and robustness. Conclusion The integrated QCBF–DAGNN-VGG16–Dense GAN-QAN–SBA framework advances the state-of-the-art in retinal vessel segmentation by effectively handling fine vessel structures and ensuring optimized training. Its consistently high performance across multiple datasets highlights its potential for reliable clinical deployment in retinal disease detection and diagnosis.