Swinsegnet: A Transformer-Driven Multi-Task Model for Robust Glaucoma Screening Across Public Datasets

Glaucoma is a leading cause of irreversible blindness, where early and accurate detection is critical for effective treatment. Traditional deep learning models for glaucoma screening often struggle with generalization across multiple datasets due to variations in imaging quality and annotation. To address this, we propose SwinSegNet, a transformer-driven multi-task model integrating Swin Transformer with SegFormer MLP to jointly perform optic disc/optic cup segmentation and glaucoma classification. The model was trained and evaluated on three benchmark public datasets ORIGA, DRISTI-GS, and REFUGE encompassing 1,951 retinal fundus images. Experimental results show that SwinSegNet outperforms existing CNN-based and U-Net variants, achieving an average accuracy of 99.49%, sensitivity of 99.24%, and specificity of 99.40%, marking an improvement of up to 4%–5% over state-of-the-art methods. These findings confirm the robustness and clinical potential of SwinSegNet for large-scale glaucoma screening.