Leveraging Segformer for Precision Segmentation of Optic Cup and Optic Disc in Fundus Imaging for Glaucoma Detection

Glaucoma is a leading cause of irreversible blindness worldwide, primarily diagnosed through the analysis of retinal fundus images by segmenting the optic cup (OC) and optic disc (OD) and calculating the cup-to-disc ratio (CDR). This study evaluates the performance of Segformer, a Vision Transformer-based model, for semantic segmentation of OC and OD in retinal fundus images using the diverse SMDG-19 dataset. The model achieved Dice Scores of 0.9451 for OC and 0.9026 for OD with a training time of just 7 hours, outperforming several state-of-the-art models. In particular, Segformer demonstrated its ability to generalize across diverse datasets without requiring data augmentation or extensive computational resources, thus providing efficient and precise segmentation results. These findings highlight the potential of Segformer as a robust tool for early detection of glaucoma, paving the way for rapid and accurate diagnosis in clinical settings. Future research could explore its adaptability to additional datasets and further optimization for clinical implementation.