Automated Optic Disc and Cup Segmentation Using Mask R-CNN for Glaucoma Assessment

Glaucoma, a leading cause of irreversible blindness, requires precise segmentation of the optic disc and optic cup in fundus images for early diagnosis and progression monitoring. This study presents an automated deep-learning-based segmentation approach using Mask R-CNN, trained on the G1020 dataset, which consists of 1,020 fundus images with corresponding labeled masks. The dataset was split into an 80% training set (816 images) and a 20% test set (204 images). The model was finetuned with a customized predictor for multiclass segmentation to effectively detect the optic disc and cup regions. Training was conducted using a stochastic gradient descent optimizer with a learning rate scheduler to enhance convergence. The proposed model achieved an average Dice coefficient of 0.9381 for the optic disc and 0.8372 for the optic cup, with corresponding Intersection over Union (IoU) scores of 0.8887 and 0.7319, demonstrating robustness in identifying these critical anatomical structures. The segmentation results show significant potential for clinical applications, aiding ophthalmologists in automated glaucoma assessment. Future work will focus on improving the model's generalizability to diverse datasets and integrating it into real-time clinical decision-making systems.