A Dilated Feature Fusion Approach to Optic Cup and Disc Segmentation for Glaucoma Detection and Progression Prediction

Glaucoma is a chronic eye disease that can cause irreversible vision loss if not detected early. To improve early detection and disease progression analysis, an advanced deep learning-based glaucoma detection framework is proposed. Fundus images are collected from benchmark datasets and preprocessed using contrast enhancement and median filtering techniques to improve image quality. High-quality images are then segmented using the Trans-Res-UNet++ model to accurately extract the optic cup and disc. For classification and progression prediction, a Dilated Feature Fusion-based Hybrid Deep Learning Network (DFFHDLNet) is introduced. In this approach, the segmented optic cup and disc images are processed separately using DenseNet to extract significant features. These extracted features are then fused and fed into a Gated Recurrent Unit (GRU) for final multi-class classification and disease progression prediction. The proposed model effectively captures essential spatial and temporal patterns, ensuring accurate detection and classification of glaucoma stages. Performance evaluation is conducted by comparing the proposed method with conventional models, demonstrating its superior accuracy and robustness. The integration of feature fusion and deep learning techniques enhances segmentation precision and classification performance, making the framework a promising solution for automated glaucoma detection and monitoring.