A hybrid deep learning model for enhanced multi-class ocular disease classification using EfficientNetB0 and ResNet50

Retinal images are essential for the diagnosis of ocular diseases, which, if left untreated, can lead to permanent vision loss. This study introduces a novel hybrid deep learning model that combines EfficientNetB0 and ResNet50 for the multi-class classification of ocular diseases using the ODIR-5K dataset. Targeting five key categories—age-related macular degeneration (ARMD), cataract, diabetic retinopathy (DR), glaucoma, and normal, the model leverages complementary CNN architectures and an advanced feature fusion strategy to enhance diagnostic precision. A key innovation of our work lies in its multimodal design, which integrates bilateral retinal images with patient demographic data (age and sex) to enrich feature representation and improve predictive robustness. The proposed approach achieves a test accuracy of 98.51%, significantly outperforming baseline models in precision, recall, and F1-score. These results highlight the strong potential of hybrid, multimodal deep learning models for advancing early detection and classification of ocular diseases, with practical implications for real-time clinical applications.