A Hybrid Ensemble Deep Learning Framework for Robust Multi-Retinal Disease Classification with Explainable AI

Age-related macular degeneration (AMD), glaucoma, and diabetic retinopathy are the retinal diseases that cause the most visual impairment worldwide. A fast and accurate diagnosis is essential to preventing cases of irreversible vision loss. Recently, CNNs have shown a lot of promise for automated detection of retinal diseases using fundus pictures. However, the particular CNN frameworks that are now in use frequently have poor generalization and uneven dataset performance. To overcome these limitations, this paper suggests an ensemble deep learning architecture for precise multi-retinal disease categorization. Numerous CNN architectures are integrated by the system, which also assesses intricate ensemble techniques like bagging, stacking, and soft voting. Explainable artificial intelligence (XAI) techniques like Grad-CAM and SHAP are also employed to improve clinical interpretability. Experimental evaluation on retinal datasets shows that ensemble models significantly outperform individual CNN architectures in classification performance. Additionally, lightweight ensemble versions are appropriate for implementation in healthcare settings with limited resources since they preserve competitive accuracy while lowering computing costs. The findings support the hypothesis put out and show that explainable AI in conjunction with ensemble deep learning offers a dependable and comprehensible method for screening for retinal diseases.