RetinaVisionMapper: A Framework for Accurate and Early Retinal Disease Identification

Retinal diseases such as Diabetic Retinopathy (DR), Glaucoma, and Age-Related Macular Degeneration (AMD) pose severe threats to vision if not identified at early stages. This study presents RetinaVisionMapper, a deep learning based framework which integrates retinal fundus images with physical metadata for precise and early disease detection. The framework relies on a dual-branch designed with a DenseNet-121-based CNN module for fundus imaging and another GRU/MLP stream to handle clinical data. Attention based feature fusion is used to fuse both modalities thus improving cross-modal learning. Advanced preprocessing methods such as CLAHE, non-local means denoising, and vessel enhancement filters are used in order to increase lesion visibility while preserving delicate vascular architecture. Integration of domain-specific data augmentation and Grad-CAM-based interpretability tools is used to achieve robustness and generalizability. The model outperforms 10 baseline algorithms with accuracy 98.72%; precision 98.10%; recall 97.80%; F1-score 97.90% and AUC-ROC 98.95%. These findings suggest that RetinaVisionMapper has therapeutic potential as a clinically viable multi-disease retinal screening tool. The proposed framework provides interpretability, scalability, and precision, suitable for integration into the diagnostic workflow of ophthalmic procedures and in a telemedicine platform for mass screening and early intervention.