Edge-AI Powered Transparent Diagnosis of Diabetic Retinopathy Using Deep Learning

Diabetic Retinopathy (DR) stands as the primary reason that causes vision impairment because early detection requires state-of-the- art deep learning technology. The current excellence of Convolutional Neural Networks (CNN)based models in terms of accuracy fails to meet clinical requirements because of their opacity. The solution involves developing an improved Explainable AI (XAI)-driven deep learning framework which detects fundus diseases. The design merges Vision Transformers (ViTs) and Hybrid CNN-ViT models to combine their excellent feature extraction methods with interpretable functionality. The integration of Layer-Wise Relevance Propagation (LRP) and SHAP (SHapley Additive exPlanations) provides clinical staff with an understanding of decision processes for model transparency enhancements. We have integrated multimodal data fusion into The system which merges images from the fundus tissue with medical data records to boost diagnosis quality. The integrated Edge AI deployment component enables fast and efficient DR detection through portable healthcare tools in real time. The training requires Generative Adversarial Networks (GANs) to perform synthetic image augmentation because of data imbalance issues. The research model achieves improved outcomes on benchmark databases while producing higher accuracy and specific results and easier interpretation compared to established CNN-based methods. This research develops a detection method for diabetic retinopathy which provides real-time, transparent and clinical compatibility allowing better trust and medical practice adoption. The framework connects advanced deep learning methods with explainability approaches to breed innovative fundus disease detection systems which offer high performance alongside easy interpretation for actual patient care activities.