An Explainable Intelligent Vision System for Diabetic Eye Disease Assessment

Diabetic retinopathy is a leading cause of preventable blindness, requiring early detection and timely intervention. Existing CNN-based solutions emphasize accuracy but often lack interpretability, efficiency, and clinical usability. This study proposes an Explainable AI–based Clinical Decision Support System using lightweight CNNs (Custom CNN) for fundus image analysis, optimized for resource-constrained environments. The framework integrates risk stratification to classify patients into Normal, Mild, and Severe stages, aiding treatment prioritization. Grad-CAM–based heatmaps enhance transparency by highlighting critical regions influencing predictions, fostering clinician trust. Deployed as a web application, the system supports secure authentication, image upload, automated screening, explainability visualization, and diagnostic report generation. Experimental results show competitive accuracy with improved usability, bridging the gap between research-focused AI models and real-world clinical practice.