Deep Learning-Driven Retinal Disease Diagnosis with Explainable AI: Development of a Comprehensive Clinical Decision Support Framework

Dr. S. W. Mohod , Rashmi Mahesh Patil , Dr.Rupesh Anna Warbhe

CNN(Convolutional neural network),Eye disease prediction, Deep Learning

Vision-threatening ocular conditions represent one of the most pressing yet underaddressed challenges in modern global healthcare. The progressive and often silent nature of diseases such as glaucoma, cataracts, diabetic retinopathy, and conjunctivitis means that millions of patients worldwide experience significant deterioration in visual function before receiving an accurate diagnosis. The consequences of delayed or missed detection are frequently irreversible, making the development of robust, accessible, and transparent automated diagnostic tools a matter of clinical urgency. While the last decade has witnessed remarkable advances in artificial intelligence, particularly in the domain of computer vision through convolutional neural networks (CNNs), the translation of these technologies into genuine clinical practice has remained disappointing. The fundamental obstacle is not classification accuracy -- modern deep learning architectures routinely achieve performance metrics comparable to trained specialists on benchmark datasets -- but rather the opacity, inflexibility, and clinical narrowness of existing systems. Current AI-based retinal diagnostic tools operate almost exclusively as black-box classifiers: they accept an image as input and produce a disease label as output, with little regard for how that decision was reached, what symptoms the patient is reporting, or what should happen next in the patient's care journey. This paper introduces a novel, end-to-end intelligent eye disease detection and management framework that fundamentally reimagines the role of AI in ophthalmological diagnosis. The proposed system moves well beyond single-modality image classification by integrating four tightly coupled components: a multi-stage CNN pipeline for hierarchical feature extraction from retinal fundus images, a voice-enabled symptom capture interface for multimodal patient input, an AI-driven clinical reasoning engine that cross-validates imaging findings against reported symptoms, and a comprehensive post-diagnosis care coordination module. The result is a system that does not merely classify retinal images but reasons about them in a manner that is transparent, patient-centered, and actionable. Clinicians receive not a binary label but a structured diagnostic narrative accompanied by confidence estimates, symptom consistency scores, risk stratification, specialist referral guidance, and automated scheduling of follow-up appointments.

Volume 14 Issue 4, April-2026

Pages : d630-d639