AI-Driven Retinal Image Analysis for Early Detection of Diabetic Retinopathy: Comparative Evaluation of Deep Learning Models and Clinical Implications

Diabetic retinopathy (DR) is a critical global public-health concern and one of the foremost causes of preventable blindness.Timely identification is vital to halt irreversible vision loss.Early detection of diabetic retinopathy proves crucial for maximizing the success of treatment because this condition stands as one of the leading blind-causing diseases.Early identification also enhances treatment precision by enabling clinicians to intervene before structural retinal damage becomes irreversible.This study aims to develop and evaluate deep-learning models for automated DR detection using large-scale retinal image datasets.The research assesses how well four architectures-Convolutional Neural Network (CNN), Res Net, VGG16, and Efficient Net were compared using publicly available Kaggle Eye PACS (35,126 images) and IDRiD (516 images) datasets, covering five severity levels.Images were pre-processed through resizing, normalization, and contrast-limited adaptive histogram equalization (CLAHE), with augmentation to balance classes.Each model was trained for 50 epochs (batch = 32, Adam optimizer).The most accurate model proved to be EfficientNet because it reached 99.4% accuracy in classification results.Based on the analysis, EfficientNet proves to be the best model for detecting diabetic retinopathy early.Efficient Net's compound-scaling design enables high performance with reduced computational cost, supporting scalable clinical screening.The research brings modern diagnostic methods through enhanced AI responses with better generalization abilities, leading to better patient health results.Limitations include computational demands, dataset imbalance, and lack of multimodal validation.The study highlights Efficient Net's clinical promise for cost-effective DR detection while emphasising the need for ethical, explainable AI in ophthalmology.