A Hybrid Approach for Diabetic Retinopathy Classification Using EfficientNetB5 and ResNet50 Models

Diabetic retinopathy (DR), a severe diabetes-related eye complication that can result in blindness if left untreated, progressively damages retinal blood vessels with varying degrees of severity. Traditional screening methods predominantly rely on manual expert evaluation, which can be time-intensive, prone to human error, and subject to interpretation variability due to inconsistent image quality. To address these challenges, our research presents an innovative hybrid deep learning solution that seamlessly combines EfficientNetB5 and ResNet50 architectures for automated DR classification. This advanced model strategically harnesses EfficientNetB5 superior feature extraction capabilities alongside ResNet50 sophisticated deep layer processing power to accurately classify DR into mild, moderate, severe, and proliferative stages. Through extensive training on a comprehensive dataset of retinal fundus images, our hybrid approach demonstrates significant improvements in early detection accuracy, reduced diagnostic inconsistency, and enhanced reliability compared to conventional methods. The results validate that this integrated approach provides healthcare professionals with a robust, efficient, and reliable automated screening tool for clinical applications, ultimately addressing the current limitations in DR diagnosis and management while potentially improving patient outcomes through earlier intervention.