Harnessing ResNet50 and EfficientNetB5 for Detection of Diabetic Retinopathy Using Explainable AI

Diabetic retinopathy, often regarded as a consequential outcome of diabetes, has a profound ability to deteriorate vision, potentially culminating in blindness if undetected in preliminary stages. Current diagnostic methods, involving manual examination of fundus images, can be time-consuming. This paper proposes an innovative approach that combines deep learning models, specifically ResNet50 and EfficientNetB5, with Explainable AI techniques, such as the SHAP method, to enhance the interpretability of diabetic retinopathy detection. This combination allows for more accurate identification of the disease stage and provides insights into the model’s decision-making process, which can guide treatment and interventions. The proposed approach aims to improve the accuracy, fairness, transparency, and outcomes of diabetic retinopathy detection, fostering trust in AI models and enabling informed decision-making in healthcare settings.