An Integrated Deep Learning Approach for Computer-Aided Diagnosis of Diverse Diabetic Retinopathy Grading

Diagnosing and screening diabetic retinopathy pose significant challenges in biomedical research. Utilizing the advancements in deep learning, computer-assisted diagnosis has emerged as a potent technique to examine medical images of patients’ eyes and detect damage to blood vessels. Nevertheless, the effectiveness of deep learning models has been impeded by factors such as imbalanced datasets, annotation inaccuracies, limited available images, and inadequate evaluation criteria. In this study, addressed these formidable challenges by employing three established benchmark datasets related to diabetic retinopathy. This approach has facilitated a comprehensive assessment of cutting-edge methodologies. As a result of our study, achieved remarkable precision scores: 93% for standard cases, 89% for mild instances, 81% for moderate conditions, 76% for severe stages, and 96% for diabetic retinopathy phases. Notably, conducted a thorough analysis of a hybrid model that integrates Convolutional Neural Network (CNN) analysis with SHapley Additive exPlanations (SHAP) model derivation. Our findings underscore the suitability of hybrid modeling strategies for detecting anomalies in blood vessels through the utilization of classification models.