Application of deep neural networks for the analysis of optical images of the vascular retina in patients with diabetes mellitus

The paper investigates the application of deep neural networks for the analysis of optical images of the retinal vascular retina in patients with diabetes mellitus in order to improve the accuracy of automated diabetic retinopathy diagnosis. Modern approaches to retinal image processing based on convolutional neural networks are considered, enabling automatic extraction of informative vascular features without manual feature engineering. A comprehensive evaluation methodology is proposed using classification performance metrics such as accuracy, sensitivity, specificity, and the area under the ROC curve (ROC-AUC), accompanied by a detailed ROC analysis with numerical calculations. Additionally, a correlation analysis between model predictions and blood glucose level indicators (HbA1c) is performed to assess the clinical relevance of the obtained results. The findings demonstrate the high diagnostic potential of deep neural networks and confirm their suitability for integration into computer-aided decision support systems in medical diagnostics.