A Deep Learning Augmented System for Automated Diagnosis of Hypertensive Retinopathy using Retinal Fundus Images

Hypertensive Retinopathy (HR) is a common, vision-threatening complication of high blood pressure. A 2023 study in India revealed a 49.33% prevalence of HR among hypertensive individuals, with 81% unaware of their condition. These figures highlight the urgent need for accessible diagnostic tools, especially since early HR signs are often invisible, making timely diagnosis difficult—particularly in resource-limited settings without specialist care. Fundus imaging is the gold standard for HR diagnosis but is often labor-intensive, inaccessible, and time-consuming. To address this, a dataset of 3,300 digitized fundus images (from both HR-diagnosed and HR-free patients) was used to train a Convolutional Neural Network (CNN) model based on YoloV8. The images were split into training (2,400), validation (450), and test (450) sets. Image processing techniques, including CLAHE and Otsu Thresholding, were applied to enhance accuracy. The CNN model achieved training, validation, and test accuracies of 95%-98%, with AUROC scores ranging from 0.96-0.99. An ML-based web app for real-time HR detection was also developed, and internal/external validation demonstrated 95%-98% accuracy. The current study demonstrated the potential of machine learning models in aiding early detection of hypertensive retinopathy in resource-limited settings, offering a valuable tool to support ophthalmologists and pathologists in clinical decision-making.