Detection of Diabetic and Hypertensive Retinopathy Using Deep Learning Models

The diagnosis of diabetic and hypertensive retinopathy, a major cause of blindness in long-term diabetes patients, is a critical topic that we address in this work. Traditional diagnostic techniques like Fluorescein angiography, an imaging test which is used to view the condition of blood vessels present in the retina of eye, are not so reliable and they are not widely available. So, we suggest a sophisticated method that makes use of two different datasets, ODIR-5K and Messidor-2 to improve diagnosis accuracy by utilizing deep learning techniques. The popular deep learning architectures used in this approach are ResNet101, ResNet152, InceptionV3, Inception ResNetv2, DenseNet201, and Xception. Adding to the foundational work, we present the Xception model, obtaining impressive 98% and 89% accuracy for Messidor-2 and ODIR-5K, respectively. This innovative method provides a more feasible and efficient alternative to existing diagnostic techniques for the timely diagnosis of hypertensive retinopathy (HR) and diabetic retinopathy (DR). The combination of sophisticated models with a variety of datasets makes easy to identify severity of disease of the patients.