An Efficient Deep Learning-Based Classification Framework for Hypertensive Retinopathy

Abstract

Hypertensive retinopathy (HR) is a well-known eye disease that is caused by high blood pressure (hypertension). In this illness, symptoms typically develop later. The AV nicking, cotton wool patches, constricted veins in the optic nerve, and blood pouring into the eye’s optic nerve all contribute to the appearance of the HR symptoms. HR disease may have different types of serious complications, including retinal artery blockage, destruction of the visual nerves, and maybe vision loss. The automated early detection of this illness can be aided by AI and deep learning models. In this research, a novel dataset for HR is collected from Pakistani hospitals (Pak-HR) and internet sources. Second, a brand-new methodology (Incept-HR) is developed to evaluate hypertensive retinopathy using InceptionV3 and residual blocks. 6,000 digital fundus images from the collected datasets were used to train the Incept-HR system. The proposed classification method, Incept-HR, has 99% classification accuracy and an f1-score of 0.99. The results show that this model produces useful outcomes and can be applied as a diagnostic testing tool. The system is not intended to replace optometrists; rather, it aims to assist professionals. The proposed methodology outperforms both the cutting-edge models VGG19 and VGG16 in terms of classification accuracy.