Diabetic Retinopathy Detection Using Real-World Datasets of Fundus Images

Diabetic Retinopathy (DR) is an eye condition that affects people who have diabetes and damages their retina over time, potentially resulting in blindness. Traditional methods of diabetic retinopathy (DR) screening, overseen by ophthalmologists, exhibit limitations such as scarcity of expert professionals, time intensiveness, and high costs. This highlights the need for a more precise and efficient approach. Our study introduces an innovative solution by harnessing deep learning. Leveraging the Residual Neural Network (ResNet) model, we analyze a diverse dataset of 7000 authorized real-time eye images of patients from “Anand Eye Hospital, Jaipur.” This methodology facilitates credible and early detection of diabetic retinopathy, a pivotal factor in averting vision loss. The severity scale separates the images into five classes: normal, mild, moderate, severe nonproliferative diabetic retinopathy (NPDR), or proliferative diabetic retinopathy (PDR), from a healthy eyeball to a proliferative diabetic retinopathy presence. Remarkably, our approach attains an impressive 93.5% accuracy, bolstering the model's reliability in diabetic retinopathy diagnosis from fundus images. With potential applications spanning urban and rural healthcare landscapes, our study underscores the transformative potential of advanced technology in reshaping retinopathy diagnosis and patient outcomes.