Development of A Novel Deep Neural Network Approach to Diabetic Retinopathy Diagnosis

Diabetic retinopathy (DR) is a prevalent ocular condition and a major contributor to vision loss in those with diabetes. The most efficient method to control the condition is by regular screening using fundus photography and prompt action. The substantial number of individuals with diabetes and their extensive need for screening has sparked interest in a computer-assisted and entirely automated detection of diabetic retinopathy (DR). Deep neural networks have made significant advancements in numerous tasks in the past few decades. In order to streamline the process of diagnosing DR and provide suitable recommendations to DR patients, we have constructed a dataset consisting of DR fundus images that have been annotated with the corresponding treatment technique necessary. We utilized this dataset to build deep convolutional neural network models for the purpose of grading the seriousness of DR fundus pictures. In the studies, we attained an accuracy of 90% during a four-degree classification job. The models we developed were implemented on a server in the cloud and used to offer pilot DR screening services to many hospitals. During the scientific assessment, the system demonstrated a high level of accuracy, achieving an accuracy rate of 93% when compared to ophthalmologists. This indicates the usefulness of our research.