Early Detection of Diabetic Retinopathy by using Deep Learning

Abstract

The eye disease diabetic retinopathy (DR) can cause blindness if not treated. Worldwide, 2.6 million persons lost their sight or had severe vision impairments in 2015 as a result of diabetic retinopathy. By 2020, experts predict that number will have jumped to 3.2 million. In high-income nations, diabetic retinopathy should be less common, but in low- and middle-income countries, finding and treating the condition early should be a top concern. Thanks to recent developments in deep learning, researchers have demonstrated that automated diabetic retinopathy screening and grading is a practical method to reduce manpower requirements. The Kaggle dataset was used to find the winners of the diabetic retinopathy detection contest. In the suggested DL-DRDC methodology, the feature vectors from the previously processed retinal fundus images were retrieved using the modified Efficient Net method. The DL-DRDC Mechanism extracts characteristics from previously studied retinal fundus pictures using a modified version of Efficient Net. The proposed GA evolution selects the best-fitting model from a model population pool in order to optimise deep CNN models. The research and the experimental findings show that the recommended approach operates more efficiently than the other basic models. This research offered a novel and better approach for the diagnosis and severity rating of diabetic retinopathy.

Keywords: Convolutional Neural Network (CNN), Deep learning, Diabetic retinopathy, Model, Patient.

View PDF