Artificial Intelligence Approach for Diabetic Retinopathy Severity Detection

Abstract

Identifying the diabetic retinopathy (DR) severity in the retina images taken under a variety of imaging conditions is more challenging. There are five classes that are commonly classified on retinal images based on severity of DR disease such as No DR, mild NPDR (non proliferative diabetic retinopathy), moderate NPDR, severe NPDR and PDR (proliferative diabetic retinopathy). Artificial intelligence is an emerging area in the medical diagnosis industry, in specific, deep learning algorithms are used for classifying retina images for accurate diagnosis of disease. The proposed work acquired retina images from publicly available Kaggle repository and loaded into improved grid search Convolutional Neural Network model for accurate diagnosis on retina images with five different classes. This novel model helps ophthalmologists to classify DR as five stages based on severity from No DR to PDR. The Experimental study showed that the proposed model has performed better than the existing Convolutional Neural Network model with the accuracy of 89%.