Early diagnosis of diabetic retinopathy using deep learning Techniques

Abstract

Health is wealth. The current changes in human lifestyle have increased Diabetics Mellitus (DM). If the anatomy does not produce enough insulin or if it has insulin resistance, DM evolves. DM will raise the blood glucose level preventing normal blood flow. This affects major organs like the eyes, nerves, and kidney. Diabetic Retinopathy (DR) is a vision disease caused due to the long-term prevalence of DM. DR develops in five stages. The earlier stage of DR will hardly show its development symptoms. The symptoms are notified when the diabetic patients start to lose their sight with blurred vision or vision with dark circles. The bitter truth behind DR is that it cannot be cured with the current healthcare advancement. However, DR can be easily controlled with proper medication if it is detected early. Hence they are asked to have regular screening of their retina either once or twice a year. Going for regular screening of the retina is costlier and time-consuming for diabetic patients. To reduce the time consumption and suggest the next screening, an automated Computer-Aided Diagnosis (CAD) is required. newlineDeep learning techniques are employed to diagnose the DR at an early stage. Designing a Convolutional Neural Network (CNN) from scratch is tedious and time-consuming. Selection of the pre-trained CNN model is done automatically using a Genetic Algorithm (GA). Several pre-trained modelsare made for the initial population. The parent selection from population, crossover, and mutation are repeated until the algorithm is converged. The model which comes out of the algorithm with flying colors will be the best-fit model for DR diagnosis. The experimental analysis showed that the model chosen by the GA is similar to the ResNet50 CNN model newline