Automatic Classification and Earlier Detection of Diabetic Retinopathy Using Deep Learning

Abstract

Glaucoma is among the most often occurring causes of irreversible visual loss. Based on fundus images, numerous methods for automatic glaucoma detection have recently been proposed. However, none of the current methods for glaucoma detection can effectively remove fundus images with a lot of redundancy, which could make glaucoma detection less reliable and accurate. A scheme based on Convolutional Neural Network is used to quickly detect glaucoma. This work aims to improve Convolutional Neural Network architectures through evolution so that they can improve glaucoma diagnosis accuracy and sensitivity by using the fundus image of the eye to investigate the efficacy of color fundus image to distinguish glaucoma’s with Deep Convolutional Neural Networks (DCNNs). The capacity for discrimination was influenced by the quality of the images, and the inclusion of images of poor quality in the analysis decreases the area under curve by 0.1 to 0.2.