Automated Staging of Diabetic Retinopathy Using a 2D Convolutional Neural Network

Abstract

An accurate detection and classification of diabetic retinopathy is critical to better assess the disease and possibly slow down its progression. Several methods are used for the diagnosis of diabetic retinopathy including dilated eye examination, fluorescein angiography, optical coherence and fundus photography. In this paper, a 2D convolutional neural network is introduced for the analysis and classification of fundus images into one of the four main stages of diabetic retinopathy. A training accuracy of 99.9% and a Leave One Out Cross Validation testing accuracy of 80.2% were achieved after training 101 fundus images representing 4 different stages of the disease for 50 epochs.