Automated Grading of Diabetic Retinopathy using DenseNet-169 Architecture

Diabetic retinopathy (DR), a severe eye disease, is a diabetes complication, and one of the world’s leading causes of blindness. Early diagnosis of DR may enable timely treatment and prevent permanent vision loss. DR has five stages of severity, namely-no DR, mild, moderate, severe and proliferative. Retinal fundus images are widely used for DR diagnosis. For the task of an early diagnosis of DR from color fundus images in an automated way we propose the use of DenseNet-169 model, a pre-trained convolutional neural network (CNN). In this study we used the publicly available Kaggle APTOS 2019 Blindness Detection dataset. In order to achieve an adequate classification performance, hype-parameter tuning, and data augmentation techniques are used. The model was also compared with two other pre-trained models DenseNet-121 and ResNet-50. Our proposed method achieved a classification accuracy of 96.54%, sensitivity and specificity of 96.23% and 99.21%, respectively, on the test set.