Chapter 18 - Deep learning-based diabetic retinopathy detection for multiclass imbalanced data

Abstract

Diabetic retinopathy (DR) is an important blinding disease and it is one of the major disabilities caused by diabetes. Automated diagnosis for DR detection is very effective for clinical usage and is an efficient method of assistance for doctors. Various approaches have been proposed by researchers to detect DR automatically using retina images. In this work, a deep learning approach is proposed to recognize DR automatically for five different imbalanced classes of DR images. The proposed model is trained using various pretrained convolutional neural network (CNN) architectures coupled with hyperparameter tuning and transfer learning. The results have demonstrated better accuracy in dealing with data imbalances using CNN architectures and transfer learning for classifying DR images.