Transfer learning for early and advanced glaucoma detection with convolutional neural networks

Abstract

Glaucoma affects millions of people worldwide and is an eye disease that can lead to vision loss if left untreated. Open-angle glaucoma is the most common type and gradually leads to vision deterioration without many early warning signs or painful symptoms. Clinical diagnosis of glaucoma by specialists is possible but the methods used are either expensive or takes a lot of time. This paper presents automatic detection of early and advanced glaucoma using fundus images. ResNet-50 and GoogLeNet deep convolutional neural network algorithms are trained and fine-tuned using transfer learning for classification. It is shown that GoogLeNet model outperforms ResNet-50 for the detection of early as well as advanced glaucoma detection.