Performance Analysis of Glaucoma Detection Using Deep Learning Models

Abstract

Glaucoma has become one of the prime reasons for blindness worldwide. It is an irremediable and persistent disease. To intercept this disease, early detection and screening of glaucoma are very important. Deep learning in this context plays a promising role. In this study, we have examined the performance of three deep learning (DL) architectures, including the Convolutional Neural Network (CNN) model using max pooling, the CNN model using average pooling, and the Transfer Learning Xception model to detect glaucoma. Public datasets containing 1250 images are used in our research. The CNN model using max-pooling achieved the highest 87.99% training and the highest 89.11% validation accuracy. The CNN model using average pooling achieved the highest 86.94% training accuracy and the highest 87.83% validation accuracy. The Xception model achieved the highest 97.63% training accuracy and the highest 98.11% validation accuracy.