Classification of Glaucoma Severity Stages Using Deep Learning

Glaucoma is one of the major causes of blindness worldwide. It can lead to visual loss and poor quality of life. Glaucoma detection, particularly in the early stages, can help prevent blindness if treated promptly. This paper studies the performance of a deep-learning model in the detection of glaucoma severity using fundus photographs. A method based on applying transfer learning to a single pretrained convolutional neural network (CNN) called EfficientNetB7 has been utilized to distinguish between normal, mild/moderate glaucoma, and severe glaucoma, unlike most of the methods where CNN was used to classify only normal and glaucoma classes. The results demonstrated that the model could predict the three classes with the accuracy, specificity, and sensitivity of 88.17%,94.08%, and 88.17%, respectively, which is comparable to the-state-of-the-art deep learning ensemble method but with much less computational cost. The results showed the possi-bility of extending the automatic glaucoma classification from two to three classes without the need to utilize sophisticated techniques.