ABSTRACT
Glaucoma belongs to a class of complex and asymptomatic group of diseases and is often characterized by temporary or sustained elevation of intraocular pressure (IOP) that damages the optic nerve (ON). Glaucoma cannot be cured fully, but the progression of the disease can be controlled by proper diagnosis and medication. The chapter presents a novel method to automatically detect and grade glaucoma based on the severity level from fundus images using deep learning approach. A 25-layer convolutional neural network (CNN) is developed and trained efficiently in extracting highly robust features from the acquired retinal fundus datasets. The extracted features are first classified into healthy and glaucomatous cases during testing. The developed algorithm is able to detect the severity level of glaucoma and grade them into three basic classes: early, moderate, and deep glaucoma. One more class, ocular hypertension (OHT) is also added, which serves as the risk factor for glaucoma occurrence. The system is able to achieve an overall accuracy of 99% using 1155 digital fundus images. Performance metrics suggest that deep learning based assessment of digital fundus images can be used as a viable decision-support system clinically in large-scale glaucoma screening and detection programs.
