• Glaucoma is a neuro-degenerative eye disease that is irreversible and accounts for the majority of visual disabilities around the globe. Glaucoma is mainly characterized by the loss of the optic nerve fibers and astrocytes. This can be examined by measuring the thickness of the neuro-retinal rim and the size of the optic cup with respect to the optic disc. Generally, the qualitative assessment of the optic nerve head, when using fundus images, has been the main focus of assessment. Manual assessment is cumbersome and contemporary automated methods were suffering some limitations.

Most of the contemporary algorithms for automatic glaucoma assessment that use fundus images rely on handcrafted features based on segmentation, which may be affected by the performance of the chosen segmentation method and the extracted features. So a more accurate and efficient automated method was essential.

We developed an automatic tool based on Convolutional Neural Network (CNN) to detect and assess glaucoma. The CNN module was trained with thousands of labeled image sets using efficient techniques. The main power of CNN relies on its ability to extract highly discriminating features at multiple levels of abstraction.

• The CNN consisted of multiple layers where the first layers were meant to extract edges at particular orientations and locations in the image.

• The middle layers detect structures composed of particular arrangements of edges and the last layers detect more complex structures that correspond to parts of familiar objects, or objects that are combinations of these parts.

• Once the fundus images are obtained and pre-processed, the CNN extracts the spatial features from retinal images and then classifies the given input by comparing it with instances on the knowledge base.

• We employed the CNN architectures namely - Inception V3 , ResNet50 ,VGG16 , and VGG19.

• Higher performance over contemporary automation methods
• Better accuracy