Glaucoma Detection using Fundus Images by Extracting Localized Disc Features

Abstract

Glaucoma is one of the leading causes of blindness worldwide. It occurs due to high pressure in the eyes and other factors such as family history, age, ethnicity, etc. It damages the optic nervous system which is irreversible damage. That's why regular screening for glaucoma is crucial and recommended. Researchers are continuously searching for better methods to identify glaucoma at early stages before it becomes worse and incurable. Significant work has been conducted on it, but there is still room for improvement. The main goal of this study is to propose a reliable system for glaucoma detection that considers the key factors contributing to glaucoma development, in accordance with the decisions made by clinical experts. In this work, the U-net model is used with EfficentNetb3 as a backbone model for optic disc and optical cup segmentation. In addition, a general deep learning model that has 1D convolution layers and other basic layers is used for glaucoma classification. Features extracted from optical disc and optical cup are used to train the deep learning model and overall 95% classification accuracy and 0.92 AUC are achieved for glaucoma classification on the RIGA dataset.