Automatic Computer-Aided Diagnosis Tool for Glaucoma Detection Using U-Net and CNN on Retinal Fundus Images

Glaucoma is a progressive optic neuropathy and a leading cause of irreversible blindness worldwide, where vision loss can be prevented only through timely diagnosis and intervention. Therefore, there is a requirement for the development of an automatic screening system to detect glaucoma at its preliminary stages. In this work, a fully automated computer-aided diagnosis (CAD) framework is proposed for early glaucoma detection using color retinal fundus images. The proposed method integrates image validation, region extraction, data balancing, data preprocessing, structural segmentation, and deep learning-based classification. The optic disc and optic cup are then segmented with a U-Net–based architecture, which allows calculation of important anatomic measures such as the vertical cup-to-disc ratio and neuroretinal rim parameters. These handcrafted features along with deep feature representations are used to train a custom convolutional neural network classifier to discriminate between glaucomatous and healthy eyes. We test the proposed system on publicly available benchmark datasets and show that it has strong robustness and generalization capacity. The achieved accuracy, sensitivity, specificity and AUC are 88.38%, 84.95%,81.43% and 93.24% respectively, in the validation dataset. The whole CAD system is applied with an easy-to-use clinical interface implemented with Streamlit to realize real-time inference and be easily integrated into the screening workflow. The proposed method is a cost-effective and reliable technique for glaucoma screening and has significant potential for mass-screening and tele-ophthalmology applications, especially in resource-limited environments.