GlaucoNet: A Novel Technique to Identify Glaucoma at its Early Stage

Glaucoma occurs due to increased intraocular pressure and damage to the optic nerves. If left untreated or undetected, it can lead to irreversible blindness. As one of the most severe eye diseases, timely detection and treatment are crucial to preventing complete vision loss. Early diagnosis plays a vital role in protecting patients from permanent blindness. Traditionally, glaucoma is diagnosed manually by ophthalmologists, a process that requires significant time, expertise, and cost. While several detection techniques are still in the experimental stage, they have not been widely implemented in real-time clinical systems. In this paper, the authors propose a novel technique based on the LeNet architecture, which can effectively detect glaucoma at an early stage. The proposed approach utilizes the gray channels of fundus images and applies data augmentation to generate a diverse and extensive dataset for training a convolutional neural network model. Leveraging the LeNet architecture, the method demonstrated good performance in detecting glaucoma across multiple datasets, including DRISHTI-GS, RIM-ONE, DRIVE, DRIONS-DB, and JSIEC datasets. On the DRISHTI dataset, the proposed model achieved a detection accuracy of 98.67%, a specificity of 91.76%, an area under the curve of 95.68% and sensitivity of 89.78%. With its high accuracy, this approach has the potential to assist clinicians in diagnosing early-stage glaucoma, enabling timely interventions and reducing the risk of irreversible vision loss.