Automated diagnosing primary open-angle glaucoma from fundus image by simulating human's grading with deep learning.

Area of Science:

• Ophthalmology
• Artificial Intelligence
• Medical Imaging

Background:

• Primary open-angle glaucoma (POAG) is a major global cause of irreversible blindness.
• Current deep learning methods for POAG diagnosis lack robustness and explainability.
• Automated, transparent diagnostic tools are needed to aid clinical decision-making.

Purpose of the Study:

• To develop and evaluate GlaucomaNet, an automated deep learning algorithm for POAG diagnosis using fundus photographs.
• To enhance the robustness and transparency of AI-driven glaucoma diagnosis.
• To address challenges of data diversity and reliance on perimetry in POAG detection.

Main Methods:

• GlaucomaNet utilizes a two-convolutional neural network architecture to mimic human grading processes.
• The algorithm learns discriminative features and fuses them for accurate POAG classification.
• Evaluation was performed on the Ocular Hypertension Treatment Study (OHTS) and Large-scale Attention-based Glaucoma (LAG) datasets.

Main Results:

• GlaucomaNet achieved high AUC scores of 0.904 (OHTS) and 0.997 (LAG) for POAG diagnosis.
• Ensemble network architectures further boosted diagnostic accuracy.
• The algorithm demonstrated high accuracy with improved transparency (comprehensiveness scores of 97% and 36%).

Conclusions:

• GlaucomaNet shows significant potential for assisting and improving clinical diagnosis of POAG.
• The developed methods increase image data diversity and reduce reliance on perimetry.
• This AI approach offers a more robust and explainable solution for automated glaucoma screening.