Artificial intelligence-assisted glaucoma detection on color fundus images: with comorbidity and cross-institutional analysis

Abstract

Background: Glaucoma is a major cause of irreversible blindness, and early detection is essential to prevent vision loss. Color fundus photography is a simple, low-cost, and noninvasive eye screening method, but diagnosis through this method can be difficult in patients with additional retinal diseases. Although artificial intelligence (AI) can address this difficulty, its effectiveness may vary between hospitals. In this study, an AI glaucoma detection system was developed and tested for reliability across different populations and clinical settings.

Methods: A stepwise AI pipeline was designed that combined image enhancement, automated identification of the optic nerve area, and deep learning-based classification. The system was trained on 1696 images from Taipei Veterans General Hospital and tested on five cross-regional external datasets. The system was also evaluated on a separate internal set of 151 images representing comorbid eye diseases.

Results: The AI system achieved a balanced accuracy of at least 80% on all external datasets. For images with other eye diseases, it achieved an area under the curve of 0.93 and a balanced accuracy of 80.9%. Its performance remained consistent regardless of differences in patient ethnicity, camera types, and image quality.

Conclusion: The proposed AI system can detect glaucoma on standard color fundus photographs with high accuracy across clinical environments and in the presence of comorbid eye diseases. The system may be a practical and affordable tool for large-scale glaucoma screening, particularly in institutions with limited resources.

Keywords: Color fundus; Comorbidities; Cross-institutional; Deep learning; Glaucoma.