Title : Early detection of glaucoma using artificial intelligence: A comparative study with ophthalmologists
Abstract:
Background: Glaucoma is the leading cause of irreversible blindness worldwide. Early detection remains challenging due to subtle clinical sings and limited access to advanced diagnostic tools. Artificial lntelligence (AI) has emerged as a promising solution; however, few studies have validated its performance using local datasets from resource-limited settings such as Kazakstan.
Objectives: To evaluate the diagnostic accuracy of a deep learning model for early glaucoma detection using fundus photographs and to compare its performance with board-certified ophthalmologists.
Methods: A diagnostic accuracy study was conducted on a local dataset of 500 fundus images (250 glaucoma,250 healthy) collected at [Your Clinic, Almaty,2025]. Images were preprocessed (normalization, augmentation) and analyzed using a Res-Net-50 convolutional neural network trained with transfer learning. Three independent ophthalmologists assessed the same dataset. Model performance was evaluated by sensitivity, specificity, accuracy, and area under the ROC curve (AUC) with 95% confidence intervals (Cl).
Results: The Al model achieved a sensitivity of 89.2%(95% Cl:85.1-92.3), specificity of 90.4% (95% Cl:86.5-93.2), accuracy of 89.8%, and an AUC of 0.91. Ophthalmologists' mean diagnostic accuracy was 88.7%, with no statistically significant difference compared to Al performance (Mc Nemar is test, p>0.05). Subgroup analysis showed stable Al performance across age, gender, and image quality categories.
Conclusion: This study shows that an Al model trained on a local Kazakhstan dataset can achieve diagnostic accuracy comparable to ophthalmologists in early glaucoma detection.These findings highlight the potential of Al-assisted screening programs, to support earlier diagnosis,improve patient outcomes,and reduce irreversible blindness in resource-limited setting.
Keywords: Glaucoma, Artificial Intelligence, Deep Learning, Optic disc, Early Diagnosis, Fundus Photography, Ophthalmology, Resource- Limited Settings.
