Mohammed Saeed
Artificial Intelligence (AI) has been increasingly improving in its accuracy and diagnostic performance in medicine in general. In the context of Keratoconus (KC), it has been heavily studied in the detection of established KC, with more recent evidence reports very high diagnostic accuracy, suggesting the value of machine learning algorithms in the management of KC (1–3). Much of the current discussion has been more focused on one question: Can AI diagnose accurately? A more important question may now be emerging: How can healthcare systems work alongside AI tools to risk stratify, plan management, and increase accessibility of care for patients that are identified as high risk?
Early diagnosis is important in KC, as delayed recognition and intervention can have significant effects on long-term visual outcomes (4). Corneal cross-linking is a well established, highly effective intervention in slowing disease progression. Unfortunately, many patients still present late, usually due to poor access to corneal specialists, and especially where socio-economic and geographical barriers can delay assessment and care. Simply improving diagnostic accuracy may not automatically improve outcomes, unless resources are allocated efficiently and are accessible to all.
Rather than directly replacing ophthalmologists, machine learning algorithms can redesign the patient pathway and guide initial assessment or screening programmes for high risk individuals, and risk stratification, especially in the context of low resources, may represent one of its most valuable applications. Future AI algorithms could be used to identify individuals at the highest risk of progression by integrating multimodal imaging with demographic data, allowing the development of automated, personalised follow-up timelines. The potential of AI-based prognostic modelling is supported by work using artificial intelligence to predict the future need for keratoplasty (5). Although current surveillance strategies are largely guided by clinical assessment, serial tomography, and risk factors of disease progression, this is heavily reliant on clinician contact.
Additionally, the future of AI use in early KC assessment would allow for dynamic risk prediction, allowing for more nuanced inputs that can create dynamic follow up regimes without the need for constant clinician evaluation. This means that patients would not have to make repeated journeys, traversing usually harsher geographical climates, which can pose a significant barrier. Through this method, those individuals identified as high risk can have more repeated tomography or earlier assessment for cross-linking, and those identified as low risk could have longer surveillance intervals, not only eliminating unnecessary clinician visits, but also allows for efficient use of resources.
While an increase in diagnostic accuracy can be very valuable for patient care and outcomes, the greatest contribution of AI may be as a tool for risk prediction, which is an important principle that underpins clinical judgement. This means that this will enable ophthalmologists to deliver the right care, to the right patient, at the right time. AI should be viewed not merely as a diagnostic assistant but as a tool capable of improving equity, efficiency and personalised care that can be accessible to all.
References
1. Afifah, A., Syafira, F., Afladhanti, P. M., & Dharmawidiarini, D. (2024). Artificial intelligence as diagnostic modality for keratoconus: A systematic review and meta-analysis. Journal of Taibah University Medical Sciences, 19(2), 296–303. https://doi.org/10.1016/j.jtumed.2023.12.007
2. Cao, K., Verspoor, K., Sahebjada, S., & Baird, P. N. (2022). Accuracy of machine learning assisted detection of keratoconus: A systematic review and meta-analysis. Journal of Clinical Medicine, 11(3), Article 478. https://doi.org/10.3390/jcm11030478
3. Maile, H., Li, J.-P. O., Gore, D., Leucci, M., Mulholland, P., Hau, S., Szabo, A., Moghul, I., Balaskas, K., Fujinami, K., Hysi, P., Davidson, A., Liskova, P., Hardcastle, A., Tuft, S., & Pontikos, N. (2021). Machine learning algorithms to detect subclinical keratoconus: Systematic review. JMIR Medical Informatics, 9(12), Article e27363. https://doi.org/10.2196/27363
4. Magklaras, E., Karamitsou, K., Diakonis, V. F., Mprotsis, T., & Tsaousis, K. T. (2026). Early detection of keratoconus: Diagnostic advances and their impact on visual outcomes—A systematic review. Medicina, 62(1), Article 42. https://doi.org/10.3390/medicina62010042
5. Yousefi, S., Takahashi, H., Hayashi, T., Tampo, H., Inoda, S., Arai, Y., Tabuchi, H., & Asbell, P. (2020). Predicting the likelihood of need for future keratoplasty intervention using artificial intelligence. The Ocular Surface, 18(2), 320–325. https://doi.org/10.1016/j.jtos.2020.02.008