Has AI Transformed Keratoconus Care?

AI holds significant promise for keratoconus and ectatic corneal disease (ECD) management, but there are still some critical gaps, according to Mazen Sinjab MD.

AI use in keratoconus and ECD is widely recognised and still under investigation, especially for diagnostics and classification, treatment planning, guidelines, research frontiers, and education. For some of these applications, AI has proven an effective and valuable ally alongside keratoconus and ECD specialists; in others, however, there is still a long way to go.

For instance, as Professor Sinjab underlined, there are already some commercially available AI-based indices for keratoconus diagnosis and classification.

“We have not yet reached absolute idealism,” he said. “AI could approach more than 0.99 area under curve (AUC) by a deep learning machine in the field of clinical keratoconus, while the sensitivity drops to 76% when talking about subclinical keratoconus due to the lack of universal definition and low certainty of evidence.”

As for treatment planning, AI is used in six domains: cross-linking optimisation, intracorneal ring selection, contact lens fitting, IOL calculation, decision support, and progression prediction. When looking at the relevant literature, there is a concentration of evidence in corneal cross-linking optimisation and progress prediction. Despite promising results, Prof Sinjab noted no AI treatment planning tool has been validated in a prospective clinical trial, and no tool replaces clinical judgement.

Several AI models have shown strong performance in predicting keratoconus progression. The Moorfields two-visit model raises area under the receiver operating characteristic curve from 0.84 to 0.93, well above age alone (0.63), while Ke Cao MD’s clustering approach identifies three natural progression phenotypes (fast, slow, and minimal).^1,2 Naoko Kato MD predicts cross-linking needs directly, and Hassan Hashemi MD’s systematic review of 10 studies confirms AUC values consistently between 0.75 and 0.93.^3,4 Together, two-visit models combined with phenotype stratification could reduce unnecessary follow-ups by 83% while maintaining 96% specificity, making it the strongest near-term case for AI adoption in keratoconus management.

A significant gap also exists in the establishment of guidelines and research due to fears of inaccuracy and hallucination. However, traditional systematic reviews are time-consuming, and AI could help the process with automation, efficient lead application, and automated analysis.

Prof Sinjab and his team are currently developing a valid AI system to speed up the process of conducting research. Nevertheless, human research judgement remains paramount. In his project, a three-layer architecture is built to ensure AI complements rather than replaces human expertise.

Finally, AI is useful in medical education. A medical education assistant can be fed with sources such as books and articles and used by students to get immediate answers to their questions through a reference-based systematic approach.

Prof Sinjab’s team created Razeen, an AI medical assistant he uses in his teaching activities. Razeen is tailored specifically for refractive surgery. It assists doctors with personalised treatment planning, standard and customised laser profiles, preoperative assessments, and diagnostic investigation.

Prof Sinjab spoke at the 3rd World Keratoconus Congress in Florence, Italy.

Mazen Sinjab MD, MSc, ABO, PhD, FRCOphth (London), CertLRS, FRCSEd, SSBO-Ed is Adjunct Clinical Assistant Professor at University of Sharjah, UAE; Consultant Ophthalmologist Surgeon at Dr Sulaiman Al Habib Hospital, Dubai, UAE; and General Secretary of the International Keratoconus Society. prof.sinjab@sinjabacademy.org

1. Balal S. “Predicting Keratoconus Progression Through Multi-Modal Deep Learning,” presented at the European Society of Cataract & Refractive Surgeons Annual Congress, Copenhagen, Denmark, 14 September 2025.

2. Cao, et al. Intelligence-Based Medicine, 2023. doi:10.1016/j.ibmed.2023.100095

3. Kato N, et al. J Clin Med, 2021. doi:10.3390/jcm10040844

4. Hashemi H, et al. Int Ophthalmol, 2025. doi:10.1007/s10792-025-03855-1