6 Minutes
Artificial intelligence (AI) can now identify which patients with keratoconus are likely to experience damaging disease progression years before conventional clinical monitoring reveals changes. Presented at the 43rd Congress of the European Society of Cataract and Refractive Surgeons (ESCRS), a new study from Moorfields Eye Hospital and University College London shows that machine learning applied to corneal scans and clinical data can accurately triage patients for early treatment or safe surveillance. The approach promises to prevent irreversible vision loss, cut corneal transplants, and improve resource allocation in ophthalmology.
Scientific background: What is keratoconus and why timing matters
Keratoconus is a degenerative corneal disease that typically begins in adolescence and progresses into adulthood, causing the normally dome-shaped cornea to thin and bulge outward. The distorted cornea produces blurred or distorted vision and can severely reduce quality of life. Estimates suggest keratoconus affects up to 1 in 350 people in some populations. Management ranges from specialty contact lenses to a minimally invasive procedure called corneal cross-linking (CXL), which uses riboflavin (vitamin B2) drops and ultraviolet light to stiffen and stabilise the cornea.
Early CXL—performed before permanent scarring or severe thinning—can halt progression and prevent the need for corneal transplant in the vast majority of cases. The clinical challenge is predicting who will progress. Today, most patients require years of regular monitoring with optical coherence tomography (OCT) and biomechanical assessments to detect progression; by the time deterioration is confirmed, some damage is irreversible.
Study design and AI performance
Researchers led by Dr Shafi Balal analysed a large retrospective cohort of patients referred to Moorfields Eye Hospital for keratoconus assessment and monitoring. The team combined 36,673 OCT images from 6,684 unique patients with routine clinical data and trained an AI algorithm to predict disease trajectory from the first clinical visit.
Key results
The model separated patients into clinically useful risk groups: approximately two-thirds were classified as low risk (suitable for continued monitoring), and about one-third were identified as high risk and recommended for prompt cross-linking. When the algorithm received imaging and data from a second visit, accuracy rose further—correctly categorising up to 90% of patients. Cross-linking success rates reported in the literature exceed 95% when performed before structural scarring, underscoring the potential of early intervention guided by AI.
Dr Balal summarised the findings: "Our research shows that we can use AI to predict which patients need treatment and which can continue with monitoring. This is the first study of its kind to obtain this level of accuracy in predicting the risk of keratoconus progression from a combination of scans and patient data." He noted that while the current work used one OCT device, the methods and algorithm can be adapted to other imaging platforms and will undergo further safety testing before clinical deployment.
Clinical implications and health-system benefits
If validated in prospective multicentre trials, algorithmic triage could shift keratoconus care from reactive to proactive. Anticipated benefits include:
- Preventing avoidable vision loss by delivering cross-linking before irreversible scarring occurs.
- Reducing the number of corneal transplants and their associated complications and recovery burden.
- Decreasing the frequency of unnecessary clinic visits for low-risk patients, freeing capacity for complex care.
- Allowing specialists to prioritise patients with the greatest need, improving overall care pathways.
Dr José Luis Güell, ESCRS Trustee and Head of Cornea, Cataract and Refractive Surgery at Instituto de Microcirugía Ocular (not involved in the study), commented: "Keratoconus is a manageable condition, but knowing who to treat, and when and how to give treatment is challenging. Unfortunately, this problem can lead to delays, with many patients experiencing vision loss and requiring invasive implant or transplant surgery." His remarks highlight the clinical urgency of better risk stratification.
Limitations, validation and next steps
The study’s limitations include its reliance on data from a single OCT device and a retrospective design. The authors acknowledge the need for prospective validation across different devices, patient populations, and healthcare systems to confirm generalisability. The algorithm will undergo safety testing and regulatory review prior to any clinical rollout.
Researchers are already planning a next-generation AI trained on millions of eye scans to expand capabilities beyond keratoconus prediction. Potential extensions include automated detection of corneal infections, early identification of inherited retinal or corneal diseases, and integration with electronic health records for longitudinal risk modelling.
Expert Insight
Dr. Maya Thompson, consultant ophthalmologist and AI-in-healthcare researcher, offers a practical perspective: "Machine learning models are only as useful as their integration into clinical workflow. For keratoconus, a validated triage tool would be transformative—allowing clinicians to offer cross-linking early to the right patients while safely reducing follow-up for others. The critical next steps are multicentre trials, transparent performance reporting by device type, and clear pathways for patient consent and data governance. Done correctly, AI can both protect vision and relieve strain on eye-care services."
Related technologies and future prospects
The work sits at the intersection of ophthalmic imaging, computational diagnostics, and translational AI. Key enabling technologies include high-resolution OCT, scalable cloud-based model training, and interoperable electronic health records. Regulatory frameworks for medical AI, applied clinical trials, and clinician training will determine how quickly diagnostic algorithms move from research to routine care.
Conclusion
This study demonstrates that AI applied to tens of thousands of OCT scans and clinical records can predict keratoconus progression early enough to alter treatment decisions. By enabling targeted, timely corneal cross-linking, the technology could prevent vision loss, lower transplant rates, and optimise ophthalmology resources. Pending further validation and device-agnostic testing, algorithmic risk stratification represents a promising step toward personalised, preemptive eye care.
Source: sciencedaily
Leave a Comment