AI Predicts Vision Loss in Young Adults Years Before Clinical Detection

Researchers have developed an artificial intelligence system capable of accurately predicting which patients with keratoconus will need urgent treatment to prevent vision loss, marking a significant advancement in eye care. The study, presented at the 43rd Congress of the European Society of Cataract and Refractive Surgeons (ESCRS), was led by Dr. Shafi Balal and colleagues from Moorfields Eye Hospital NHS Foundation Trust and University College London (UCL), UK. Keratoconus, a condition affecting up to one in 350 people, causes the cornea to thin and bulge outward, leading to distorted vision. While it often begins in adolescence and can be managed with contact lenses, some cases progress rapidly, requiring corneal cross-linking—a treatment that strengthens the cornea using ultraviolet light and riboflavin drops—or even a corneal transplant. Currently, doctors rely on long-term monitoring to determine which patients are at risk of progression, delaying treatment until damage has already occurred. This new AI model changes that by analyzing optical coherence tomography (OCT) scans of the cornea, combined with patient data, to predict disease progression from a single visit. The system was trained on 36,673 OCT images from 6,684 patients monitored over two years or more. Using deep learning, the algorithm successfully categorized patients into low-risk and high-risk groups. At first visit, it correctly identified two-thirds as low-risk—those who could safely avoid treatment—and one-third as high-risk, who would benefit from early cross-linking. When data from a second visit was added, the accuracy rose to 90%. The implications are profound. Early intervention with cross-linking can halt progression in over 95% of cases, preventing the need for invasive surgery and long recovery times. For low-risk patients, the AI could reduce unnecessary follow-ups, freeing up healthcare resources and improving patient experience. Dr. Balal emphasized that this is the first study to achieve such high accuracy using a large, real-world cohort and multi-modal data. Although the model was tested on one type of OCT device, the researchers believe it can be adapted for other imaging systems. The AI is now undergoing further safety and validation testing before clinical deployment. The team is also developing a more advanced version trained on millions of eye scans, with potential applications in detecting other eye conditions like infections and inherited diseases. Dr. José Luis Güell, an independent expert from Spain, praised the study, calling it a breakthrough in managing keratoconus. He noted that the ability to predict progression early could prevent vision loss and reduce reliance on complex, high-risk procedures. He emphasized that AI could transform patient care by enabling timely, personalized treatment and reducing the burden on healthcare systems. The research was supported by the ESCRS Digital Research Award, Frost Trust, and the UK’s National Institute for Health and Care Research (NIHR). With further validation, this AI tool could become a standard part of eye care, offering a proactive, data-driven approach to preserving sight in young adults affected by keratoconus.