Keratoconus loves teenagers. You're 16, worried about SATs and college applications, then your eye doctor drops this bomb: your corneas are warping and nobody knows how fast they'll get worse.
Current medical approach? "Come back every 3 months so we can watch your vision degrade." By the time doctors catch progression, scarring has already started. Once that happens, cross-linking treatment only works 60% of the time instead of 95%.
Your insurance won't pay for preventive treatment until after damage shows up on scans. So kids spend 2-3 years getting their corneas photographed every quarter, waiting to see if they'll need transplant surgery before graduation.
AI Catches What Doctors Miss Because It's Seen 7,000 Cases
Moorfields fed their system 36,673 corneal scans from 6,697 patients. The AI learned to spot micro-patterns that predict which corneas will warp, even when the current scan looks fine to human eyes.
One visit gets 67% accuracy at risk prediction. Two visits, three months apart, hits 90%. Your corneal specialist has maybe seen 100 keratoconus cases in their career. The AI analyzed nearly 7,000.
That's the difference between "looks stable to me" and "I can see the warping will start in 8 months based on patterns invisible to humans."
The Treatment Works Great - If Doctors Don't Wait Too Long
Cross-linking is basically UV-curing your cornea like a 3D printer part. Riboflavin drops, then UV light for 30 minutes. Costs $2,500 per eye but stops 95% of cases dead in their tracks.
Compare that to corneal transplant: $15,000 per eye, plus lifelong anti-rejection drugs, plus the fun of waiting for a donor eyeball.
But here's the fucked part: doctors wait for "measurable progression" before they'll treat you. That means letting your cornea warp 1.5 diopters or thin by 20 microns before insurance approves treatment. By then, some vision loss is already locked in forever.
Why You Can't Get This Tomorrow
Right now it only works with Pentacam scanners. If your eye clinic has some other brand (most do), you're shit out of luck until they train a universal version.
FDA wants 12-24 months of data proving AI-guided early treatment actually prevents vision loss, not just predicts it. So we're talking 2027 at the earliest before this shows up in regular clinics.
Meanwhile, teenagers keep going blind waiting for "measurable progression" that could've been prevented 2 years earlier. Fucking brilliant healthcare system we've built.