Machine Learning Shows Promise but Limited Clinical Validation in

A systematic review finds machine-learning approaches can classify keratoconus severity and forecast aspects of progression and surgical outcomes, but the evidence is still too limited to prove clinical benefit. The review identified 41 studies published from January 2015 to October 2025: 20 focused on severity classification, 10 on progression assessment or prediction, and 11 on treatment outcomes. Most evidence came from retrospective, single-center work, with fewer independent multicenter evaluations and prospective clinical workflows. Keratoconus involves corneal thinning, protrusion, and irregular astigmatism. Reported accuracies vary across models and devices; one random-forest model classified 644 eyes with 98.6% accuracy, but calibration and direct comparability remain challenges.