Machine learning in keratoconus: severity grading, progression prediction, and surgery outcomes

To review and evaluate machine learning (ML) applications in keratoconus (KC), with a focus on severity classification, progression prediction, and surgical outcome prediction.

A systematic literature search was conducted in PubMed, Scopus, Web of Science, IEEE Xplore, and Google Scholar for English-language studies published between January 2015 and October 2025. Studies applying ML to KC grading, progression prediction, or surgical outcome prediction were included. Data on clinical tasks, imaging modalities, model architectures, validation strategies, and performance were extracted and evaluated.

Forty-one reports were identified, including 20 on severity classification, 10 on progression assessment, and 11 on surgical outcome prediction. Severity models mainly reproduced clinical or device-derived grades, while progression models addressed different endpoints, including tomographic change and subsequent corneal cross-linking. Surgical outcome models predicted optical changes, treatment response or graft outcomes at different follow-up intervals.

ML showed promising internal performance for KC severity classification, while evidence for progression and surgical outcome prediction remains limited. Clinical translation will require standardized endpoints, patient-separated evaluation, multicenter external validation, calibration assessment, and direct comparison with existing clinical decision rules.