Development and validation to predict visual acuity and keratometry two years after corneal crosslinking with progressive keratoconus by machine learning

Purpose To explore and validate the utility of machine learning (ML) methods using a limited sample size to predict changes in visual acuity keratometry 2 years following corneal crosslinking (CXL) for progressive keratoconus. Methods The study included all consecutive patients with keratoconus who underwent CXL from July 2014 December 2020, year follow-up period before 2022 develop model. Variables collected patient demographics, acuity, spherical equivalence, Pentacam parameters. Available case data were divided into training testing sets. Three ML models evaluated based on their performance predicting corrected distance (CDVA) maximum (K max ) compared actual values, as indicated by average root mean squared error (RMSE) R-squared ( R values. Patients followed validation set. Results A total 277 eyes 195 sets 43 35 baseline CDVA (26.7%) ratio steep flat /K 1 ; 13.8%) closely associated changes. K 20.9%) was Using these metrics, best-performing model XGBoost, which produced predicted values closest both set = 0.9993 0.9888) 0.8956 0.8382). Conclusion Application approach incorporation identifiable parameters, considerably improved variation prediction accuracy after treatment