Machine learning predictive model based on the corneal biomechanics of clinical data

OBJECTIVE: To develop and validate a machine learning model for predicting postoperative corneal ectasia risk after refractive surgery by integrating multidimensional preoperative and postoperative biomechanical parameters. METHODS: A retrospective cohort of 200 patients who underwent LASIK or SMILE at Shenyang Xingqi Eye Hospital between January 2023 and December 2024 was analyzed. Variables including corneal morphology, biomechanics, and early postoperative changes were screened using Least Absolute Shrinkage and Selection Operator (LASSO) regression. Multiple algorithms-generalized linear model (GLM), gradient boosting machine (GBM), and support vector machine (SVM)-were compared via 10-fold cross-validation. The optimal SVM model was fine-tuned through grid search and further tested in a prospective temporal validation cohort (n = 40, January-September 2025). Model performance was evaluated by AUC, calibration, and decision-curve analysis. RESULTS: LASSO identified PreopCCT, PreopKmax, CH, CRF, and early postoperative changes (PostopCCTChange1m, PostopKmaxChange3m) as key predictors. The SVM model achieved the best discrimination among all models (AUC = 0.88 in cross-validation). In prospective validation, the locked SVM maintained high accuracy (AUC = 0.984, 95 % CI 0.944-1.000; sensitivity = 0.88; specificity = 0.90) with good calibration (slope = 0.96; Brier = 0.098) and net clinical benefit across threshold probabilities of 0.1-0.7. CONCLUSION: The optimized SVM model provides a reliable, data-driven approach for individualized ectasia-risk assessment in refractive surgery. By combining biomechanical and tomographic features, it enables early identification of high-risk patients and supports personalized surgical planning and postoperative monitoring. Further multi-center and long-term validation is warranted to enhance generalizability and clinical implementation.