Corneal Endothelium and Tear Film Metrics Enhance the Accuracy of Machine Learning Prediction in Implantable Collamer Lens Surgery Outcomes.

PURPOSE: To develop and evaluate a machine learning (ML) model to predict postoperative vault and residual refractive error following Implantable Collamer Lens (ICL) implantation by incorporating corneal functional parameters, and to investigate their influence on prediction accuracy.

METHODS

This study included 282 eyes implanted with ICLs from 142 patients. A Marine Predators Algorithm combined with a support vector machine (MPA-SVM) for feature selection and optimal modeling was trained to predict postoperative vault residual refractive error using preoperative ocular parameters, with and without corneal functional parameters, including tear film quality and corneal endothelium metrics. Model performance was compared using mean absolute error (MAE), median absolute error (MedAE), R-squared (R2), and Wilcoxon signed-rank tests.

RESULTS

Corneal endothelium metrics significantly improved MPA-SVM performance in vault prediction (P < .05), achieving lower MAE (138.55) and MedAE (110.41) and higher R2 (0.26) and prediction accuracy. For refractive error prediction, combining tear film quality, endothelium metrics, and other ocular parameters yielded the best results for cylinder (MAE = 0.38; MedAE = 0.32; R2 = 0.19), significantly outperforming models using only ocular parameters (P < .05). Endothelium metrics also improved sphere prediction, with numerically lower MedAE and a higher percentage of prediction errors within ±0.25, ±0.50, and ±0.75 diopters.

CONCLUSIONS

In the MPA-SVM ICL prediction model, corneal endothelium metrics significantly improves vault prediction and, when combined with tear film quality parameters, enhances cylinder prediction after ICL implantation. This enhanced model offers ophthalmologists a valuable tool for improving the safety and planning of ICL procedures.