Assessment of IOP After Corneal Refractive Surgery Based on AI

IOP Changes and Value Prediction after FS-LASIK and SMILE A total of 350 patients undergoing SMILE and 350 undergoing FS-LASIK between October 1, 2023, and January 2, 2025, will be enrolled. Parameters including spherical equivalent (SE), central corneal thickness (CCT), mean keratometry (Km), residual stromal bed thickness (RST), percent tissue altered (PTA), and IOP values preoperatively and at 1 week, 1 month, and 2 months postoperatively will be recorded. These parameters will be included in a multiple linear regression model to analyze factors influencing postoperative IOP and derive a preliminary estimation formula for postoperative IOP measurements. Differences in IOP changes between the two surgical groups will be compared, and other influencing factors will be explored. Assessment of IOP Measurement Levels after CRS Based on Machine Learning Data from approximately 10,300 SMILE and 11,000 FS-LASIK patients treated between October 1, 2020, and April 1, 2025, will be collected. Parameters affecting IOP, such as SE, CCT, Km, RST, and PTA, will be included to explore a predictive ML model for post-CRS IOP. The information gain algorithm will quantify the influence of various factors on postoperative IOP. Follow-up IOP measurements (preoperative, 1 week, 1 month, 2 months) and Pentacam Shah-corrected IOP will be recorded. The estimation formula from Study 1 will also be used to calculate postoperative IOP. Several different ML models will be constructed based on features with higher weights to build predictive models for judging IOP levels at follow-up timepoints. Clinical validation will be performed. A cross-validation method will be employed to verify the model's effectiveness. Internal validation will first be conducted using the training dataset. The entire modeling dataset will be randomly divided into 10 equal folds. Nine folds will be used for model training, and the remaining fold will serve as the test set. The trained model will predict outcomes on the test set, and prediction errors will be calculated (e.g., sum of squared errors). This process will be repeated 10 times, each time with a different fold as the test set. The average performance across all 10 iterations will be reported.