Artificial intelligence-based refractive error prediction and EVO-implantable collamer lens power calculation for...

Area of Science:

• Ophthalmology
• Biomedical Engineering
• Artificial Intelligence

Background:

• Implantable collamer lenses (ICLs) are effective for myopia correction, but accurate phakic intraocular lens (IOL) power calculation remains challenging, particularly for low-to-moderate myopia.
• Existing methods for ICL power calculation require refinement to improve postoperative refractive outcomes.

Purpose of the Study:

• To develop and evaluate a novel stacking machine learning (ML) model for predicting postoperative refractive errors.
• To establish a new method for calculating EVO-ICL lens power using ML.

Main Methods:

• A total of 2767 eyes from 1678 patients who underwent non-toric (NT)-ICL or toric-ICL (TICL) implantation were analyzed.
• Stacking ML models (SVR, LASSO, random forest, XGBoost) were trained using ocular dimensional parameters to predict postoperative spherical equivalent (SE) and sphere.
• Model accuracy was compared to the modified vergence formula (MVF) using mean absolute error (MAE), median absolute error (MedAE), and prediction error percentages.

Main Results:

• For NT-ICL, the random forest model achieved the lowest MAE (0.339 D) for SE prediction, while the SVR model had the lowest MAE (0.386 D) for sphere prediction.
• For TICL, the XGBoost model demonstrated the lowest MAE for both SE (0.325 D) and sphere (0.308 D) prediction.
• ML models showed comparable or improved accuracy over MVF, with notable advantages in eyes with low-to-moderate myopia.

Conclusions:

• Stacking ML models offer comparable accuracy to existing MVF models for refractive error prediction after ICL implantation.
• These ML models present potential advantages for patients with low-to-moderate myopia, providing a novel nomogram for refractive error prediction and lens power calculation.