Big-data and artificial-intelligence-assisted vault prediction and EVO-ICL size selection for myopia correction

Area of Science:

• Ophthalmology
• Artificial Intelligence
• Big Data Analytics

Background:

• Implantable collamer lens (ICL) surgery requires precise sizing for optimal outcomes.
• Accurate prediction of postoperative vault height is crucial for ICL safety and efficacy.

Purpose of the Study:

• To develop and evaluate artificial intelligence (AI) models for predicting ICL size and vault height.
• To leverage big data analytics for improved surgical planning in ICL procedures.

Main Methods:

• Utilized a dataset of 6,297 eyes from 3,536 patients implanted with ICLs.
• Employed Random Forest, Gradient Boosting, and XGBoost algorithms for prediction tasks.
• Investigated feature importance using permutation and impurity-based methods.

Main Results:

• Random Forest achieved the highest accuracy in vault prediction (R²=0.315) and ICL sizing (82.2%).
• Classification accuracy for vault prediction reached 0.828 with a mean AUC of 0.765.
• Gradient Boosting and XGBoost also demonstrated strong performance in ICL sizing (81.5%-81.8% accuracy).

Conclusions:

• AI models, including Random Forest, Gradient Boosting, and XGBoost, are effective for vault and ICL size prediction.
• AI can assist ophthalmologists in enhancing ICL surgery safety, planning, and predicting clinical results.