Personalized prediction of post-SMILE refractive outcomes using a machine-learning nomogram

Area of Science:

• Ophthalmology
• Biomedical Engineering
• Data Science

Background:

• Small Incision Lenticule Extraction (SMILE) is a popular refractive surgery for myopia and myopic astigmatism.
• Accurate prediction of refractive outcomes is crucial for patient satisfaction and surgical success.

Purpose of the Study:

• To develop and validate a personalized, machine learning-driven nomogram for predicting refractive outcomes after SMILE.
• To enhance the precision and predictability of refractive surgery planning.

Main Methods:

• Retrospective analysis of 1253 eyes undergoing SMILE.
• Feature selection using clinical expertise and statistical methods.
• Training and evaluation of four machine learning models (linear regression, decision tree, random forest, neural network).

Main Results:

• The random forest model achieved the highest predictive performance (MAE: 0.18 D, RMSE: 0.24 D, R2: 0.92).
• The final nomogram incorporated 11 key predictors and showed strong external validation.
• 92.3% and 99.4% of eyes achieved outcomes within ±0.50 D and ±1.00 D of intended correction.

Conclusions:

• A machine learning-derived personalized nomogram offers a highly accurate tool for forecasting SMILE refractive outcomes.
• This nomogram can improve precision and predictability in clinical refractive surgery planning.
• The tool is transferable and may aid in optimizing surgical planning for individual patients.