Machine learning predictive model based on the corneal biomechanics of clinical data

Area of Science:

• Ophthalmology
• Biomedical Engineering
• Data Science

Background:

• Corneal ectasia is a sight-threatening complication after refractive surgery.
• Accurate preoperative risk assessment is crucial for patient safety and optimal surgical outcomes.
• Existing methods may not fully capture the complex interplay of factors contributing to ectasia development.

Purpose of the Study:

• To develop and validate a machine learning model for predicting postoperative corneal ectasia risk.
• Integrate multidimensional preoperative and postoperative biomechanical and tomographic parameters.
• Enhance individualized risk assessment for refractive surgery patients.

Main Methods:

• Retrospective analysis of 200 patients undergoing LASIK or SMILE.
• Feature selection using Least Absolute Shrinkage and Selection Operator (LASSO) regression.
• Comparison of Generalized Linear Model (GLM), Gradient Boosting Machine (GBM), and Support Vector Machine (SVM) algorithms.
• Prospective validation of the optimal SVM model.

Main Results:

• LASSO identified key predictors: preoperative corneal thickness (PreopCCT), maximum keratometry (PreopKmax), corneal hysteresis (CH), corneal resistance factor (CRF), and early postoperative changes.
• The SVM model demonstrated superior performance with an AUC of 0.88 in cross-validation.
• Prospective validation showed high accuracy (AUC = 0.984), sensitivity (0.88), and specificity (0.90) with good calibration.

Conclusions:

• The optimized SVM model offers a reliable, data-driven approach for individualized ectasia risk assessment.
• Combining biomechanical and tomographic features enables early identification of high-risk patients.
• Supports personalized surgical planning and postoperative monitoring, though further validation is needed.