Machine learning algorithm improves accuracy of ortho-K lens fitting in vision shaping treatment.

Area of Science:

• Ophthalmology
• Biomedical Engineering
• Machine Learning

Background:

• Orthokeratology is a vision shaping treatment (VST) that uses specialized lenses to reshape the cornea.
• Accurate estimation of the alignment curve (AC) curvature is crucial for effective VST lens fitting.
• Current methods for AC curvature calculation can be time-consuming and may require multiple lens trials.

Purpose of the Study:

• To develop and evaluate a machine learning (ML)-based model for estimating AC curvature in VST.
• To improve the efficiency and accuracy of orthokeratology lens fitting by minimizing lens trials.
• To compare the performance of ML models against a previously published calculation method.

Main Methods:

• Retrospective data from 1271 myopic subjects were analyzed.
• Four ML algorithms (robust linear regression, SVM, bagging decision trees, Gaussian processes) were trained to predict AC curvatures.
• Input variables included patient demographics, refractive data, and ocular measurements (HVID, K1, K2, ACD, AL).

Main Results:

• Linear SVM and Gaussian process models demonstrated the highest predictive performance.
• R-squared values for AC1K1, AC1K2, and AC2K1 predictions were 0.91, 0.84, and 0.73, respectively.
• The ML methods showed excellent consistency with previous methods, performing best with flat K and e values.

Conclusions:

• The developed ML model offers an efficient approach for estimating VST lens AC curvatures.
• This method can reduce the number of lens trials, enhancing clinical workflow.
• The ML model can also decrease the risk of cross-infection from trial lenses, particularly relevant during pandemics.