Prediction of Subjective Refraction From Anterior Corneal Surface, Eye Lengths, and Age Using Machine Learning...

Area of Science:

• Ophthalmology
• Biomedical Engineering
• Machine Learning

Background:

• Accurate refractive error prediction is crucial for vision correction.
• Traditional methods rely on subjective patient responses and extensive measurements.
• Developing objective, data-driven prediction models is an active area of research.

Purpose of the Study:

• To develop and evaluate a machine learning regression model for predicting subjective refractive prescription.
• To identify the minimum set of ocular biometry and corneal topography features required for accurate prediction.
• To assess the achievable accuracy of such models.

Main Methods:

• Utilized anterior corneal surface parameters (Zernike coefficients, keratometry), axial length, anterior chamber depth, and age from 355 eyes.
• Split data into training (75%) and test (25%) sets.
• Trained and optimized various machine learning regression algorithms using 10-fold cross-validation, with Gaussian process regression performing best.

Main Results:

• Gaussian process regression models achieved mean absolute errors of approximately 1.00 D for the spherical component and 0.15 D for astigmatic components.
• Neighborhood component analysis indicated that keratometry, age, and axial length were the most significant predictors.
• Increased topographic detail did not significantly improve prediction accuracy, especially for the spherical component.

Conclusions:

• Subjective refraction can be predicted using a minimal set of biometric data (keratometry, age, axial length).
• High-order corneal topography details do not substantially enhance prediction accuracy over simpler measures.
• While machine learning shows promise, its application in predicting subjective refraction for individuals at statistical extremes may be risky and impractical.