Improving the estimation of the spherical equivalent subjective refraction using objective information on...

Area of Science:

• Ophthalmology
• Optometry
• Computer Science

Background:

• Machine learning (ML) and deep learning (DL) have shown limited success in predicting subjective refraction endpoints objectively.
• Accurate subjective refraction is crucial for optimal vision correction.

Purpose of the Study:

• To enhance the predictive accuracy of objective refraction measurements.
• To develop improved ML models incorporating accommodative response and optical quality data.

Main Methods:

• Trained linear regression models using normal equations.
• Utilized data from a Hartmann-Shack aberrometer and an autorefractor.
• Tested three models on 176 eyes.

Main Results:

• The best model significantly reduced mean absolute error by 40% compared to commercial autorefractometers.
• Achieved 95% limits of agreement with subjective refraction of ±0.54 D.
• Demonstrated performance approaching inter-examiner variability in subjective refraction.

Conclusions:

• Incorporating accommodative response and optical quality data substantially improves objective refraction prediction.
• The developed ML models offer a more accurate and reliable method for determining subjective refraction endpoints.
• This approach shows promise for advancing objective eye examination techniques.