AI lens simulation improves agreement with children's reference refraction measurements

Children who need glasses usually undergo more than one kind of optical measurement. An automatic instrument estimates the refraction of the eye, but a clinician's subjective refraction—checking which lens corrections give the clearer result—remains a reference for prescribing. Researchers tested an AI-assisted system that uses a detailed optical measurement to simulate candidate lenses for an individual eye. In a prospective study involving 183 children, its estimates agreed more closely with clinician-derived prescriptions than those produced by a conventional autorefractor.[1]

The work, published in Translational Vision Science & Technology, involved one right eye from each of 183 children, with a mean age of 9.9 years and a standard deviation of 2.0 years. Testing took place after cycloplegia, the use of eye drops to temporarily relax focusing. Each participant underwent traditional autorefraction, the proposed wavefront-based method and masked subjective refraction, with the testing sequence randomised. The masking matters because the clinician's reference measurement should not be knowingly adjusted to match the device being assessed.[1]

The system starts from wavefront measurements produced with a microelectromechanical Shack–Hartmann sensor, checks whether the data are of suitable quality and then simulates the optical effects of trial lenses. A patient-specific optimisation step accounts for small measurement perturbations before suggesting an estimate of the prescription. The study describes this as digital-twin-inspired because it creates a computational model of the eye's optical response. It does not reconstruct every feature of the patient's visual system, and it does not automatically replace the clinician's final judgement.[1]

The central comparison used spherical equivalent, a commonly reported combined measure of spectacle lens power. Conventional autorefraction differed from subjective refraction by a mean 0.21 dioptres, with a standard deviation of 0.52. The AI wavefront method had a mean difference of 0.04 dioptres and a standard deviation of 0.35. The 95% limits of agreement were also narrower for the AI estimate: approximately −0.67 to 0.74 dioptres, compared with −0.81 to 1.22 for the conventional instrument.[1]

The improvement is measurable but not universal

For spherical equivalent, 66.1% of the wavefront estimates were within 0.25 dioptres of subjective refraction, compared with 54.6% for traditional autorefraction: an improvement of 11.5 percentage points. Within 0.50 dioptres, the proportions were 91.3% and 86.9%, a 4.4-point difference. These comparisons describe closeness to the reference measurement, rather than children's experience wearing the resulting glasses.[1]

The researchers also measured two Jackson cross-cylinder components, J0 and J45, to represent aspects of astigmatism. Both methods were compared with subjective refraction across these parameters; the mean difference for the wavefront method was close to zero, with dispersion that still matters for individual patients. A small average error does not imply that any particular child's astigmatism is measured correctly. Prescribing decisions depend on the complete optical correction, age, symptoms and clinical context—not a single summary statistic.[1]

The design offers useful strengths for a feasibility study. It was prospective, involved masked reference assessment and included randomisation of test sequence. But it was still an agreement study at the level of instrument estimates. Children did not wear glasses prescribed independently by each method for a sustained period, and the investigators did not measure whether the new system improved reading, comfort, school participation or long-term visual development. Nor was the system tested as an autonomous service in a clinic with diverse staffing and equipment arrangements.[1]

The findings support using the model to suggest lens settings before a clinician completes the prescription. Testing spectacle wear and follow-up over time would establish whether closer measurement agreement translates into better accepted prescriptions or more efficient appointments.[1]

Report an error or suggest a correction