Predicting contrast sensitivity functions with digital twins.

Area of Science:

• Ophthalmology and Vision Science
• Computational Biology
• Biomedical Engineering

Background:

• Contrast sensitivity function (CSF) is a key measure of visual performance.
• Accurate and efficient CSF assessment is crucial for diagnosing and managing visual impairments.
• Current methods for CSF testing can be time-consuming and require substantial data collection.

Purpose of the Study:

• To develop and validate digital twins (DTs) for contrast sensitivity function (CSF) prediction.
• To assess the accuracy, precision, and efficiency of DT-generated CSF predictions.
• To evaluate the potential of DTs in reducing data collection burden for qCSF testing.

Main Methods:

• Utilized a data-driven, generative model approach based on a hierarchical Bayesian model (HBM).
• Computed joint distributions of CSF hyperparameters and parameters at population, subject, and test levels.
• Generated predicted CSFs using posterior distributions from unmeasured conditions as input for the CSF generative model.

Main Results:

• DTs demonstrated high accuracy in group-level predictions across 12 tasks.
• Individual subject-level accuracy was maintained with new data, comparable to observed data.
• DT predictions could reduce data collection burden by over 50% in qCSF testing.
• DT predictions showed potential as informative priors for generating synthetic or rescoring existing qCSF data.

Conclusions:

• Digital twins show significant potential for enhancing vision assessment accuracy, precision, and efficiency.
• DTs can reduce the data collection burden in quantitative CSF testing.
• This approach supports personalized medicine by enabling more efficient and effective patient care solutions.