Abstract
We developed and validated digital twins (DTs) for contrast sensitivity function (CSF), using a data-driven, generative model approach based on a Hierarchical Bayesian Model (HBM). The HBM was trained with the trial-by-trial responses obtained from quantitative CSF (qCSF) testing of an observer population across three luminance conditions (N = 112). HBM analysis yielded the joint posterior probability distribution of CSF hyperparameters and parameters at the population, condition, subject, and test levels. A generative model, which combines this joint posterior distribution with newly available data, yields DTs that predict CSFs for new or existing observers in unmeasured conditions. The DTs were tested and validated across 12 prediction tasks. In addition to their accuracy and precision, these predictions were evaluated for their potential as informative priors that enable generation of synthetic qCSF data or rescore existing qCSF data. The HBM captured covariances at all three levels of the hierarchy, which enabled the DTs to make highly accurate predictions for individuals and group. DT predictions could save more than 50% of the data collection burden in qCSF testing. DTs hold promise for revolutionizing the quantification of vision, which can better serve assessment and personalized medicine, offering efficient and effective patient care solutions.