Including the nonlinear response of neurons to improve the prediction of visual acuity across levels of contrast, luminance, and blur

Abstract

We present a theoretical model that predicts visual acuity changes over extended ranges of stimulus contrast, luminance, and optical blur. We highlight the significance of neuronal response nonlinearity to optical contrast in achieving model agreement with experimental data. The model operates by computing, for each experimental condition, a parameter termed data separability within the framework of statistical decision theory. We assume a theoretical model observer that utilizes sharp image templates for optotype identification, consistent with our previous work for small ($<0.5$ D) optical aberrations (Leroux et al., 2024). The model incorporates the nonlinear response of visual neurons to contrast stimuli in the simulation of visual images. We digitalized measurements from Johnson and Casson (1995), who studied the combined effects of stimulus contrast (6 to 97%), luminance (0.075 to 75 cd/m${}^2$), and blur (0 to 8 D positive lens), and compared our model’s predictions to their data. The model achieved an overall root-mean-square residual of 0.048 logMAR for measurements spanning 1.73 logMAR. Accounting for nonlinearity proved critical in predicting acuity across these extended ranges of experimental conditions. This approach may also be necessary for modeling acuity under non-standard experimental conditions and/or for subjects with pathologies.