Abstract
Purpose :
Visual acuity (VA) and contrast sensitivity function (CSF) are distinct characteristics of visual function. When CSF is measured using gratings or filtered optotypes, CS acuity (CSA) refers to the cutoff spatial frequency that is the highest visible for grayscale targets (x-axis intercept). In principle, testing VA with black optotypes on a white background should yield similar metrics. However, due to their broad frequency spectrum, sharp-edged optotypes are not recognized solely by their fundamental spatial frequency, but also much lower frequencies (Huang et al, 2007; McAnany, et al. 2008). Many studies report that patients in early disease stages can exhibit reduced CSA and CS deficits at intermediate frequencies, when VA is normal (Wai, et al., 2022). In this study, we propose an explanation for this phenomenon, by employing machine learning to reveal the functional relationship between VA and CSF.
Methods :
After re-parameterizing the CSF model (Watson & Ahumada, 2005) with peak gain (PG), peak frequency (PF), and CS acuity (CSA), we estimated these parameters and VA from a dataset with 14 eyes tested in 4 Bangerter foil conditions with both qVA and qCSF (Zhao, et al., 2023). Next, with Gram–Schmidt orthogonalizations we extracted CSA, PF, and PG as the 1st, 2nd and 3rd components and used their z-scores as the normalized nCSA, nPF, and nPG. A polynomial regression on VA was then conducted using all linear, quadratic, cubic and quartic combinations of the normalized parameters as predictors (Figure 1). The best fitting model was determined through model comparison.
Results :
The best fitting model consisted of a linear combination of nCSA and nPF (Figure 2): VA=0.294 - 0.160 * nCSA + 0.147 * nPF, accounting for 90.4% of the variance (49.0% from nCSA, and 41.4% from nPF). The RMSE of the model prediction was 0.07 logMAR, comparable to the test-retest SD of 0.06 logMAR of the qVA data.
Conclusions :
This study provides the first quantitative functional relationship between VA and CSF, in which VA is negatively correlated with CSA but positively correlated with PF. This simple functional form can potentially explain the clinical phenomena of normal or near-normal VA but deficient CSF, although the exact functional relationship may vary and still need to be investigated in different clinical populations.
This abstract was presented at the 2024 ARVO Annual Meeting, held in Seattle, WA, May 5-9, 2024.