The present study followed the tenets of the declaration of Helsinki and was part of a series of experiments that had been approved by the local institutional review board. A non-blinded counterbalanced crossover design was chosen.
In total, 17 participants (6 males) participated after providing informed consent. All had normal or corrected-to-normal visual acuity. One participant was excluded due to excessive eye blinks, leaving 16 participants that were included in the analysis.
VEP recording and evaluation
VEP-based acuity estimates were obtained using the procedure described by Bach et al. , which has been successfully used in its original or modified version in several studies (e.g., [33,34,35,36]). For each experimental condition (i.e., for a single acuity estimate), the recording duration was typically in the order of 5–10 min, depending on the number of eye blinks, which were rejected based on a 120-μV threshold criterion. The procedure was in agreement with the respective ISCEV extended protocol . In short, steady-state VEPs were recorded to six different check sizes. The response at the first harmonic was obtained through Fourier analysis and corrected for noise , and the corresponding statistical significances were estimated . This results in a tuning curve which relates the response amplitude to the dominant spatial frequency of the stimulus. An algorithm selects the appropriate data points based on the statistical significance of the response, fits a straight line and determines the abscissa intercept, which represents the “VEP SF limit” as proposed by the ISCEV extended protocol. In the degraded vision condition, the large number of check sizes that were too small to be resolved increased the likelihood of spurious significances (multiple testing problem). To reduce this effect for the purpose of the present study, we manually corrected for this in obvious cases by defining these points as non-significant before the heuristic algorithm was applied.
A conversion factor is applied to the intercept spatial frequency with the aim of making the result numerically comparable to standard subjective decimal acuity values. In our current implementation, which is also used for clinical routine applications, the result is clipped to ≤ 1.6 (logMAR ≥ − 0.2), because larger values, far beyond the range covered by the stimulus check sizes, have a relatively high likelihood of being imprecise, while at the same time a differentiation between decimal acuity values above 1.6 is irrelevant for the typical application of the method in cases of unexplained visual loss.
For comparison, we also estimated acuity using a machine learning approach that we have recently proposed . This was only done for the condition with undegraded acuity and only after the heuristic algorithm revealed an effect of contralateral occlusion (see “Results” section) in order to test whether the effect is specific to the type of analysis after we failed to find no systematic change of the tuning curve. The machine learning approach uses a neural network that has been trained with previous VEP tuning curves and corresponding behavioral acuity data to estimate acuity from new tuning curves. Application of the machine learning approach to the data of the degraded acuity condition would not have resulted in meaningful acuity estimates because the training data set did not include acuity levels in that range.
Specific study procedure
In each participant, one eye was selected randomly as study eye and used in all experimental conditions. With this eye, VEP-based acuity estimates were obtained either with normal vision or with vision degraded by placing a diffusing filter (1° Light Shaping Diffuser, Luminit, Torrance, CA, USA) in a trial frame in front of the study eye. This filter produced Gaussian blur and reduced acuity to around 0.09 decimal acuity (logMAR = 1.06) as measured behaviorally in a recent study .
The fellow eye was either covered with an eye patch as used for amblyopia treatment (ORTOPAD, Trusetal Verbandstoffwerk GmbH, Schloss Holte-Stukenbrock, Germany; light transmission measured to be less than 1%, although transmittance of the adhesive material at the rim of the patch is somewhat higher), or was supplied with a strongly diffusing, albeit translucent occluder made from polymethyl methacrylate with light-diffusing beads embedded throughout the material (PLEXIGLAS Satinice 0D010 DF, Evonik Performance Materials GmbH, Darmstadt, Germany; thickness 3 mm, light transmission 83% as per the data sheet). This occluder was also inserted into the trial frame. Thus, the experiment involved two diffusors, one for occluding the contralateral eye (used alternatingly with the opaque eye patch), which completely nullifies any perception of shape, and one for the study eye (used solely in the degraded acuity condition), which only moderately reduces acuity. Both eyes were also supplied with the appropriate corrective lenses (individual refraction and near addition for the stimulus distance).
The rationale for choosing the two different types of occlusion for the fellow eye is that they represent the extremes of a continuum of light levels to which an occluded eye could be exposed in clinical practice. The case of a non-translucent occluder inserted into a trial frame (allowing straylight to enter from the sides) would be expected to have an effect that is in between these extremes.
The two acuity conditions in the study eye and the two occlusion conditions in the fellow eye resulted in a total of 4 conditions in a 2 × 2 design. The order of conditions was counterbalanced across participants to minimize sequence effects.
All analysis was performed with IGOR Pro 7 and 8 (Wavemetrics, Inc.). Statistical testing (repeated-measures comparison of medians) was performed with a permutation test, and confidence intervals were bootstrapped . Because the machine learning approach was only a secondary analysis, it was not included in the correction of multiple testing. At one occasion, we supplementarily assessed the mean instead of the median, as it better reflected certain characteristics of the data (see Results section) .