Flicker adaptation improves acuity for briefly presented stimuli by reducing crowding

Abstract

Adaptation to flickering/dynamic noise improves visual acuity for briefly presented stimuli (Arnold et al., 2016). Here, we investigate whether such adaptation operates directly on our ability to see detail or by changing fixational eye movements and pupil size or by reducing visual crowding. Following earlier work, visual acuity was measured in observers who were either unadapted or who had adapted to a 60-Hz flickering noise pattern. Participants reported the orientation of a white tumbling-T target (four-alternative forced choice [4AFC], ⊤⊣⊥⊢). The target was presented for 110 ms either in isolation or flanked by randomly oriented T's (e.g., ⊣⊤⊢) followed by an isolated (+) or flanked (+++) mask, respectively. We measured fixation stability (using an infrared eye tracker) while observers performed the task (with and without adaptation). Visual acuity improved modestly (around 8.4%) for flanked optotypes following adaptation to flicker (mean, −0.038 ± 0.063 logMAR; p = 0.015; BF₁₀ = 3.66) but did not when measured with isolated letters (mean, −0.008 ± 0.055 logMAR; p = 0.5; BF₁₀ = 0.29). The magnitude of acuity improvement was associated with individuals’ (unadapted) susceptibility to crowding (the ratio of crowded to uncrowded acuity; r = −0.58, p = 0.008, BF₁₀ = 7.70) but to neither fixation stability nor pupil size. Confirming previous reports, flicker improved acuity for briefly presented stimuli, but we show that this was only the case for crowded letters. These improvements likely arise from attenuation of sensitivity to a transient low spatial frequency (SF) image structure (Arnold et al., 2016; Tagoh et al., 2022), which may, for example, reduce masking of high SFs by low SFs. We also suggest that this attenuation could reduce backward masking and so reduce foveal crowding.

Introduction

Converging anatomical and physiological evidence suggests that there are two independent visual processing pathways in the brain, with distinct physiological and functional characteristics: the magnocellular (MC) and the parvocellular (PC) pathways (Callaway, 2005; Goodale & Milner, 1992; Livingstone & Hubel, 1988; Ungerleider & Mishkin, 1982). This distinction may, however, not be as clear cut as is presented in textbooks. Psychophysical evidence suggests that PC/form-selective mechanisms can influence temporal resolution (Yeshurun, 2004) and, conversely, that MC/motion-selective mechanisms can improve form perception (Burr, 1979; Nishida, 2004; Wallis, Williams, & Arnold, 2009). 

Perceptual adaptation is a powerful psychophysical tool to probe visual function (Kohn, 2007). On the whole, prolonged exposure to a visual stimulus leads to poorer visual performance (presumably reflecting reduced sensitivity of visual neurons), but there are instances where adaptation can improve visual function. Such instances of superior visual performance are useful because, unlike poorer visual performance that can arise from non-sensory factors, they tend to have rather specific explanations in terms of sensory mechanism (Chapman & Chapman, 1973). 

Adaptation to blur (Mon-Williams, Tresilian, Strang, Kochhar, & Wann, 1998; Pesudovs & Brennan, 1993), motion (Lages, Boyle, & Jenkins, 2017), and flicker (Arnold, Williams, Phipps, & Goodale, 2016) have all been shown to improve a fundamental measure of visual function: recognition acuity for letters. For example, adaptation to spiral motion produces a motion after-effect in the opposite direction to adaptation (Anstis, Verstraten, & Mather, 1998) that can produce an illusory change in object size (either enlargement or shrinking for contracting and expanding adapters, respectively). Lages and colleagues (2017) showed that such motion adaptation can enhance the recognition of letters, a finding that they attributed to illusory size change. Recently, we used a similar paradigm to confirm the reported modest acuity improvement (around −0.04 logMAR, or two extra letters on a Snellen eye chart) but observed no association between this change and perceived target size (Tagoh, Hamm, Schwarzkopf, & Dakin, 2022). We also considered whether changes in pupil size, or fixation stability, might explain the improvement. Smaller pupils reduce incident light, thus reducing aberrations and improving spatial resolution (Atchison, Smith, & Efron, 1979). We, however, report no evidence of significant changes in pupil size following adaptation. In terms of eye movements, fixational movements and ocular drifts contribute to visual acuity (Intoy & Rucci, 2020; but for an opposing view see Packer & Williams, 1992), but we report that changes in fixation stability are unrelated to changes in acuity following motion adaptation (Tagoh et al., 2022). Finally (because target letters were always presented flanked by two distractor letters), we looked at, and ruled out, acuity change being supported by a reduction in foveal visual crowding. Specifically we showed that removing flanking letters produced only a modest improvement in acuity (0.011 logMAR) that was less than the acuity improvement induced by adaptation; thus, motion adaptation cannot improve acuity wholly by reducing crowding. 

Similar to motion adaptation, adaptation to fast flickering random noise patterns (at 75 Hz) leads to modest but significant enhancement of vernier (∼1.2 arcmin) and letter acuity (around −0.033 logMAR) (Arnold et al., 2016). This is not the case for slow (0.67 Hz) flicker. Arnold et al. (2016) proposed that flicker adaptation reduces the contribution of coarse spatial information (delivered by the fast/transient MC pathway) thereby effectively “sharpening” spatial pattern perception (by boosting the effective contribution of the PC pathway, which signals fine-scale information) and so improving acuity. We note that the stimuli used by Arnold et al. (2016) were three-letter test words, presented for 110 ms and immediately followed by a post-stimulus mask of 400-ms duration. This configuration (unlike, for example, the stimuli used by Lages et al and ourselves) is likely prone to high levels of foveal crowding. For example, Lev, Yehezkel, and Polat (2014) used a similarly backward masked stimulus (an E letter target flanked by tumbling-E’s) with interstimulus intervals (ISIs) ranging from 30 ms to 120 ms. They showed that a post-stimulus mask exacerbated foveal crowding, producing around 0.15 logMAR acuity loss (!), or (for context) over an order of magnitude more crowding than in the study summarized above (Tagoh et al., 2022). Given the high levels of crowding likely to have been present in the Arnold et al. (2016) study, it is therefore possible that the acuity changes reported following flicker adaptation might be partly attributable to changes in foveal crowding (e.g., by reducing the efficacy of the post-stimulus mask). 

In this study, we set out to investigate changes in acuity following flicker adaptation and to specifically investigate the role of crowding/masking, oculomotor activity, and pupil size. To this end, we modified the paradigm described by Arnold et al. (2016): 

• • We measured acuity with both flanked and isolated letters (before and after adaptation to flicker) to quantify the impact of crowding.
• • We measured gaze position and pupil diameter while observers performed the task.
• • Observers adapted only to fast flicker (as we were interested only in adaptation that improved acuity).
• • We used a longer adaptation period (30 seconds initially, with a 4-second top-up) consistent with our previous work (Tagoh et al., 2022).
• • We used an adaptive staircase procedure (QUEST) (Watson & Pelli, 1983) to improve efficiency of acuity estimation.

General methods

Observers

Observers were recruited from the University of Auckland staff and student body. In Experiment 1, we recruited 20 observers (nine males and 11 females), with ages ranging from 14 to 53 years (average age, 25 ± 9 years). For Experiment 2, we enrolled 35 participants, including 20 males and 15 females between the ages of 19 and 53 years (average age, 25 ± 8 years). Among the observers in Experiment 2, seven individuals had previously taken part in Experiment 1. All observers, except four individuals (including two authors), were unaware of the purpose of the study. All participants reported having normal or corrected-to-normal visual acuity in both eyes. The University of Auckland Human Participants Ethics Committee reviewed and approved the study protocol (reference no. 024231), and all participants provided informed consent before participating. The experiments adhered to the tenets of the Declaration of Helsinki. COVID-19 restrictions in place at the time of data collection meant that we had fewer participants in Experiment 1 compared with Experiment 2. 

Apparatus

Stimuli were presented on a 65-inch organic light-emitting diode (OLED) display (LG 65E6T; LG Electronics, Seoul, South Korea) with a resolution of 3840 × 2160 pixels and a screen refresh rate of 60 Hz. The display was positioned at a viewing distance of 4 meters, resulting in the screen subtending 20.27° × 11.42°. The monitor output was calibrated using luminance measurements obtained with a photometer (LS100; Konica Minolta, Tokyo, Japan), and the background luminance of the display was set to 67 cd/m². 

Observers were seated in a dimly lit room, and the head of each observer was supported by a chinrest to maintain a stable viewing position and angle. They viewed the screen monocularly using their dominant eye while the non-dominant eye was covered with an occluder comprised of an infrared pass filter. This arrangement allowed us to present stimuli monocularly, while monitoring the position of both eyes in eyetracking conditions. In Experiment 2, which incorporated eye tracking, we monitored the gaze location and pupil diameter of observers using a remote near-infrared eye-tracking device (Tobii 4C; Tobii Gaming, Stockholm, Sweden; Gibaldi, Vanegas, Bex, & Maiello, 2017), which was positioned on a tripod at a distance of 65 cm from the observer's head. The eye-tracking system was calibrated using a standard Tobii calibration process. Eye-tracking data, encompassing both eye movement and pupil diameter, were continuously recorded during each run at a sampling rate of 90 Hz. 

Stimuli

We used MATLAB R2019b (MathWorks, Natick, MA) and elements of the Psychophysics Toolbox 3 (Brainard, 1997) to create and present stimuli. 

To measure acuity, we used a white (134 cd/m²) tumbling-T acuity target (Figure 1 and Supplementary Movie S1, which shows a typical stimulus following a 4-second top-up adaptation stimulus), oriented in one of four possible directions (⊤⊣⊥⊢). In uncrowded conditions, the target-T appeared in isolation (⊤). In crowded conditions, the target letter was horizontally flanked by one randomly oriented T on either side (e.g., ⊣⊤⊢), separated from the target by an edge-to-edge spacing of half the target width. In other words, the flankers around our target stimulus were located at a spacing of 1.5× letter width from the target. This target–flanker distance is critical for our procedure because it allows us to interrogate actual foveal crowding as the primary phenomenon mediating acuity changes following adaptation to flicker rather than overlap masking, which occurs at a maximum target–flanker distances of 1.4× letter width (Pelli, Palomares, & Majaj, 2004). The height of the tumbling-T target was set to 5× its stroke width to match the dimensions of conventional Sloan letters. To determine the trial-by-trial size of the tumbling-T, we employed QUEST, an adaptive staircase procedure (Watson & Pelli, 1983), combined with a four-alternative forced choice (4AFC) psychophysical task to estimate threshold acuity. The QUEST procedure adjusted the threshold performance level to achieve a 62% correct target orientation identification. 

Figure 1.

 

The experiment protocol for the adapted condition. A 30-second flicker-adaptation phase preceded the first trial, followed by a 50-ms ISI, a 110-ms test, and then a 400-ms post-stimulus mask. Finally, a yellow fixation marker appeared that prompted the observer to respond. The second (and subsequent trials) were similar, except that the adaptation phase lasted only 4 seconds. Unadapted conditions were identical, except that the stimulus appeared immediately on each trial. For the uncrowded letter conditions, the central T-optotype target (and mask) appeared without flankers.

The flicker adaptation stimulus was a circular pattern of 60-Hz dynamic white noise, 2.48° in diameter, with individual pixels subtending 0.014 deg². During the adaptation phase of each experiment (30 seconds on the first trial, 4 seconds for top-up on subsequent trials), participants were instructed to fixate on a centrally presented 3-arcmin-diameter red disk. After the offset of the adaptor, there was a 50-ms ISI, during which the fixation marker disappeared (to prevent occlusion of the subsequently presented target). Following this, a 110-ms stimulus movie played followed by a visual mask. The mask was comprised of a series of crosses in the same configuration as the string of test tumbling-T optotype letters. After 400 ms, it was replaced by the fixation marker, and participants were instructed to report the original orientation of the central tumbling-T target. Further details regarding the individual target stimuli, presentation sequences, and variables measured in each experiment are provided in the Methods section of each experiment. 

Psychophysics procedure

We first measured acuity for isolated T's (with and without adaptation) and then measured acuity for flanked T's (with and without adaptation). Each observer did at least three runs (of 35 trials) in each of the four conditions (except for one participant where a third run for the unflanked adapted condition could not be conducted due to observer time constraints). Prior to the main testing phase, each observer performed at least two practice runs, which were not included in the final data analyzed. Unflanked conditions were, however, run first followed by flanked conditions. This was in order to further facilitate participants learning the task. The sequence of adapted and unadapted runs (within flanked or unflanked letter conditions) was counterbalanced for each observer to reduce the impact of order and practice effects on our results. We incorporated 2-minute breaks between runs to allow washout of adaptation. Observers responded at their own pace. No feedback was given during any of the tasks. The response-versus-size data were fit with a cumulative normal function to estimate threshold attained within a single set of 35 trials. The average threshold estimate across runs was reported as the final acuity estimate. The procedure for Experiment 2 was identical to that of Experiment 1 with the addition of eye tracking. 

Analysis

We used MATLAB R2019b and the Bayesian statistical framework of JASP 0.16.3 (JASP Team, 2022). To evaluate the influence of specific conditions outlined for each experiment on performance and to uncover potential associations, we used two-tailed paired t-tests and Pearson correlation tests. To present a comprehensive outline of our findings, we reported both p values and Bayes factors for the comparisons made. All acuity data analyzed in Experiments 1 and 2 met the assumption of normality according to the Kolmogorov-Smirnov test (KS limiting form). 

The Bayesian framework allowed us to quantify evidence in favor of the null hypothesis. For our analyses, we selected a prior based on the default Cauchy prior in JASP (r = √2/2). To ensure clear interpretation of our results, we adopted the classification system for Bayes factors as proposed by van Doorn et al. (2021). As per this categorization scheme, Bayes factors conveyed the following effects in support of the null hypothesis (BF₀₁)—weak (1–0.33), moderate (0.33–0.1), and strong (0.1–0.03)—or in favor of the alternative hypothesis (BF₁₀)—weak (1–3), moderate (3–10), and strong (10–30). All of the data and analyses from this project are available on the Open Science Framework (OSF) via this link: https://osf.io/4w7up/?view_only=dd9e6d0f6e3944af94fd4ad7b3da44c8. 

Experiment 1. Recognition acuity following fast flicker adaptation

We quantified acuity change observed following adaptation to flicker. To explore the role of crowding, we evaluated visual acuity with letters presented in isolation or flanked by distractor symbols. Whereas we anticipate an improvement in acuity following adaptation to flicker, if this effect operated solely by reducing crowding, then it should not occur for isolated letters. 

Methods

Twenty observers participated in this experiment. Due to the limitations posed by COVID-19 restrictions, we set out to enroll as many participants as feasible in a specified period (February 2020 to June 2021). 

Results

Figure 2 illustrates the impact of flicker adaptation on visual acuity. For flanked letters, 80% of observers (16/20) showed an improvement in acuity following adaptation to flicker (Figure 2A). Our statistical analysis revealed that adaptation to flicker significantly improved acuity compared to baseline unadapted acuity: adapted, 0.011 ± 0.12 logMAR; unadapted, 0.049 ± 0.13 logMAR; mean difference, −0.038 ± 0.063 logMAR; t(19) = 2.87; p = 0.015; BF₁₀ = 3.66. In contrast, for isolated letters, only 55% of participants showed an improvement in acuity following adaptation to flicker (Figure 2B). These improvements were not statistically significant: adapted, −0.122 ± 0.13 logMAR; unadapted, −0.114 ± 0.14 logMAR; mean difference, −0.008 ± 0.055 logMAR; t(19) = 0.69; p = 0.5; BF₁₀ = 0.29. 

Figure 2.

 

Impact of flicker adaptation on visual acuity. Gray bars represent mean acuity across all observers in each condition. The error bars denote ±1 SEM, and each pair of colored discs represents data for one participant. Acuity is measured in logMAR: higher values on the ordinate represent poorer acuity, and lower values better acuity. (A) For flanked/crowded T targets, flicker adaptation improved acuity (by around two letters on a Sloan chart; mean gain, −0.038 logMAR; p = 0.015). (B) Removing the flankers produced an overall improvement in unadapted performance (mean gain, 0.16 logMAR; compare the first bars of parts A and B). However, acuity improvement derived from adaptation for unflanked (isolated) letters was not statistically significant (mean gain, −0.008 logMAR; p = 0.5).

These results suggest that flicker adaptation could at least in part be improving acuity for these rapid/masked stimuli by reducing crowding. If this is the case, then there are several means by which this improvement could arise. First, flicker adaptation could shrink the crowding interference zones surrounding the target (Toet & Levi, 1992). Second, the adaptor could reduce the impact of the post-stimulus mask, effectively sustaining the target and supporting superior recognition (Levi, 2008). Either way, if changes in crowding are supporting improvement from flicker adaptation, then (a) our stimulus should induce substantial crowding (i.e., acuity without adaptation should be better for isolated compared to flanked letters), and (b) there should be an association between acuity improvement and individual susceptibility to crowding (people who are more susceptible to crowding should show more benefit from adaptation). 

Figure 3 plots susceptibility to crowding under unadapted conditions (plotting conventions are as in Figure 2), susceptibility to crowding under adapted conditions, and individual susceptibility to crowding (unadapted) versus acuity change following adaptation. Comparing unadapted flanked and unflanked acuity, we observed significant crowding: unadapted flanked letters, 0.049 ± 0.13 logMAR; unadapted unflanked letters, −0.11 ± 0.14 logMAR; mean difference, 0.16 ± 0.11 logMAR; t(19) = −6.42; p < 0.001; BF₁₀ = 9.0 × 10³ (Figure 3A). The disruptive effect of flankers corresponds to a loss of around eight letters on a Sloan chart. Similarly, we observed significant crowding when comparing flanked and unflanked letters under adapted conditions: adapted flanked letters, 0.011 ± 0.12 logMAR; adapted unflanked letters, −0.122 ± 0.13 logMAR; mean difference, 0.13 ± 0.09 logMAR; t(19) = −7.03; p < 0.001; BF₁₀ = 1.7 × 10⁴ (Figure 3B), where the disruptive impact of crowding corresponds to a loss of around 6.6 letters on a Sloan chart. 

Figure 3.

 

Susceptibility to crowding in our experiment. (A) Unadapted visual acuity for isolated versus flanked letters; plotting conventions are as shown in Figure 2. Mean visual acuity was significantly better when measured with isolated compared to flanked letters (mean change, 0.16 logMAR). (B) Adapted visual acuity for isolated versus flanked letters; mean visual acuity was significantly better when measured with isolated compared to flanked letters (mean change, 0.13 logMAR). (C) The correlation between acuity change following adaptation to flicker and susceptibility to crowding was statistically significant (r = −0.58; p = 0.008). Each solid disc represents a plot of average acuity change versus mean change in susceptibility to crowding for each individual, and the shaded area indicates a worsening of acuity. Positive numbers on the abscissa denote greater susceptibility to crowding. The dotted line shows the trend line, and the error bars represent ±1 SEM.

The impact of crowding can also be quantified in terms of crowding ratios (flanked divided by unflanked size threshold). We did this for each pair of unadapted or adapted size thresholds and report the average. Under these conditions, the mean crowding ratios were 1.45 and 1.36, respectively. The sets of unadapted or adapted crowding ratios are not statistically different from one another, t(19) = 1.34, p = 0.2, BF₁₀ = 0.5. Thus, although we did not observe significant acuity reduction with uncrowded stimuli but did with crowded stimuli, the ratio of crowded to uncrowded performance was not statistically different following adaptation (this is discussed further below). As predicted, however, there is a robust correlation between acuity change following flicker adaptation and individual differences in susceptibility to crowding, r(20) = −0.58, p = 0.008, BF₁₀ = 7.70 (Figure 3C). Consequently, observers who experienced a greater magnitude of crowding gained a greater acuity benefit from flicker adaptation. 

Experiment 2. Fixation stability and pupil diameter following flicker adaptation

We have confirmed that acuity (for rapidly presented flanked letters) improves modestly following adaptation to flicker (Arnold et al., 2016) and that participants who showed greater crowding benefited more from adaptation. We next consider if changes in fixation (Intoy & Rucci, 2020) or pupil size (Atchison et al., 1979) could contribute to this effect. If the improvements in acuity that we observed following flicker adaptation stem (at least in part) from fixational and/or pupillary behavior, then we anticipate the following outcomes:

• • Adaptation will improve fixation stability, increase drift curvature, and reduce both the speed of drift eye movements and the distance covered by the eye, as these changes will support superior acuity performance (Intoy & Rucci, 2020). Individual changes in fixation characteristics should also correlate with acuity change.
• • Flicker adaptation will reduce pupil diameter, which has been shown to enhance visual acuity (Atchison et al., 1979). Individual changes in pupil size will correlate with acuity change following flicker adaptation.

Methods

Thirty-five observers ages 19 to 53 years (mean age, 25 ± 8; 20 males; 15 females) were recruited from the University of Auckland staff and student body, seven of whom had participated in previous Experiments. All but four were naïve to the purpose of the experiment. 

Apparatus and stimuli

Experimental setup was similar to Experiment 1, except we also measured gaze position while observers performed the task. 

Psychophysical paradigm

The experimental procedure closely resembled that of Experiment 1, including the 4AFC design, 35 trial setups, and QUEST staircases. All targets were flanked. 

Analysis

Eye tracking data were preprocessed by first disregarding data from the first 567 ms (34 frames) of each run. This aimed to eliminate erroneous data collected while the observer located the fixation dot. Next, we filtered out blinks by detecting instances where eye position data were absent. Additionally, any trials that commenced or concluded with a blink or where more than a quarter (25%) of the captured gaze position estimations were affected by blinks were excluded from the analysis. Data were then broken up into sequences punctuated by blinks before being subjected to analysis. 

Within each sequence, we used a third-order, low-pass Savitzky–Golay polynomial filter (Savitzky & Golay, 1964) to smooth the x and y eye-position data. The speed of eye movement was the product of the difference between two consecutive gaze positions and the sampling frequency of the eye tracker. We then used speed estimates to partition gaze-position estimates into sequences of drift (i.e., eye movements with speed <10°/s) and microsaccades (i.e., eye movements with speed >10°/s). 

Fixation stability was estimated with a bivariate contour ellipse area (BCEA) formula (Crossland, Sims, Galbraith, & Rubin, 2004), which estimates the area of the ellipse that encompasses a given proportion of the gaze position data:  

\begin{eqnarray} {\rm{BCEA}} = 2k\pi {\sigma _H}\sigma_{v}\sqrt {1 - {\rho ^2}}\quad \end{eqnarray}

(1)

where σ_H and σ_V are the standard deviations of gaze point locations over the x and y dimensions, respectively, and ρ the product-moment correlation of x and y positions; k is a constant and can be derived from  

\begin{eqnarray} P = 1 - {e^{ - k}}\quad \end{eqnarray}

(2)

where P is the probability that fixation points will lie within the ellipse, and e is the natural logarithm. We used a k value of 1.14 corresponding to a P value of 0.68, consistent with previous research (Crossland et al., 2004). We estimated mean BCEA for each trial separately and averaged across trials (within each run for each condition and observer). Mean pupil diameter was estimated directly by the Tobii device for each trial and averaged across all trials for all runs for each condition and observer. 

We also quantified the curvature of drift eye movements (Intoy & Rucci, 2020). For this analysis, the sequence of angles between adjacent eye positions was first calculated and then the angular difference between these measures was calculated using the MATLAB function angdiff. We calculated drift curvature as the absolute value of the angular subtense between adjacent gaze positions, and drift distance was calculated as the total distance traversed by the eye in the course of a given trial. All measurements were estimated for each trial and averaged across trials for all runs in a given condition and for each observer. 

Prior to conducting the analysis, we examined such measures derived from eye-tracking data for adherence to the normality assumption. Any data that deviated from the normal distribution assumption were subjected to a log transformation to align with this requirement. To ascertain the presence of any differences across conditions we used paired-samples t-tests. 

In our approach, we incorporated Bayesian inference, which enabled us to quantitatively assess the evidence presented by our data in favor of either the null or alternative hypotheses. In exploring the variables associated with changes in acuity, we calculated the Pearson correlation between alterations in acuity and changes in other metrics related to eye movement. 

Results

Figure 4 plots fixation stability (Figure 4A) and pupil size (Figure 4C) and their relationship to visual acuity (Figures 4B and 4D). We again report a significant improvement in visual acuity following adaptation to flicker: unadapted, 0.11 ± 0.16 logMAR; adapted, 0.030 ± 0.16 logMAR; mean difference, −0.080 ± 0.060 logMAR; t(34) = 7.78; p < 0.001; BF₁₀ = 2.61 × 10⁶. Note that this is a larger improvement than reported in Experiment 1. In Experiment 2, seven out of the 35 observers had already participated in Experiment 1, possibly gaining more experience with the task. Moreover, each observer in Experiment 2 conducted fewer trials (a minimum of 210 except one observer) compared with Experiment 1 (a minimum of 420 except one observer), potentially reducing the impact of observer fatigue on the results. 

Figure 4.

 

Effect of flicker adaptation on fixation and pupil size. (A, B) Plotting conventions are as depicted in Figure 2. Fixation stability is somewhat poorer (quantified by the BCEA measure) following adaptation to flicker, but there was no association between individual differences in BCEA and acuity change. (C) Comparison of pupil size before and following adaptation to flicker; pupil size was significantly smaller following adaptation. (D) However, there was no significant link between individual differences in pupil size and acuity change.

Although we anticipated that flicker adaptation would improve fixation stability (lower BCEA score), we found no significant enhancement in fixation stability (larger BCEA score) following adaptation: mean difference, 0.24 ± 0.71 deg²; t(34) = −2.04; p = 0.05; BF₁₀ = 1.13 (Figure 4A), with the Bayes factor providing only weak support. We found no association between fixation stability and better acuity following adaptation to flicker: r(35) = 0.04; p = 0.84; BF₁₀ = 0.21 (Figure 4B). These results suggest that fixation stability has negligible impact on the acuity gains reported here. 

Observers’ pupil size was significantly reduced by adaptation to flicker. Specifically, pupil size reduced from an unadapted diameter of 5.12 ± 0.89 mm to an adapted diameter of 4.88 ± 0.92 mm. Evidence from our data strongly support the observed pupil size reduction following adaptation to flicker: mean difference, −0.25 ± 0.35 mm; t(34) = 4.23; p < 0.001; BF₁₀ = 159.68 (Figure 4C). However, we did not find compelling evidence for a link between changes in pupil size and acuity change following adaptation to flicker: r(35) = −0.24; p = 0.17; BF₁₀ = 0.53 (Figure 4D). 

Although we anticipated flicker adaptation might increase drift curvature, our analysis did not reveal a statistically significant increase in drift curvature: unadapted, 1.45° ± 0.12°; adapted, 1.39° ± 0.21°; mean difference, −0.061° ± 0.19°; t(34) = 1.96; p = 0.058; BF₁₀ = 1, with the Bayes factor predicting a weak effect. However, neither drift speed: unadapted, 9.97 deg/s ± 8.22 deg/s; adapted, 9.95 deg/s ± 8.07 deg/s; mean difference, −0.024 deg/s ± 4.82 deg/s; t(34) = 0.03; p = 0.98; BF₁₀ = 0.18, nor drift distance measurements: unadapted, 31.47° ± 20.29°; adapted, 35.65° ± 32.0°; mean difference, 4.22° ± 24.95°; t(34) = −1.00; p = 0.33; BF₁₀ = 0.29, displayed significant differences between the unadapted and flicker-adapted conditions. 

Furthermore, none of these metrics displayed a strong association with changes in visual acuity: drift curvature, r(35) = −0.21, p = 0.88, BF₁₀ = 0.42; drift speed, r(35) = −0.16, p = 0.36, BF₁₀ = 0.32; or drift distance, r(35) = −0.24, p = 0.17, BF₁₀ = 0.53. 

On the suggestion of an anonymous reviewer, we tested our datasets for homoscedasticity, revealing that eye-movement curvature and distance data had unequal variances. As a result, we ran a two-sample t-test which does not assume equal variances, and (in agreement with the previous test) we found no statistically significant difference in curvature (p = 0.13) and distance (p = 0.56) of eye movement following adaptation to flicker. 

Taken together, these results do not support our hypothesis that alterations in fixational eye movements or pupil size contribute significantly to changes in visual acuity arising from flicker adaptation. 

Discussion

Adapting to dynamic stimuli, such as the rapid motion of spiral/circular grating or elements within a random noise flicker pattern, can enhance the recognition of spatial form (Arnold et al., 2016; Lages et al., 2017; Tagoh et al., 2022). Here, as in our earlier work on motion adaptation, we set out to investigate the influence of eye movements, pupil size, and crowding on acuity change arising from flicker adaptation. Overall, our experiments confirm the finding by Arnold et al. (2016) that exposure to flickering stimuli leads to a modest but significant enhancement of acuity for briefly presented (masked) crowded stimuli. Like our earlier research on motion adaptation (Tagoh et al., 2022) we again show that neither fixational eye movements nor pupil size contribute to acuity enhancement resulting from adaptation. Unlike our earlier work, we show here that a reduction in crowding does contribute to such acuity gains when we use post-stimulus masked targets. 

Acuity gain following adaptation to flicker is not inherited from changes in oculomotor activity or pupil size

Intoy and Rucci (2020) showed that decreased drift distance and slower and more curved drift eye movements made during performance of a high-acuity task improve visual acuity. Our results are inconsistent with such factors contributing to any improvements in acuity arising from flicker adaptation. First, rather than an improvement, we demonstrated a significant impairment in fixation stability following adaptation to flicker. Second, flicker adaptation did not alter the specific metrics of oculomotor activity in a manner that supports improved visual acuity. Although the precision of our eye-tracking device is poorer than that of Intoy and Rucci (2020), our estimate of fixation stability (0.24 deg²) is reliable and consistent with a previous estimate (0.25 deg²) for an identical group of observers using the same eye-tracking hardware (Tagoh et al., 2022). Intoy and Rucci (2020) registered large changes in their eye-tracking metrics when observers switched from fixation to recognition tasks. Although our analysis did not reveal any links between changes in these metrics and changes in acuity, we cannot rule out the possibility that subtle changes in these metrics (unmeasurable by our eye-tracking system) contributed to the subtle acuity gains we observed. What is clear is that this cannot be the whole story, as we observed a reliable association between changes in acuity and susceptibility to crowding. 

Regarding pupil size, we estimate that in order to achieve an improvement in acuity comparable to the magnitude of acuity enhancement observed in Experiment 2 (∼−0.08 logMAR), observers’ pupil size must decrease from an unadapted state of ∼5.12 mm to an adapted state of ∼2.63 mm (Atchison et al., 1979). Although we found a significant reduction in pupil diameter (−0.25 mm), it is an order of magnitude less than the change (−2.49 mm) required to produce a corresponding improvement in acuity or to be associated with acuity change. 

Magnitude of acuity change arising from adaptation or from crowding

In Experiments 1 and 2, flicker adaptation led to a substantial acuity improvement of ∼−0.04 and ∼−0.076 logMAR, respectively. These reliable gains are consistent with those reported in earlier studies using motion (around −0.04 logMAR) (Lages et al., 2017; Tagoh et al., 2022) and blur adaptation (around −0.04 to −0.12 logMAR) (Cufflin, Mankowska, & Mallen, 2007; Ghosh, Zheleznyak, Barbot, Jung, & Yoon, 2017; Mon-Williams et al., 1998; Pesudovs & Brennan, 1993; Poulere, Moschandreas, Kontadakis, Pallikaris, & Plainis, 2013). Here, we observed high levels of foveal crowding (0.16 logMAR) that exceeded the gains observed following adaptation, leaving open the possibility that acuity gains could originate from crowding reduction. This high degree of foveal crowding is consistent with an estimate of 0.15 logMAR measured with similar stimuli (Lev et al., 2014) but not with our own work with unmasked targets (∼0.011 logMAR) (Tagoh et al., 2022). 

It is likely that at least some of the interfering effect of the flankers on recognition of our target stimuli arises from crowding and not, for example, overlap masking. This is because our target–flanker distance (1.5 letter widths) is larger than the target–flanker distance at which overlap masking occurs in the fovea (1.4 letter widths or less). Additionally, because crowding zones are smaller and acuity limits are hit before crowding limits in the fovea, it is difficult to study foveal crowding with regular Sloan letters because they must be 4–5 arcmin in size at threshold (Jackson & Bailey, 2004). If the Sloan letter has to be this large for observers to be just able to recognize it (i.e., relatively large compared to other types of target), then positioning flankers to avoid overlap masking means that flankers are likely to fall outside the tiny 6- to 8-arcmin crowding zones (Pelli et al., 2016). This was not the case in our experiment, as our target at threshold was smaller (about 3.88 arcmin), likely because we used a 4AFC tumbling-T task, which is less complex than a 10AFC Sloan recognition task and thus can be performed at smaller letter sizes. 

Individual differences in acuity change are associated with changes in susceptibility to crowding following adaptation to flicker

In addition to high levels of foveal crowding, we found a significant correlation between susceptibility to foveal crowding and changes in visual acuity following adaptation. What we did not do is measure the effect of flicker adaptation without a post-stimulus mask to ascertain the magnitude of crowding contributed individually by either the flanking letters or the post-stimulus mask. Because the magnitude of foveal crowding we observed here results from both target flankers and the post-stimulus mask, we speculate that the individual contributions of the target flankers or the post-stimulus mask to this foveal crowding will be smaller compared to the magnitude of crowding we observed here. Moreover, an experimental design that excludes the post-stimulus mask constrains our ability to ensure that the target is presented for only a brief duration, as we were able to achieve with our experimental design. Our account of a contributory role for foveal crowding in flicker-induced acuity gains is further strengthened by the finding of a statistically significant impact of crowding when we compute crowding ratios within both unadapted (Figure 3A) and adapted (Figure 3B) conditions. We note that the comparison between crowding ratios for adapted and unadapted conditions is not statistically significant. This is likely because the observers’ unflanked adapted acuity performance was so good that it approached another limit (e.g., optical), precluding further acuity improvement. At baseline, then, our stimulus could be inducing such high levels of crowding that, even when adaptation reduces crowding sufficiently to explain the acuity, enough crowding remains to prevent our reporting a statistically significant reduction in crowding. This result does not argue against our view that motion adaptation improves acuity partly through a reduction in foveal crowding, but rather suggests that adaptation-induced acuity improvement likely results from a combination of factors, including a reduction in the contribution of coarse-scale image structure relayed by fast/transient MC mechanisms (Arnold et al., 2016; Mon-Williams et al., 1998; Tagoh et al., 2022) and a reduction in foveal crowding. 

Mechanism

Crowding (and not acuity per se) limits human peripheral vision (Levi, 2008). That is to say, the distance of a flanking to target letter at which one first observes a breakdown in target recognition is greater than the size of a target that is legible in isolation (Bouma, 1970). This is not the case in the fovea, where the distance of a flanking to a target letter at which one first observes a breakdown in target recognition is smaller than the size of a target that is legible in isolation. In the fovea, then, acuity and not crowding limits recognition (Danilova & Bondarko, 2007) in contrast to the periphery. Recent studies have successfully measured foveal crowding, albeit at smaller spatial extents (Coates, Levi, Touch, & Sabesan, 2018; Lev et al., 2014), and others have likened foveal crowding to masking (Chung, Levi, & Legge, 2001; Levi, Klein, & Hariharan Vilupuru, 2002). 

With respect to the mechanisms underlying gains in acuity following adaptation to flicker, our results taken together demonstrate that it is unlikely that flicker adaptation improves acuity only by reducing spatial crowding in the fovea, as the magnitude of crowding is almost 3× and 2× the gains measured in Experiments 1 and 2, respectively. 

Lev et al. (2014) argued that, based on a two-stage model of crowding (Neri & Heeger, 2002)—where targets are located first by larger V2/V3 receptive fields and then identified (after about ∼100 ms) using smaller receptive fields in V1—recognition of a crowded target requires sufficient processing time. This means that limiting stimulus availability with a post-stimulus mask (e.g., Arnold et al., 2016, and the current study) will increase crowding, the effects of which can be ameliorated by increasing stimulus availability, such as by increasing stimulus duration (Lev et al., 2014). We argue that, under unadapted flanked target conditions, the post-stimulus mask interferes with target recognition, reducing available processing time and exacerbating foveal crowding/masking. Research by Petry, Grigonis, and Reichert (1979) found that the impact of meta-contrast masking was reduced following adaptation to flicker. Assuming flicker adaptation has a similar impact on backward masking, it is possible that flicker adaptation reduced the efficacy of the post-stimulus mask in our experiment by decreasing sensitivity to the MC transient mechanisms that respond to the mask. This should improve stimulus availability and so reduce foveal crowding. 

Conclusions

We have confirmed that adaptation to flicker enhances visual acuity, particularly in the case of briefly presented flanked (masked) targets. Our experiments have systematically eliminated potential contributions from fixational behavior and pupil size. We conclude that a reduction in crowding/masking effects, stemming from both the flanked target stimuli and the post-stimulus mask signals, by flicker adaptation contributed to the reported increase in acuity. However, further studies are required to untangle the distinct contribution of the flankers or the post-stimulus mask to the reported effect. 

Acknowledgments

The authors thank Soheil M. Doustkouhi, Catherine Morgan, Tony Han, and Aryaman Taore for their suggestions. This research was supported by a grant from the Marsden Fund Council from government funding, managed by Royal Society Te Apārangi (3716355). 

Commercial relationships: none. 

Corresponding author: Steven C. Dakin. 

Email: [e-mail]. 

Address: School of Optometry & Vision Science, The University of Auckland, Auckland 1142, New Zealand.