Introduction

Gaze-contingent paradigms are widely used in visual research to study visual perception and cognitive processes. In gaze-contingent paradigms, the gaze location is monitored with an eyetracker and is then used to manipulate the visual information the observer is currently looking at. In one variant of such gaze-contingent paradigms, the moving mask technique (Rayner & Bertera, 1979), foveal information is degraded in order to investigate scene and object perception (Henderson, McClure, Pierce, & Schrock, 1997; Larson & Loschky, 2009; van Diepen, Ruelens, & d'Ydewalle, 1999), eye movement patterns in visual search (Bertera, 1988; Bertera & Rayner, 2000; Cornelissen, Bruin, & Kooijman, 2005) and reading (Fine & Rubin, 1999; Scherlen, Bernard, Calabrese, & Castet, 2008), and high-level cognitive functioning such as visual context learning (Geringswald, Baumgartner, & Pollmann, 2012) when the observer is faced with a central scotoma.

The results of studies with simulated scotomata are often interpreted in the light of foveal contribution to specific visual processes or the consequences of impaired foveal vision and may further be related to visual functioning in patients with maculopathies suffering from the loss of central vision. However, it is often not reported whether and how the reliability of the scotoma simulation was tested. In the case of the disruption of an effect of interest in the scotoma, as compared with a control condition, interpretation of the results may be straightforward, indicating that foveal vision is required for a given task. However, in experiments faced with null effects in the scotoma condition, the interpretation of the outcome may be flawed if the implementation of the simulation is not tested and reported. The preferable explanation for a null effect in the scotoma condition may be that foveal vision is not necessary for or, for example, a result of adaptation of visual strategies to the task. However, the alternative explanation that needs to be ruled out in order to draw this conclusion is that the simulated scotoma fails in serving its purpose of disrupting foveal perception of visual stimuli, thus resulting in a type II error caused by the poor quality of the scotoma implementation.

One way of determining the quality of gaze-contingent scotoma simulations may lie in the measurement of the temporal accuracy that concomitantly affects the spatial precision. The end-to-end latency of the system—that is, the temporal delay between the retrieval of the eye position and the corresponding update of the visual stimulus—mainly depends on the inherent properties of the eyetracker, such as sampling rate and algorithms concerning pupil and corneal reflex detection in video-based eyetrackers, data-smoothing, or online saccade detection; on routines used for the rendering of the visual stimulus; and on the refresh rate of the monitor. The total temporal delay of these processes is difficult to measure but may be sufficiently approached by a worst-case estimate if events are synchronized and their individual duration is known. Work on gaze-contingent paradigms suggests that observers do not become aware of display changes when delays up to 12–16 ms (Inhoff, Starr, Liu, & Wang, 1998; Loschky & McConkie, 2000), 20–30 ms (McConkie & Loschky, 2002), or even 60 ms (Loschky & Wolverton, 2007) in gaze-contingent multiresolutional displays are considered. However, this does not sufficiently answer the critical question of whether the simulated scotoma effectively disrupts foveal processing. In addition to the end-to-end latency of the system, many other factors can influence the spatial accuracy of the simulated scotoma. Oculomotor events such as eye blinks or similar signal distortions may introduce periods in which the position of the scotoma cannot reliably be calculated. Furthermore, the measured eye position may be inaccurate due to unreliable detection of the pupil, especially in video-based eyetrackers, because of occlusion of the eye, low contrasts between pupil and iris, or reflections of visual aids such as glasses or lenses.

The interaction of all these factors makes it difficult to estimate the validity of the scotoma simulation even if some of the interfering parameters can be measured or controlled for. Thus, without testing whether a gaze-contingent scotoma actually results in behavioral impairments in tasks that require foveal resolution, interpreting null effects in scotoma simulations may still be flawed. Only if it can be experimentally shown that a foveally demanding task is disrupted by the scotoma simulation while the main task of interest remains unaffected by the scotoma can the conclusion that the main task does not require foveal vision be statistically supported against a possible type II error. However, if the control task requiring foveal vision produces a null effect in the scotoma condition, a null effect in the main task might be due to improper implementation of the scotoma simulation.Footnote 1 Hence, we developed a method for testing the validity of the simulated scotoma behaviorally. Our method consists of a high-acuity visual discrimination task that is performed under free viewing and compared with a scotoma condition within the same observer. If the scotoma simulation prevents foveal processing, target detection performance in the scotoma condition should be significantly impaired, as compared with free viewing. In addition, we evaluated the spatial accuracy of the eye movement recording during the experiment and measured the participants' visual acuity after they had completed the search task.

Method

Experimental paradigm

The spatial resolution of the visual system decreases with increasing distance from the fovea (e.g., Anstis, 1974; Green, 1970; Millodot, 1972; Wertheim, 1980; an overview is provided by Marmor & Marmor, 2010). While the relative visual acuity of foveal vision is equal to 1, acuity of peripheral vision rapidly decreases to 0.1 at a retinal eccentricity of 10°. In order to test the validity of the scotoma, we utilize the Landolt ring as the target stimulus that is the international standard optotype for visual acuity testing (International Organization for Standardization, 2009). According to the international standards, visual acuity is defined as the reciprocal of the smallest resolvable gap width measured in minutes of arc (MAR, "minimum angle of resolution"; visual acuity = gap of Landolt ring [′]−1). Thus, a visual acuity of 1 is given if an observer is able to discriminate a Landolt ring with a gap size of 1′. If the Landolt ring needs to be enlarged to a gap width of 2′, the visual acuity measures 0.5.

Regarding the validation of an artificial scotoma, we can thus calculate the size of a Landolt ring whose identification should be impaired if the scotoma covers the retina to a given extent. Table 1 represents the relation between eccentricity, visual acuity, and the size of the gap of the Landolt ring that should still be resolvable at the given eccentricity. If, for example, one wants to evaluate a scotoma with a radius of 4°, a gap size smaller than 0.29−1 = 3.5′ should not be resolvable by the observer if visual stimuli are efficiently blocked out. One minute of arc equals one sixtieth of 1°, which leads to a gap size of 3.5′/60′ = 0.058°. The other way around, the minimum sized scotoma necessary to constrict peripheral identification of a given Landolt gap can be calculated. The gap of the Landolt ring should be at least 2 pixels large for accurate presentation (Bach, 1997). Thus, the minimal scotoma size that can be evaluated using this method depends on the size of 2 pixels on the screen in minutes of arc.

Table 1 Relation between eccentricity, decimal visual acuity, and corresponding size of the resolvable Landolt ring gap
Full size table

Participants

Ten healthy participants (2 males, 8 females; 23 years average age) took part in the experiment. All participants were right-handed and reported normal or corrected-to-normal visual acuity. The experiments were approved by the Ethics Committee of the University of Magdeburg. Informed written consent was obtained prior to the experiments. Participants were compensated with course credits or received a compensation of Euro 6.

Apparatus

Stimulus presentation and response recording were controlled using the Psychtoolbox (Brainard, 1997; Pelli, 1997) and iViewXToolbox extension based on the EyelinkToolbox (Cornelissen, Peters, & Palmer, 2002) under MATLAB on a PC with a 22-in. Iiyama Vision Master Pro 511 CRT monitor and a MSI NX6200AX-TD512H graphics card. The monitor was 400 mm (1,280 pixels) wide and 300 mm (960 pixels) high, and the vertical refresh rate was 100 Hz. Participants viewed stimuli binocularly from a distance of 60 cm, leading to a pixel size of 0.03° of visual angle. Eye position was recorded using an iViewX Hi-Speed eye-tracking system (SensoMotoric Instruments GmbH, Teltow, Germany) with a temporal resolution of 240 Hz. Head movements were minimized using a chin and forehead rest. Participants were tested individually in a dimly lit, sound-attenuated chamber. Participants' visual acuity was measured with the Acuity Test implemented in the Freiburg Vision Test (FrACT; Bach 1996, 2007), using the standard settings implemented in the software. Participants viewed the stimuli on an LCD monitor measuring 340 mm (1,280 pixels) × 270 mm (1,024 pixels) at a distance of 3 m. The participants’ responses were recorded by the experimenter.

Stimuli and procedure

The participant’s task was to find a Landolt ring under seven closed circles and to indicate the side of its gap that was varied according to the four cardinal positions (bottom, left, top, right) by pressing one of the four arrow keys on the keyboard. All stimuli were black presented on a gray background and placed equidistantly on an imaginary circle with a radius of 9° of visual angle. The position of the Landolt ring and the direction of its gap were randomized and counterbalanced across blocks. An example of a stimulus display is given in Fig. 1.

Fig. 1
figure 1

Stimulus displays and schematic trial timeline. a Search displays were composed of one target Landolt ring and seven closed circles serving as distractors. The simulated scotoma is shaded dark for the purpose of visualization, whereas it was colored the same gray as the background in the experiment. Relations between the size and spatial position of stimuli on the left are the same as in the experiment. The section on the right depicts enlarged stimuli, which do not correspond to displays used in the experiment regarding size and spatial relation. b Schematic depiction of the sequence of stimulus presentation. After drawing the search display into the backbuffer, the current gaze coordinates are retrieved from the eyetracker and used to draw the scotoma at the corresponding coordinates. The backbuffer is made visible at the next vertical blank. This procedure is repeated for every screen refresh until a response by the participant has been recorded

Full size image

We used a Landolt ring with the smallest possible gap that could be resolved by our monitor (0.06°), reflecting a relative visual acuity of (0.06 * 60)−1 = 0.28. Given this value, we created a minimum sized scotoma necessary to constrict peripheral identification of the Landolt gap. Using linear interpolation on the values in Table 1, the gap of the Landolt ring should not be perceptible at retinal eccentricities greater than 4.18°. Thus, the scotoma was composed of an opaque area with a radius of 4.18° and a transparency gradient following the increasing slope of a Gaussian distribution with a deviation of 0.5° leading to a smooth fading out at the edges, resulting in a total scotoma radius of 4.68°. On the basis of the Landolt acuity norms, participants should thus not be able to resolve Landolt rings outside the scotoma, and potentially resolvable Landolt rings would be occluded by the scotoma. The scotoma was superimposed on the search display and colored the same gray as the background. Its position was updated with the gaze coordinates retrieved from the iViewX Hi-Speed tracking system. Raw gaze coordinates were spatially smoothed by the heuristic filter implemented in the eye-tracking software, adding a latency of one sample to the gaze data. No other additional filter algorithms—for example, for fixation or saccade identification—were implemented. The machine delay of the eyetracker was estimated to be about 10 ms. The estimated worst-case latency until the update of the display was two frames on the CRT (20 ms), adding up to a total of 30 ms. In case no gaze sample was available—for example, due to eye blinks or signal losses—the scotoma statically remained on the last known valid position until a new gaze sample became available. The timeline of stimulus presentation is illustrated in Fig. 1.

Each trial started with the presentation of a fixation cue for 1,000 ms, followed by the search display that remained on screen until the participant responded or for a maximum of 5,000 ms. Auditory feedback was provided for correct (a 2000Hz high-pitch tone) and wrong (a 500-Hz low-pitch tone) answers. After the search display, a blank was presented for 500 ms.

Before the start of the experiment, participants were calibrated using a 13-point gaze calibration. Each participant completed six search blocks each including 64 trials. The scotoma/no-scotoma viewing conditions were alternated over blocks and counterbalanced across participants for the first block. After each block, we validated the spatial accuracy of the gaze data. Thirteen points of 0.3° of visual angle were presented in succession on an imagery grid covering 23.5° × 17.6° for 2 s, and participants were asked to fixate them as accurately as possible. Participants' visual acuity was measured after completing the search experiment. One complete experimental session lasted approximately 30 min.

The MATLAB code for the experiment can be downloaded from the supplemental materials.

Gaze data analysis

Gaze data were analyzed by a custom-made Python script using a velocity-based algorithm. We used the procedures described in Nyström and Holmqvist (2010) for filtering the data with the Savitzky-Golay FIR smoothing filter (Savitzky & Golay, 1964) algorithm with second-order polynomials and a filter length of 20 ms, the removal of invalid gaze samples due to eye blinks or signal losses, and the calculation of adaptive velocity thresholds. We then identified saccade events by velocities that exceeded the adaptive peak velocity threshold. The saccade start was calculated as the first sample before the velocity peak that fell beyond the individual saccade velocity threshold and after which the velocity of the eye increased monotonically. The saccade end was calculated as the closest sample after the velocity peak that fell beyond the individual saccade velocity threshold and was the last sample of monotonically decreasing velocity. Fixations were then identified from the intersaccadic events that exceeded a duration of 100 ms.

Spatial accuracy of the gaze data was determined by calculating mean distances between the physical screen coordinates of search stimuli and the measured eye position, using the intermediate fixation task. Although participants were explicitly instructed to strictly maintain fixation at the dot presented, some participants made several fixations or anticipated the location of the consecutive fixation dot. We thus weighted the fixations by their duration and distance to the fixation dot and chose those fixations that were the longest in duration and the closest to the validation target, with a priority weighting on duration. After determining the Euclidean distances between the selected fixations and the validation points for all 13 trials, we fitted a second-order local polynomial surface to the distances at the given coordinates in order to approximate deviations between physical screen coordinates and measured gaze location across the whole display. Following this procedure, we could estimate the spatial accuracy of the gaze data at the specific positions of the search stimuli. The code for obtaining the deviations between physical screen coordinates and measured gaze location is provided in the supplemental materials.

Data exclusion

The first two blocks served as training in order to familiarize participants with the search task and were not included in data analysis. We further removed trials on which more than 20% of the gaze samples were missing due to signal losses.Footnote 2

Results

One participant’s data contained 34.4% of invalid trials in the unimpaired condition and 17.2% in the scotoma condition, leading to the exclusion of the participant from all further analysis. The average percentage of data excluded for the remaining 9 participants was 1.7% (SD = 2.9%) in the unimpaired and 0.6% (SD = 1.2%) in the scotoma conditions. There was no significant difference in the amount of data excluded between viewing conditions, t(8) = 1.178, p = .273.

Spatial accuracy of the measured gaze data

The mean deviation score across all positions and participants was 0.84° of visual angle (SD = 0.27°). The mean deviation scores for the eight stimulus positions ranged from 0.68° to 1.02°, and an ANOVA on deviations did not indicate any significant differences between the stimulus positions, F(7, 56) = 1.186, p = .326.

Response accuracy

In order to test the validity of the simulated scotoma, we carried out an ANOVA on response accuracy with the factors viewing condition (no scotoma vs. scotoma) and target position (1–8). It revealed a significant main effect of viewing condition, F(1, 8) = 203.900, p < .05, reflecting a severe impairment in stimulus discrimination when vision was impaired by the artificial scotoma. Neither the main effect of target position, F(7, 56) = 1.493, p = .189, nor the interaction, F(7, 56) = 1.255, p = .289, was significant. Thus, stimulus discrimination was equally impaired across all eight positions with the scotoma (Fig. 2). The average performance was 96.4% (SD = 2.7%) in the unimpaired condition and dropped to 43.3% (SD = 12.7%) in the scotoma condition.

Fig. 2
figure 2

Connected radial vectors representing the average response accuracy (grid labels, in percentages) of all participants at the target locations (outer labels). Accuracy was close to 100% in the unimpaired condition (blue), in which participants could foveate the target, but dropped significantly below 50% when participants were forced to rely on eccentric viewing due to the scotoma (red). The shaded area represents the standard error of the mean

Full size image

We next investigated whether participants with a higher visual acuity performed better in the discrimination task. The decimal visual acuity measured with the FrACT was converted to the LogMAR (logarithmic "minimum angle of resolution") equivalent to approximate the logarithmic metric of the visual percept corresponding to the arithmetic scale of visual acuity (Bach & Kommerell, 1998; Holladay, 1997). Decimal visual acuities ranged from 1.05 to 2.00, with a reconverted mean of 1.42. Response accuracy in the unimpaired condition did not correlate significantly with logMAR acuity, r = .26, p = .49. However, we found a significant correlation between response accuracy in the scotoma condition and logMAR acuity, r = .68, p < .05, indicating that participants with a higher foveal acuity performed better in peripheral stimulus discrimination.

Eye movement patterns

The analysis of the response accuracies clearly shows that the scotoma significantly impaired target identification when observers searched the displays with the scotoma. This should also be apparent in the eye movement data. If the scotoma prevented observers from foveating the stimuli, visual exploration of the displays should be more difficult and result in unstructured search patterns, as compared with the control condition. We expected the increased difficulty in target detection to be reflected in the number of fixations as well as the distance of the last fixation to the Landolt ring when the response was given. As compared with unimpaired search, participants were expected to commit more fixations until responding with the scotoma. In order to investigate this question, we selected only trials on which participants responded during the search interval (99.8% of data in the unimpaired and 95.0% in the scotoma conditions). On average, participants made 6.95 fixations (SD = 0.88) in the unimpaired and 8.62 fixations (SD = 1.39) in the scotoma condition. An ANOVA on the number of fixations, with the factors viewing condition (no scotoma vs. scotoma) and target position (1–8), confirmed that participants made significantly more fixations under search with the scotoma [main effect of viewing condition, F(1, 8) = 15.65, p < .05], independently of the target position [main effect of target position, F(7, 56) = 0.421, p = .885; interaction, F(7, 56) = 0.871, p = .535].

If the scotoma effectively disrupted target identification, the position of the last fixation of the trial should be placed equally often on distractors and the Landolt ring alike, whereas participants should foveate the Landolt ring under unimpaired viewing in order to discriminate the direction of its gap. This should lead to larger distances of the last fixation to the target in the scotoma condition, as compared with the control condition. The analysis of the fixation data revealed that in some cases, observers immediately moved their eyes back to the center of the screen awaiting the following trial during or slightly after pressing the response key. Thus, if the last fixation of the trial was within 3° of the center of the screen and if the distance of this fixation to the center was smaller than that of the second to last fixation, we included the latter as the critical fixation in the analysis. An ANOVA on the median distances to the target with the factors viewing condition (no scotoma vs. scotoma) and target position (1–8) revealed a significant main effect of viewing condition, F(1, 8) = 79.67, p < .05, indicating that the distance of the last fixation was significantly higher when observers searched with the scotoma. Neither the main effect of target position, F(7, 56) = 0.536, p = .803, nor the interaction, F(7, 56) = 1.215, p = .310, was significant. Without the scotoma, the median distance was, on average, 1.03°, meaning that the participants foveated the target when making their response. With the scotoma, the average median distance was 9.52°, and last fixations were distributed across the target and distractor positions (Fig. 3).

Fig. 3
figure 3

Probability density plots of the distance of the last fixation to the target in degrees of visual angle for the unimpaired condition (blue) and for the scotoma condition (red). Plotted are all trials on which the participants responded during the response interval. The arrows show the Euclidean distances of the distractors to the target, which were roughly 7°, 13°, 17°, and 18° of visual angle

Full size image

Discussion

The goal of this work was to develop and test a visual high-acuity task that can be used to examine the validity of gaze-contingent scotoma simulations behaviorally. The question of whether the scotoma effectively disrupts foveal processing is especially important in the case of null effects in scotoma simulations, since researchers will need to rule out the possibility that such a null effect was due to an improper implementation. Thus, it needs to be shown that the scotoma does have an effect in a secondary task that requires foveal resolution, such as discriminating high-acuity visual targets like the Landolt rings used in our paradigm. To this end, we measured visual discrimination accuracy and eye movement patterns in a visual search task without and with the scotoma.

Response accuracy patterns show that the scotoma severely impaired visual discrimination of the opening of target Landolt rings. Accuracy dropped from almost perfect discrimination in unimpaired viewing to approximately 43% in search with the scotoma. Since the discrimination of the target stimuli required the spatial resolution of those parts of the fovea that were blocked from visual input by the scotoma, we can conclude that the scotoma effectively interfered with foveal processing. Eye movement patterns further support this result. Under unimpaired viewing, participants foveated the Landolt ring to discriminate its gap direction in the great majority of trials. This was not the case when visual input was blocked by the scotoma. The distance of the last fixation to the target of 9.5° and its distribution across distractor positions may indicate that participants misidentified distractors as Landolt rings. Taken together, these results show that visual discrimination of fine targets was significantly reduced when foveal processing was disrupted by the scotoma. This indicates that the method we developed provides immediate evidence that the scotoma impairs foveal vision and can generally be implemented as described in order to test the effectiveness of the experimental (hard- and software) setup.

Limitations and future directions

It is important to point out that, although the data clearly show a severe disruption of foveal processing of the target Landolt rings with the scotoma, response accuracies were higher than chance performance (25%). Thus, target identification was not completely eliminated in our experiment. Furthermore, participants with a higher foveal acuity performed better at peripheral discrimination of the targets. In order to ensure that the simulated scotoma completely diminishes foveal processing, we suggest considering the following procedures in the construction of the behavioral validation task.

The use of interpolated values of eccentric visual acuity (Marmor & Marmor, 2010) can serve as a well-defined approximation to determine the minimal size of the scotoma necessary to disrupt discrimination of the Landolt rings. However, they may not be an optimal choice for all participants. The interindividual variance in acuity is especially prominent at eccentricities below 10° (Marmor & Marmor, 2010), and similarly, it has been shown that visual performance in a variety of tasks, including orientation discrimination, contrast sensitivity, and vernier acuity, can vary greatly between participants (Halpern, Andrews, & Purves, 1999). We found a positive correlation between detection performance and acuity in the scotoma condition. Thus, testing foveal acuity yielded an indication of detection performance of peripheral targets. While measurement of individual discrimination performance at a given eccentricity with an adaptive threshold protocol is the best way to determine individual test stimuli, standard tests of foveal acuity may at least hint at peripheral discrimination performance. This becomes particularly important for the evaluation of small scotomata, since differences in visual resolution of the retina are largest up to an eccentricity of 5°.

Second, the spatial inaccuracy of the specific eyetracker model should be taken into account. The average deviation between the measured and the actual gaze position was well below 1° of visual angle in our setup, indicating a good spatial accuracy of the system. However, the spatial error in eye position measuring can vary greatly within trials depending on physiological and cognitive factors, as well as the participant’s compliance during the task. For example, it has been shown that the measured gaze position using video-based eyetrackers is influenced by the size of the pupil (Drewes, Masson, & Montagnini, 2012; Wyatt, 2010), which, in turn, depends on the stimulus luminance, which might vary between and within trials but can also be influenced by task difficulty or cognitive load (e.g., Beatty, 1982; Granholm, Asarnow, Sarkin, & Dykes, 1996; Hess & Polt, 1964). The radius of the scotoma should thus be increased by the averaged spatial error of the eye-tracking system, which should be separately measured—for example, via fixation validation, as in our experiment. We note that our method for the selection of fixations used for the calculation of the spatial accuracy may be biased in favor of selecting fixations closer to the validation target, rather than sampling the overall performance of each participant. In order to avoid this problem, participants should be asked to press a button when they are fixating the target, and the fixation temporally associated with the buttonpress should be used in the calculations.

Third, the validation paradigm should be adapted to the specific needs of the intended main experiment. For example, if the filtering or distorting algorithms for the creation of the simulated scotoma are more complex than in our study (e.g., Marmor & Marmor, 2010; Perry & Geisler, 2002; Vinnikov, Allison, & Swierad, 2008), the same algorithms should also be used in the validation experiment, since they can introduce an additional temporal delay. Similarly, the presentation of large pictures can introduce delays in display presentation and should be carefully evaluated. We also suggest adapting the position of the test stimuli to corresponding positions important to the intended main experiment, since we have previously observed variable spatial accuracies of our eye-tracking system, particularly close to the corners of the screen.

Conclusion

Gaze-contingent displays offer a promising method for the investigation of visual perception. We would like to stress the merit of behavioral tests of gaze-contingent simulations. Since the scotoma simulation will always meet technical limits, such as the delay between gaze retrieval and scotoma movement or the spatial accuracy of gaze tracking, there will be a need to validate how effectively vision is impaired by the simulation. This is even more the case when additional, sample-specific factors, such as the individual physical attributes of the eye, the influence of vision aids, or individual strategies, come into play. Visual tasks in which the dependent variable is modulated in the scotoma condition, as compared with a control condition, might need a different paradigm to test the simulation, in order to control for the possible influence of other factors, such as inducing saccadic inhibition by global transients during fixation. However, we suggest running a behavioral paradigm such as the one we propose here before carrying out the main experiment, potentially in the scope of a pilot study, since it directly addresses the critical question of whether the scotoma does affect foveal perception and, thus, reduces the costs of investigating confounds post hoc in case of obtaining null effects.Footnote 3