Abstract
purpose. To study how central visual motion integration and segmentation processes are influenced by the congruence or incongruence of peripheral contextual moving surrounds and to determine their clinical relevance.
methods. Nine subjects participated in experiments 1 and 2 (12-second blocks containing 2-second static fixation and 10-second surface plaid movement) and 15 in experiment 3 (72-second blocks, with 12-second fixation, and 60-second motion). Observers reported whether they perceived nontransparent (corresponding to visual integration of motion cues into one surface) or transparent (segmentation of two surfaces from motion cues) plaid motion within a 5° central circular region. Surround stimuli were 20° transparent or nontransparent moving plaids.
results. Contextual effects required the presence of both local and global ambiguity. If central local motion became unambiguous, then surrounds became ineffective. Under local and global ambiguity, transparent surrounds invariably induced central congruence while also strongly suppressing incongruent percepts. Nontransparent surrounds produced similar but less consistent congruent bias, especially for longer viewing times. In the latter case, however, suppression of incongruent central interpretations became barely detectable compared to the observed significant facilitation of congruent percepts.
conclusions. Local ambiguity is critical in contextual modulation, and the peripheral enhancement or suppression of central motion integration depends both on transparency bias and center-surround congruence. The importance of local ambiguity in contextual modulation is clinically relevant, because it implies that contextual effects will be stronger in disorders with impaired central vision, such as macular degeneration. Moreover, the increased efficacy of global context under conditions of increased local ambiguity may be useful in future rehabilitation approaches.
The study of center-surround interactions in visual perception is of great relevance in health and disease, in particular for understanding different forms of visual impairment in which central and peripheral vision are differentially affected, such as macular degeneration
1 2 and scotomas of cortical origin.
3 Previous literature has shown that what patients can see is not necessarily correctly anticipated from their visual fields or neurophysiological data.
3 4 5 Indeed, we have reported previously that patients with acquired or neurodevelopmental disorders could integrate coherent motion representations despite the presence of local disruption of magnocellular information processing; thus, we hypothesized that such patients were integrating contextual information over space to solve for local ambiguity.
3 4 5 Our own work on Stargardt and Best disease has in fact elucidated that local impairment is strongly dependent on how much spatial information can be integrated over space (explaining why magnocellular impairment is more confined to central vision than parvocellular impairment).
1 2
In this study, we explored the well-known concept that visual context can influence the perception of local stimuli,
6 7 8 9 10 an effect that is observed even if the experimental subject is not aware of the presence of a modulatory stimulus.
11 The most commonly explored and discussed types of center-surround interactions are the contextual sensitivity of human contrast, orientation discrimination, and vernier thresholds, because they can be directly related to neurophysiological studies in monkey V1 and also because the role of primary visual cortex in contour integration is relatively well understood. It is known that contrast detection can be improved up to 40% by suprathreshold contextual information, the effect being modulated by low-level properties such as relative orientation and collinearity.
12 Moreover, it is well established that responses of visual neurons may be markedly modulated by stimuli outside the classic receptive field (i.e., stimuli that do not themselves evoke responses of such neurons), and such modulation is dependent on the relative orientation, direction of motion, and contrast of stimuli presented in surrounding regions.
13 14 15 16 Accordingly, there is also evidence that relative-motion stimuli represent important contextual influences.
17 Most of these interactions can be explained by models that postulate contour integration mechanisms through long-range horizontal connections
9 18 19 or competition processes based on surround suppression and/or binocular rivalry.
20 21 22
However, the rules governing peripheral contextual influences on the interpretation of ambiguous central motion stimuli remain largely unexplored. Comprehending the local and global rules that constrain the facilitatory effects of visual context in solving visual integration problems is equally relevant in the understanding of normal and pathologic vision, since ambiguity in perception is a common denominator in diseases causing visual impairment. A better insight on these mechanisms can lead to an improved understanding of disease pathophysiology and also to the development of new rehabilitation strategies.
In this study, we have explored peripheral contextual influences on the perceptual competition between moving surface segmentation and integration processes. The effects of context were studied using ambiguous bi-stable stimuli: Subjects were to report their interpretation of two superimposed gratings moving in different directions (plaid stimuli, see
Fig. 1for experimental design and stimulus conditions, and the Methods section for further details). Plaids may be perceived either as two surfaces, one being transparent and sliding on top of the other (transparent or component motion) or as a single, coherent pattern whose direction of motion is intermediate to the component vectors (nontransparent or pattern motion). The degree of perceived transparency depends on the luminance of the grating intersections, the angle between movement directions, and the speed of the components.
23 24 25 Our question was whether the type of perceptual congruence of peripheral contextual surrounds could modulate suppression and enhancement of center surface motion signals, and whether local and global ambiguity had distinct roles in this process.
Experiment 1: Surround Modulation under Short Viewing Times.
Stimuli were presented in subject-initiated, 12-second blocks containing 2 seconds of fixation and 10 seconds of plaid movement
(Fig. 1a) . We defined 11 center conditions for all luminance/contrasts sets with graded levels of perceptual coherence: eight were obtained by varying the luminance of grating intersections (for details, see
Table A1in the Appendix) and three by applying local dynamic texture on the gratings
(Figs. 1c 1d) . Surround conditions were defined as either no surround (to establish perceptual ambiguity of the central stimuli per se) or as a 20° diameter moving plaid patch surrounding the central patch, having similar spatiotemporal properties as the central stimulus, thus differing only in terms of intersection luminance and/or local texture
(Fig. 1d) , yielding categories of surround stimuli biased toward either nontransparent or transparent motion. Center and surround regions were separated by a 0.25° wide red annulus of fixed size, so that they were not adjacent.
The spatiotemporal parameters of plaid movement were kept constant throughout the experiments: movement with 30% duty-cycle gratings, 1 cyc/deg spatial resolution, 1 dps velocity, and 120° difference in movement direction with upward coherent motion direction in experiments 1 and 2; and coherent movement in all the cardinal directions in experiment 3. Only luminance parameters, texture, and the presence or absence of the surround patch were varied across conditions.
The direction of motion of dots forming the textures was the following (when referring to the upward motion used in experiment 1): three directions (one vertical, two horizontal) for ambiguous textures (Txd ambiguous), two directions (both horizontal) for textures unambiguously biased for transparency (Txd component), and 1 direction (vertical) for textures unambiguously biased for (pattern) coherence (Txd patt). Note, that ambiguous textures were constructed such that 50% of the superimposed dots provided bias towards transparency and 50% towards nontransparency.
Experimental blocks were organized into three repeated runs for each of the three luminance conditions, each run containing 66 blocks in randomized presentation order.
Experiment 2: Ruling Out Patch Size as the Explanation of the Main Effect.
Experiment 3: Surround Modulation under Longer Viewing Times.
All subjects were experienced observers and were selected on the basis of fixation ability on concurrent eye movement experiments. We nevertheless ran further control eye movements experiments using plaid stimuli in eight subjects (iViewX High-speed Eye-Tracker; SMI, Munich, Germany) which allowed for data collection at 240 Hz. Data time series were searched for fixation, blinks, and saccadic events (Begaze Software; SMI) and subsequently exported to one of two standard statistical packages (SPSS; SPSS, Chicago, IL; StatView; SAS Institute, Cary, NC). Analysis of the recorded data showed that fixation could be stably held across all conditions, with very rare saccades, and reduced the number and duration of blinks, which was similar across conditions. Under the conditions of the experiment, we could therefore ascertain that the subjects did not foveate the surround region.
The continuous recording of responses allowed us to estimate the overall duration of single-percept types (transparent or nontransparent), the relative ratio of these perceptual states during the presentation periods, and the number of perceptual switches as a possible measure of percept stability. Note that the total amount of time in a given state is dependent both on the number of switches to that state and single-state durations.
In describing the overall perceptual state duration during the presentation period, we provide statistics based on the percentage of “transparent” responses, because these statistics are complementary to those based on the percentage of nontransparent responses (taking into account that % transparent + % nontransparent + % unsure = 100%), and the percentage of unsure responses, in which subjects pressed either both buttons or none of them, because their uncertainty in describing the percepts was low and stable across all conditions (see
Supplementary Fig. S1). The latter observation proved that the perceptual response patterns were indeed not influenced by the level of subject uncertainty.
We have applied ANOVA statistics (except as otherwise stated) after excluding potential violations of its statistical assumptions (including Kolmogorov-Smirnov normality verification, and Levene homogeneity tests).
In the case of experiment 3, we performed GLM (general linear model)/ANOVA repeated measures and random-effects analyses (given a relatively large number of subjects), using the subjects as the random variable. Effects were tested by using both between-subjects (with subjects entered as random effects) and within-subject comparisons. Fisher PLSD correction for post hoc comparisons was applied. The same dichotomy of contextual influence was found using within-subjects GLM and nonparametric statistical models (Friedman and Wilcoxon tests, for within-subject comparisons), thereby proving the robustness of our results and their independence of the statistical model used. In the special case of assessment of the number of percept switches as an indicator of perceptual stability we have used the Kruskal-Wallis test given the gross violations of ANOVA assumptions even after variable transformations.
Although our main analyses focused on the overall time spent in each perceptual state, it is also important to specifically analyze the stability of a single percept. In this part of the analysis, we therefore departed from considering the total amount of time spent in a given perceptual state and used the duration of single perceptual states. This approach helps in differentiating between the inherent perceptual bias and the ambiguity (in the sense of stability) of a given stimulus, thus providing an independent measure of percept stability or stimulus ambiguity. Note that a highly ambiguous stimulus with a given perceptual bias can solely differ from a similarly biased but less ambiguous stimulus in the number of perceptual switches. Accordingly, a given surround type may lengthen specific single-percept-duration events without a change in the overall summed duration, depending on the number of switches during that condition. This notion may also be clinically relevant in the study of visual impairment and ageing processes.
Moreover, response type (pattern or component) ordering of single-percept durations provides means for describing the effects of surround modulation in terms of enhancement or suppression depending on perceptual congruency and incongruency between centers and surrounds compared to the no-surround condition. In this kind of analysis, there is a signature of enhancement of congruent percepts if the average single-percept duration for the perceptually congruent surround condition is longer than that of the no-surround condition. Similarly, active suppression is verified if the average single-percept duration for the perceptually incongruent surround condition is shorter than that of the no-surround condition.
We investigated the influence of peripheral visual context on the perception of central motion. We found that contextual effects are strongly dependent on the presence of both local and global ambiguity. These findings have strong implications in the understanding of normal and impaired vision, because in the latter ambiguity is increased in general.
Natural scenes present concomitant challenges to central and peripheral vision, and it is widely known that this is of particular relevance in normal vision, as well as in neurodevelopmental disorders and diseases related to aging.
3 4 5 30 Central-peripheral interactions, in particular those concerning magnocellular and motion processing, are of strong importance both in different eye diseases and aging, where temporal sensitivity is often decreased, and center-surround visual motion antagonism reduced.
20
Our findings show that what subjects can perceive in the foveomacular regions is strongly influenced by the level of congruence/incongruence of peripheral visual information. These results extend our previous findings in patients,
3 5 showing that contextual information may be integrated over space to solve for local ambiguity even in normal vision.
We found that the dynamics of central motion integration significantly depended both on the type of motion perception in the contextual peripheral surround and on the presence of central local motion disambiguation cues. Transparently perceived surrounds (two perceived peripheral moving surfaces) evoked more consistent effects (enhancement of congruent percepts and suppression of incongruent percepts) compared with nontransparent (just one perceived peripheral surface) surrounds (which only showed a clear enhancement effect of congruent percepts, at the level of single-percept durations). This influence was stronger with textured surrounds, in line with the fact that they are inherently unambiguous. Indeed, unambiguous feature motion provided by overlaid random dots completely determines the perceived direction of local contours, and thereby provides a solution to the aperture problem.
27 This rendered textured peripheries to have a powerful contextual influence.
A remarkable property of central textured stimuli was their own resistance to contextual modulation. This is probably because their unambiguousness renders them less susceptible to be influenced by the surround. We were surprised, however, to find that even when central textured stimuli were designed to be globally ambiguous (e.g., containing bimodally disambiguated local motion; 50% of the dots providing bias toward one percept type, and 50% toward the other), surround modulation was not effective. This surprising effect is explained by the absence of local ambiguity despite the presence of global ambiguity. Indeed, whenever local motion became unambiguous, the effect of surround modulation diminished regardless of the overall luminance/contrast.
These observations have strong basic and clinical research implications. First, they suggest that the shifting balance between the coherent and transparent (two-surface) percepts are attributed to processing stages before any integration or combination of local motion signals, in agreement with previous work.
31 Second, these findings are also clinically relevant because motion may often become ambiguous in different visual disorders. Furthermore, in patients with macular diseases, central ambiguity occurs more frequently, which implies that the surround has a more powerful effect, which may be used clinically in rehabilitation approaches.
Contextual effects were similar both for short (10 seconds) and long (60 seconds) stimulus presentations, which shows that they generalize across levels of motion adaptation and were independent of pattern motion directions.
A possible locus of the described contextual modulation effects would be MT, a pattern selective region playing a key role in surface motion integration,
25 with well-described center-surround modulation properties.
16 22 However, an earlier point of interaction could be attained via an MT-V1 feedback, analogous to the one suggested by the model of Bayerl and Neumann,
32 in which localized V1 motion signals are the target of feedback modulation, by means of velocity-matching operations. Since component (local) motion selective neurons represent the overwhelming majority in V1 and V2,
33 34 35 36 37 the proposed feedback route would produce an imbalance of early response distributions producing more robust shifts toward transparent (component) than coherent interpretations on the population level. This model would thus explain the differential contextual modulation of surface integration processes for pattern (nontransparent) and component (transparent) motion conditions.
This notion is consistent with results in previous work,
38 emphasizing that integration of local motion signals across space is a relevant mechanism in vision that might be implemented by the existence of a cooperative network linking neurons sensitive to different directions/speeds and different spatial locations. Furthermore, this view is not inconsistent with the issue of spatial integration/segregation within a visual area (in fact pattern and component neurons also coexist in MT).
As stated earlier, the fact that locally disambiguated
27 but globally ambiguous (due to local bimodal motion distributions) textured stimuli could escape surround influences—possibly posed either by top-down feedback
19 39 or by collinear facilitation mechanisms
17 that were probably enabled by our experimental conditions—does in any case provide strong evidence for an early neural locus underlying surround modulation.
Given the notion that contextual effects may occur even if the experimental subject is not aware of the presence of the modulatory stimulus,
11 which is also the case when stimuli are placed peripherally, future studies should determine the role of high level top-down mechanisms in modulating such peripheral effects.
We were able to measure how congruent and incongruent context influences central visual dominance and suppression durations.
40 41 In this way, we have found further evidence of distinct mechanisms underlying pattern and component surround modulation. Component surrounds lead to a higher frequency of perceptual switches (indicating higher perceptual instability, with consequentially shorter single-percept durations), enhanced dominance of congruent stimuli, and enhanced suppression of incongruent stimuli. Pattern surrounds are associated with fewer switches (i.e., increased percept stability), and although they enhance the individual duration of congruent stimuli, they are less effective in suppressing incongruent percepts (possibly because their main effect is to increase stability of all percept types). We conclude that context may distinctly influence both dominance and suppression of single-percept durations, in parallel or not with percept stability, depending on the type of surround. Moreover, we found that the switch dynamics in our center-surround plaid displays were different from those found without contextual modulation
28 (in particular, the first percept was not always pattern).
This notion of percept stability is also relevant for future studies seeking clinical applications for patients with scotomas. These studies should also elucidate how surround effects can be integrated into reciprocal inhibition based perceptual switch models similar to the ones suggested for binocular rivalry,
42 43 which are relevant in strabismus and amblyopia research. Such a unified framework may also be useful to explain why nontransparent surrounds (resulting from perceptual integration) show such different dynamics compared with transparent surrounds (resulting from perceptual segmentation).
We do believe that these results will help develop new rehabilitation strategies that take advantage of improved knowledge of the rules governing peripheral modulation of visual foveomacular signals. Center-surround interactions may for example be stronger in diseases such as macular degenerations
1 2 where central ambiguity is increased, but effective surrounds can still be processed in the visual periphery and influence center processing.
Taken together, the results in our study extend the knowledge on low-level contextual influences on the perception of local visual stimuli
6 7 8 9 10 and integration of form and depth information in extracting surface representations in early visual areas,
44 by demonstrating a role of peripheral perceptual bias in global surface segmentation/segregation processes and the role of enhancement and suppression mechanisms in this process. Finally, our study also clarifies the relative influence of local and global disambiguation, showing that peripheral influences on motion integration in foveal vision are strongly modulated by local/global ambiguity, a knowledge that can be applied in novel approaches to low-vision rehabilitation.