Perceptual consequences of centre-surround antagonism in visual motion processing

Transcription

1 1 Perceptual consequences of centre-surround antagonism in visual motion processing Duje Tadin, Joseph S. Lappin, Lee A. Gilroy and Randolph Blake Vanderbilt Vision Research Center, Vanderbilt University, st Ave S, Nashville, TN Intuition tells us that increasing the size of a moving object should make its motion more visible. The receptive fields of neurons throughout the visual system, however, often exhibit centre-surround antagonism, which inhibits the responses of such neurons when stimulated with increasingly larger objects 1. Here we report that increasing the size of a high-contrast moving object can make its motion very difficult to perceive. We further show that this counterintuitive finding is consistent with receptive field properties of neurons in the middle temporal visual area (MT) and provide converging evidence that impaired motion discrimination for large stimuli is a perceptual correlate of centre-surround antagonism in MT 2-4. The spatial suppression of motion signals observed at high contrast gradually changes to spatial summation as contrast decreases, indicating that integration of motion signals over space crucially depends on stimulus properties. In this way, the visual system processes motion efficiently and adaptively by employing computationally advantageous inhibitory mechanisms only when the sensory input is sufficiently strong to ensure visibility. When evolution stumbles onto an adaptive biological implementation, that implementation often appears in other, related applications. In the case of vision, one such adaptation is centre-surround receptive field organisation, a ubiquitous property in mammalian visual systems presumably tailored for extracting image features that are differentially distributed over space 1. In motion perception, this is evident as an antagonistic interaction between centre and surround regions of the receptive fields of most neurons in MT 2-4 a cortical region importantly involved in motion processing. A centre-surround MT neuron responds well to motion localised within the central region of its receptive field, but its response is gradually attenuated as the spatial extent of stimulation expands to include the surround region. These neurons are believed to be critically involved in perceiving object motion 3,5, and probably play a role in figureground segmentation 5-7 and shape-from-motion processing The surround inhibition enables MT neurons to perform these important computational tasks, but is there a perceptual downside to the decreased activity of centre-surround neurons when stimulated with a large moving object? Intuition leads one to expect that increasing the size of a moving stimulus should improve its visibility, rather than impair it as the decrease in the activity of centresurround neurons suggests. Indeed, increasing the size of a low-contrast moving stimulus does enhance its visibility 11,12, presumably owing to spatial summation. While psychophysical estimations of receptive field properties tend to be based on lowcontrast or noisy stimuli, physiological descriptions of centre-surround motion neurons have been obtained with high-contrast moving stimuli. Moreover, in visual cortex, the nature of centre-surround interactions is often dependent on contrast, with surround suppression stronger at high contrast and spatial summation more pronounced at low

2 contrast Thus, threshold contrast measurements may not fully describe the spatial properties of human motion perception, especially at high contrast. With this in mind, we devised alternative approaches for measuring motion sensitivity. In our first experiment, we measured the threshold exposure duration required for human observers to accurately discriminate motion direction of a drifting Gabor patch. In separate conditions, observers viewed foveally presented Gabor patches of various sizes and contrasts. Spatial frequency and speed were fixed to 1 cycle/ and 2 /s, respectively. The contrast of a Gabor patch was ramped on and off with a temporal Gaussian envelope, allowing the presentation of brief motion stimuli. The spatial integration of motion signals importantly depends on contrast (Fig. 1). At low contrast (2.8%), duration thresholds decreased with increasing size, reaching a lower asymptote around 40 ms (Fig. 1a, c). This result, implying spatial summation of motion signals, is consistent with earlier reports 11,12. At high contrast, however, duration thresholds increased four-fold as the Gabor patch width increased from 0.7 to 5. This result is highly surprising and implies neural processes fundamentally different from spatial summation. Duration threshold elevation was greatest for Gabor patches larger than 2.6 in width. At this size, the Gabor patch is large enough to impinge on the surrounds of MT neurons with foveal receptive fields (Fig. 1d). We speculate that surround inhibition is responsible for the observed decrease in motion sensitivity and explore this hypothesis in the experiments that follow. It is intriguing that increasing stimulus contrast dramatically changes the way motion signals are integrated over space. Re-plotting the data as a function of contrast (Fig. 1b) reveals that, for small Gabors, the duration thresholds decreased with increasing contrast. For large Gabors, however, duration thresholds increased more than three-fold as the contrast increased from 2.8 to 92%. This result is compatible with the notion that the relative strength and/or spatial extents of the excitatory centre and inhibitory surround change with contrast At high contrast, the computational benefits of surround suppression 5-10 probably outweigh the obligatory decrease in neuronal activity and reduced sensitivity. At low contrast, high sensitivity is essential, so it makes functional sense that receptive field organisation shifts from surround suppression to spatial summation. Our data suggest that the transition from summation to suppression occurs around 5% contrast (Fig. 1c), which is the contrast where an average MT neuron attains about 25% of its maximum response 17. The results reported here are robust. We obtained similar patterns of results with faster moving stimuli (8 /s) and when the spatial bandwidth was held constant by scaling spatial frequency with Gabor width (1 cycle/s). Next, we manipulated the effective stimulus contrast by adding variable amounts of dynamic noise to a fixed contrast (46%) Gabor patch. The results (Fig. 2a) indicate spatial summation when the stimulus is noisy (46% noise contrast). In contrast, when noise is less salient (or absent), increasing the size results in poorer performance. For large Gabor patches, increasing the noise actually improves visibility of stimulus motion. 2

3 Motion strength of a periodic Gabor patch may also be varied by adjusting the magnitude of an abrupt phase shift 18 : increasing phase shifts from 0 to 90 enhances the visibility of motion. We conducted an experiment where observers identified the motion direction of a fixed contrast Gabor patch that abruptly shifted in phase in the middle of a 100 ms presentation interval. Threshold phase shift was obtained for a range of contrasts and sizes with spatial frequency fixed at 0.5 cycles/. Results (Fig. 2b) replicated the duration threshold findings, again demonstrating the spatial suppression of motion signals. There is an ongoing debate about the role of MT in the analysis of isoluminant moving stimuli (i.e. varying in hue rather than luminance contrast) 19-21, but it is clear that MT s response to such stimuli is typically attenuated 20. Given a weak response of MT near isoluminance, we may expect spatial summation of isoluminant stimuli similar to that observed for low-contrast stimuli. Using methods similar to those of the previous experiment, we compared phase shift thresholds for isoluminant (red-green) and luminance contrast (yellow-black) stimuli for a range of Gabor patch sizes. Luminance contrast stimuli showed surround suppression similar in magnitude to that in the previous experiments (Fig. 3a). Thresholds for isoluminant stimuli, however, decreased with increasing size, exhibiting spatial summation. This result is also unexpected because it shows that, for large stimuli, colour-defined motion is perceived more accurately than luminance-defined motion, presumably because the latter is affected by surround suppression. This superiority of colour-defined motion contrasts with previous reports indicating that perception of motion at isoluminance is generally less reliable 19. Surround inhibition in motion-sensitive neurons is characterized by a decrease in neural activity in response to a large moving stimulus. This decreased activity may be evident in weaker adaptation of motion-sensitive centre-surround neurons during prolonged exposure to motion. The motion aftereffect (MAE) an illusory perception of motion following prolonged adaptation to motion is thought to reflect the adaptation of motion-sensitive neurons 22. As adaptation is associated with neuronal activity, our results suggest that adapting to a large high-contrast moving stimulus should result in a weaker MAE. We explored this prediction by inducing MAE with moving Gabor patches of various sizes and contrasts and testing for MAE with a small test stimulus. As predicted, MAE strength decreased with increasing size when observers adapted to a high-contrast Gabor patch, suggesting spatial suppression (Fig. 3b). This result is consistent with earlier observations 23,24 that were restricted to highcontrast adapting stimuli. On the other hand, when a low-contrast adapting stimulus was used, MAE strength increased with increasing size. These results support the notion that neurons involved in perceiving motion direction are inhibited when stimulated with a large, high-contrast moving stimulus. Contrast dependency of this effect again indicates that spatial properties of motion perception are dynamic and stimulus-dependent. A prominent attribute of MT receptive fields is their increasing size with retinal eccentricity 4,25. Similarly, the critical size beyond which surround suppression begins to affect neuronal response increases with eccentricity, although not as rapidly as receptive 3

4 field size 4. Thus, we would expect that the detrimental effect of stimulus size should diminish with increasing eccentricity. To test this prediction, we investigated the effect of size for a range of eccentricities with contrast fixed to 92%. The results for foveal presentation (0 ) once again show surround suppression (Fig. 4). As eccentricity increased, the duration thresholds dropped for all sizes (Fig. 4a). More importantly, the size dependency of duration thresholds systematically changed with eccentricity, with almost no effect at the largest eccentricity tested (Fig. 4b). A closer examination of the results reveals that for each eccentricity (except 54 ) there is a size near which a small increase in size had a relatively large effect on the duration thresholds (crossed circles in Fig. 4b). This critical size increases with eccentricity, paralleling the eccentricitydependent increase in critical stimulus size required for surround suppression to occur in macaque MT 4. The present study reveals the detrimental effect of increasing stimulus size on human motion perception. This finding stands in contrast to earlier attempts to characterise the receptive field properties of human motion perception. Spatial summation has often been assumed as a basic characteristic of motion processing 11,12, but we show that spatial summation is restricted to low visibility conditions. Several previous reports are consistent with parts of the results reported here. Verghese and Stone showed that dividing a large high-contrast object into smaller parts actually improved performance in a speed discrimination task and suggested surround suppression as one possible explanation 26. Derrington and Goddard reported that increasing the contrast of a brief large drifting grating reduced performance 27. This result agrees with a portion of our findings (filled diamonds in Fig. 1b), but their explanation differs from ours. They emphasised that brief motion stimuli, by virtue of their broad temporal frequency spectrum, should stimulate motion filters tuned to opposing directions of motion. At high contrast, these paired filters could saturate, minimising the imbalance of activation, and causing the impaired motion discrimination. This account provides a possible explanation for the effects of contrast reported here, but it fails to capture the effects of size that are central to this paper, and it cannot explain the MAE results (Fig. 3b) that generalise our principal findings to prolonged motion stimuli. The series of results reported here both contradicts everyday intuition and defies accepted ideas about spatial properties of motion perception. These highly counterintuitive results, however, become much less surprising when considered in the context of the receptive field properties of neurons in cortical area MT. Indeed, our results provide converging evidence consistent with the current understanding of how MT neurons respond to moving objects differing in size, eccentricity, prolonged exposure, luminance contrast, and colour contrast. Taken together, these results suggest that impaired motion direction discrimination for large high-contrast stimuli is a perceptual correlate of centre-surround mechanisms in MT. Furthermore, our results generate a series of predictions that can be tested using neurophysiological and neuroimaging approaches. For example, the results with isoluminant stimuli (Fig. 3a) predict that surround inhibition in MT should diminish at isoluminance. The observed contrast dependence of size effects is particularly interesting. It agrees with contrast-dependent changes in receptive field properties observed in primary visual cortex 13-16, and it predicts that similar effects may also occur in MT. Our results clearly indicate that the 4

5 integration of motion signals over space is an adaptive process that enables the visual system to more efficiently process moving stimuli by employing computationally important suppressive mechanisms only when the sensory input is sufficiently strong to guarantee visibility. Methods Patterns were displayed on a linearized monitor with 800 x 600 resolution at 120 Hz (main results were confirmed at 160 and 200 Hz). Viewing was binocular at 83 cm. Ambient illumination was 4.8 cd/m2. Background luminance was 60.5 cd/m2. The size was defined as 2s width of a Gabor patch (Fig. 1d) a drifting vertical sine grating windowed by a stationary two-dimensional Gaussian envelope (s = standard deviation of the Gaussian). Duration was defined as 2s of the temporal Gaussian envelope. Threshold exposure durations were obtained for a range of Gabor patch sizes and peak contrasts. On each trial, a drifting Gabor patch was presented foveally and observers indicated perceived direction (left vs. right) by a key press. Feedback was provided. Duration thresholds (82%) were estimated by interleaved Quest staircases. For each condition, five observers ran four pairs of interleaved staircases, with the first pair discarded as practice. All experiments complied with institutionally reviewed procedures for human subjects. Added noise experiment Speed was increased to 4 /s. A Gabor patch was presented foveally and masked by of a 20 by 20 noise pattern. Noise consisted of dynamic dark and light elements (2.5 by 2.5 arcmin) that were randomly generated at 120 Hz. Other methods were the same as in the first experiment. Isoluminant motion experiment The Gabor patch consisted of spatially overlapping isoluminant red and green gratings. The red-green isoluminant point was obtained with the minimum motion technique 28. If isoluminant red and green gratings are presented spatially in anti-phase, the resulting red-green compound grating has no luminance contrast. Presenting the same two gratings in phase produces a yellow-black luminance grating. Spatial frequency was 1 cycle/. A control experiment measuring duration thresholds yielded results qualitatively similar to those in Fig. 3a. MAE experiment Observers (n = 3) adapted to a foveally presented 1 cycle/ Gabor patch moving at 4 /s. In separate conditions, MAE was measured for various sizes and contrasts. The test stimulus consisted of two overlapping Gabor patches drifting in opposite directions (stimulus parameters were the same as the adapting Gabor except for width, which was fixed to 1 ). If the contrasts of two Gabors were identical, the motion was ambiguous and the test stimulus appeared to flicker. Adapting to a motion in one direction effectively reduces motion strength of the Gabor moving in the adapted direction and the test stimulus will appear to move in the opposite direction. Then, to restore the 5

6 perception of flicker, the contrast of the Gabor moving in the adapted direction is increased and the contrast of the other Gabor lowered. The contrast ratio required to restore flicker perception is taken as a measure of MAE strength (higher contrast ratio = stronger MAE). Initial adaptation was for 30s (10s after the first trial), followed by a 0.3s blank screen and a 1s test stimulus. After viewing the test stimulus, observes indicated the perceived direction. The contrast ratio of two Gabor patches composing the test stimulus was then adjusted under control of two interleaved 1-up 1-down staircases. Each staircase converged after 6 reversals, with the average of the last 4 reversals taken as the result. Eight measurements were made for each condition. Eccentric presentation experiment To present visually larger stimuli within the limited monitor area, a Gaussian spatial envelope was replaced with the 2D raised cosine envelope. Size was defined as the distance between two diametrically opposing points on the raised cosine envelope where the contrast is 60.7% of the peak contrast (analogous to the 2s of a Gaussian distribution). Spatial frequency was lowered to 0.5 cycles/, increasing the speed to 4 /s. In this experiment only, motion directions were vertical (up vs. down). Other methods were the same as in the first experiment Allman, J., Miezin, F. & McGuinness, E. Stimulus specific responses from beyond the classical receptive field: neurophysiological mechanisms for local-global comparisons in visual neurons. Annu. Rev. Neurosci. 8, (1985). 2. Allman, J., Miezin, F. & McGuinness, E. Direction- and velocity-specific responses from beyond the classical receptive field in the middle temporal visual area (MT). Perception 14, (1985). 3. Born, R. T. & Tootell, R. B. Segregation of global and local motion processing in primate middle temporal visual area. Nature 357, (1992). 4. Raiguel, S. E., van Hulle, M. M., Xiao, D. K., Marcar, V. L. & Orban, G. A. Shape and spatial distribution of receptive fields and antagonistic motion surround in the middle temporal area (V5) of the macaque. Eur. J. Neurosci. 7, (1995). 5. Born, R. T., Groh, J. M., Zhao, R. & Lukasewycz, S. J. Segregation of object and background motion in visual area MT: effects of microstimulation on eye movements. Neuron 26, (2000). 6. Nakayama, K. & Loomis, J. M. Optical velocity patterns, velocity-sensitive neurons, and space perception: a hypothesis. Perception 3, (1974). 7. Gautama, T. & Van Hulle, M. M. Function of center-surround antagonism for motion in visual area MT/V5: a modeling study. Vision Res. 41, (2001). 8. Buracas, G. T. & Albright, T. D. Contribution of area MT to perception of three-dimensional shape: computational study. Vision Res. 361, (1996). 9. Liu, L. & Van Hulle, M. M. Modeling the surround of MT cells and their selectivity for surface orientation in depth specified by motion. Neural Comput. 10, (1998). 10. Xiao, D. K., Raiguel, S., Marcar, V., Koenderink, J. & Orban, G. A. Spatial heterogeneity of inhibitory surrounds in the middle temporal visual area. Proc. Natl. Acad. Sci. U.S.A. 92, (1995). 11. Anderson, S. J. & Burr, D. C. Spatial summation properties of directionally sensitive mechanisms in human vision. J. Opt. Soc. Am. A 8, (1991). 12. Watson, A. B. & Turano, K. The optimal motion stimulus. Vision Res. 35, (1995). 13. Kapadia, M. K., Westheimer, G. & Gilbert, C. D. Dynamics of spatial summation in primary visual cortex of alert monkeys. Proc. Natl. Acad. Sci. U.S.A. 96, (1999). 14. Levitt, J. B. & Lund, J. S. Contrast dependence of contextual effects in primate visual cortex. Nature 387, (1997).

7 7 15. Sceniak, M. P., Ringach, D. L., Hawken, M. J. & Shapley, R. Contrast's effect on spatial summation by macaque V1 neurons. Nature Neurosci. 2, (1999). 16. Somers, D. C. et al. A local circuit approach to understanding integration of long-range inputs in primary visual cortex. Cereb. Cortex 8, (1998). 17. Sclar, G., Maunsell, J. H. & Lennie, P. Coding of image contrast in central visual pathways of the macaque monkey. Vision Res. 30, 1-10 (1990). 18. Nakayama, K. & Silverman, G. H. Detection and discrimination of sinusoidal grating displacements. J. Opt. Soc. Am. A 2, (1985). 19. Dobkins, K. R. & Albright, T. D. in High-level motion processing (ed. Watanabe, T.) (MIT Press, Cambridge, MA, 1998). 20. Gegenfurtner, K. R. et al. Chromatic properties of neurons in macaque MT. Vis. Neurosci. 11, (1994). 21. Thiele, A., Dobkins, K. R. & Albright, T. D. Neural correlates of chromatic motion perception. Neuron 32, (2001). 22. Huk, A. C., Ress, D. & Heeger, D. J. Neuronal basis of the motion aftereffect reconsidered. Neuron 32, (2001). 23. Murakami, I. & Shimojo, S. Modulation of motion aftereffect by surround motion and its dependence on stimulus size and eccentricity. Vision Res. 35, (1995). 24. Sachtler, W. L. & Zaidi, Q. Effect of spatial configuration on motion aftereffects. J. Opt. Soc. Am. A 10, (1993). 25. Albright, T. D. Direction and orientation selectivity of neurons in visual area MT of the macaque. J. Neurophysiol. 52, (1984). 26. Verghese, P. & Stone, L. S. Perceived visual speed constrained by image segmentation. Nature 381, (1996). 27. Derrington, A. M. & Goddard, P. A. Failure of motion discrimination at high contrasts: evidence for saturation. Vision Res. 29, (1989). 28. Cavanagh, P., MacLeod, D. I. & Anstis, S. M. Equiluminance: spatial and temporal factors and the contribution of blue-sensitive cones. J. Opt. Soc. Am. A 4, (1987). 29. Rees, G., Friston, K. & Koch, C. A direct quantitative relationship between the functional properties of human and macaque V5. Nature Neurosci. 3, (2000). 30. Kastner, S. et al. Modulation of sensory suppression: implications for receptive field sizes in the human visual cortex. J. Neurophysiol. 86, (2001). Acknowledgements We thank Jeff Schall, Stephanie Shorter-Jacobi and Anup Panduranga for helpful comments on the manuscript. Correspondence and requests for materials should be addressed to D.T. ( duje.tadin@vanderbilt.edu).

8 Tadin et al., Figure a 92% b 5 Duration threshold (ms) Log threshold change c % 22% 11% 5.5% 2.8% 92% 46% 22% 11% 5.5% 2.8% Contrast d 2 deg Figure 1 Effects of size and contrast on motion discrimination. Individual data points are averages for five observers. Error bars are SEM. a, Duration thresholds as a function of stimulus size at different contrasts. b, Duration thresholds as a function of contrast for a range of stimulus sizes. c, Log threshold change as a function of stimulus size at different contrasts. For each observer, log threshold change was calculated relative to the duration threshold for the smallest size (0.7º) at each contrast level. d, A Gabor patch 2.7º wide shown relative to an average foveal macaque MT receptive field. The dashed dark circle illustrates the stimulus size beyond which an average foveal MT center-surround neuron exhibits surround suppression 4. The radius of the surround is usually about 3 times larger than the center radius, as indicated by the full circle. Full spatial extent of the Gabor patch (r = 3σ = 4º) is indicated by the light dashed circle. This comparison assumes that the properties of human and macaque MT are comparable 29, and that the receptive field sizes are similar for two species 30.

9 Tadin et al., Figure a 80 b Duration threshold (ms) 40 46% 30 16% 0% Threshold phase shift (deg) % % 92% 22% 11% 5.5% Figure 2 Results from added noise and phase shift experiments. a, Duration thresholds as a function of stimulus size at different noise contrasts. b, Phase shift thresholds as a function of stimulus size at different contrasts. Error bars are SEM.

10 Tadin et al., Figure 3. Threshold phase shift (deg) a Isoluminant Luminance contrast MAE strength (contrast ratio) b AP RB DT Adapting Figure 3 Results from isoluminant motion and MAE experiments. a, Phase shift thresholds as a function of stimulus size for luminance contrast and isoluminant stimuli. b, Effects of stimulus size and contrast on the MAE strength for three subjects. Filled symbols show MAE strength for low-contrast adaptation (2.8%). Empty symbols show results for high-contrast adaptation (26%). Error bars are SEM.

11 Tadin et al., Figure 4. Duration threshold (ms) a Log threshold change b Figure 4 Effects of size and eccentricity on motion discrimination. a, Duration thresholds as a function of stimulus size at different eccentricities. Error bars are SEM. b, Log threshold change as a function of stimulus size at different eccentricities. For each observer, log threshold change was calculated relative to the duration threshold for the smallest size at each eccentricity. For clarity, error bars (SEM) for each eccentricity were averaged and are shown at the end of each curve. Crossed circles mark the points where increasing the stimulus width by 1.3º had a particularly large effect on the duration thresholds.