Physiological Mechanisms Underlying Motion-Induced Blindness
Camilo Libedinsky
1
, Tristram Savage
1
, Margaret Livingstone
1*
1. Dept. Neurobiology, Harvard Medical School, 220 Longwood Avenue, Boston, MA 02115
Visual disappearance illusions such as motion-induced blindness (MIB) - are
commonly used to study the neural underpinnings of visual perception. In such
illusions a salient visual target becomes perceptually invisible. Previous studies are
inconsistent regarding the role of primary visual cortex (V1) in these illusions. Here
we provide physiological and psychophysical evidence supporting a role for V1 in
generating MIB.
Some of the most striking visual illusions fall into the category of multistable
phenomena. These are situations in which an unchanging stimulus generates alternating
perceptual states. Some examples are Necker cube reversals, binocular rivalry,
ambiguous structure from motion and motion-induced blindness
1
.
Motion-induced blindness (MIB) is a phenomenon of visual disappearance in
which a salient target becomes intermittently invisible when surrounded by a field of
rotating distractors
1
. Several explanations have been proposed to explain this illusion:
slowdown of the attentional switch
1
, interhemispheric competition
2
, depth ordering,
surface completion
3
and perceptual filling-in
4
among others. The aim of our first,
perceptual, experiment was to test whether having a large surface-inducing mask is
necessary for MIB to occur. To do this we compared the effects on the rate of target
disappearance of a full mask, a mask that only just surrounded the target and flashing
bars around the target
5
(Figure 1A, 1B and 1C). There was no difference between the
rates of disappearance or the time the target remained invisible under these conditions
Nature Precedings : hdl:10101/npre.2008.1506.1 : Posted 9 Jan 2008
(Figure 1E). To further explore how early in the visual system MIB suppression
originates, we asked how MIB is affected by segregating the mask and target across the
vertical midline. We arranged target and mask as shown in Figure 1D with the mask only
visible at a distance of 1 degree to the left of the target. Then we varied the fixation
location so that the mask and target were both on the same side of the midline, or on
opposite sides of the midline, and we found that the target disappeared significantly less
often when it was on the opposite side of the vertical midline from the mask, compared to
the same-side condition (Figure 1F). This result further supports the idea that MIB is
generated by suppressive interactions occurring at early visual areas, because only in
early visual areas are receptive fields (and their inhibitory surrounds) restricted to one or
the other hemisphere
6,7
.
Since the perceptual studies summarized in Figure 1 indicate a role for early
visual areas in generating MIB, we looked at the firing patterns of individual V1 neurons
in two alert fixating macaques while they viewed the MIB stimulus. One of these
monkeys was trained to report the visibility of a peripheral yellow target in the presence
of an MIB mask while maintaining fixation on a small spot. Each trial started with the
target present in the cell's receptive field, and the monkey was trained to move a lever
rightward when he saw the target disappear, and to move it leftward when the target re-
appeared. In some trials the target actually disappeared and reappeared, and in some trials
it remained present throughout the trial. The monkey was rewarded at the end of the trial.
Great care was taken to ensure that the lever pulls reflected perceptual reports (see
supplementary methods). The pattern of the monkey's reports indicates that macaques,
Nature Precedings : hdl:10101/npre.2008.1506.1 : Posted 9 Jan 2008
like humans, perceive disappearances of the salient target in the presence of a moving
field of dark blue crosses.
We recorded from single units in V1 while the monkey viewed the MIB stimulus
shown in Figure 1A, with the target centered on the receptive field of each cell recorded.
A protection zone surrounding the target prevented the mask from entering the activating
zone of the V1 cells. We compared V1 neural activity preceding lever presses in trials
when the target actually disappeared and re-appeared to the activity in trials when the
monkey moved the lever even though the target was continuously present throughout the
trial (and we will refer to the lever presses in these latter trials as indicating illusory
disappearances). We observed, as expected, an increase in neural activity around 500 ms
before lever presses in response to actual target appearances and disappearances, but we
also observed a similar, but smaller, average increase in activity before lever presses
indicating illusory transitions. However the increases in neural activity preceding illusory
transitions were much smaller than the peaks of activity preceding real target transitions
and did not reach statistical significance (Fig 2A). We therefore cannot explain the
illusory disappearances simply by parallel changes in the activity of V1 cells.
However, we noticed that in the presence of the MIB mask, the responses of the
V1 cells to actual appearances and disappearances of the target were often attenuated
compared to the mask-absent condition, so we asked whether the mask might weaken or
interfere with neural responses to the target in V1. We measured the responses of 25
single cells in V1 to the presentation of the same target with and without the MIB mask
in two monkeys during passive fixation. On average, the neurons gave smaller responses
to both appearances and disappearances of the target in the presence of a surrounding
Nature Precedings : hdl:10101/npre.2008.1506.1 : Posted 9 Jan 2008
mask compared to the no-mask control condition (Fig 2B). On average, there was a
significant decrease in the initial peak response to both target ON and target OFF in the
presence of the MIB mask (paired t-test, p<0.05), and no significant difference in the
sustained responses (300-500ms, paired t-test, p>0.05).
Our physiological recordings from macaque V1 thus showed that although V1
target responses did not parallel target visibility, early signals from V1 in response to
target transitions were significantly reduced in the presence of the MIB mask, but the
sustained phases of the responses were unaffected. Our failure to observe a reduction in
the sustained responses to the target indicates that the perceptual disappearances might
not be attributable to the reduction of signals from early visual areas reflecting merely the
presence of the target, but rather to changes in signals indicating target transitions. That
is, this result raises the question of whether the mask actually does render the target "less
visible" or whether it makes the target "more likely to disappear". Therefore we explored
this issue perceptually by sinusoidally modulating the luminance of the target in the
presence and absence of the MIB mask around two values (a high and a low luminance
value). If the MIB mask simply renders the target "less visible" then we expect the target
to disappear more frequently during the dimmest periods of the luminance cycle; if the
MIB disappearances are caused by changes in the likelihood of "disappearances" then we
would expect the target to disappear more frequently during the decreasing brightness
phases of the brightness cycle. We found that subjects reported target disappearances
much more often right after the target started dimming in the presence of the MIB
mask, for both high and low luminance levels (Figure 1G), even though in the absence
of the mask the target simply appeared to dim, not disappear. This suggests that
Nature Precedings : hdl:10101/npre.2008.1506.1 : Posted 9 Jan 2008
regardless of the absolute value of luminance of the target, under MIB conditions, small
transients induce the disappearance of the target, whether this transient originates in the
target or within the brain itself.
In summary, we found both perceptual and physiological evidence that MIB can
originate in early visual areas. We established that macaque monkeys, like humans,
perceive the MIB illusion, and, even though the activity of V1 cells did not correlate
directly with the illusory disappearances of the target, the responses in V1 to the target
were diminished by the MIB mask. Furthermore, decreases in target luminosity,
regardless of absolute luminosity level, induced perceptual disappearances of the target.
Such decreases in target luminosity should cause transient OFF responses in a
subpopulation of V1 cells. Since perceptual disappearances tended to occur right after the
target decreased in luminance, we deduce that these disappearances were caused by OFF
responses. Because we also found that under MIB conditions the initial transient
responses of V1 cells were reduced (thus bringing the peak response closer to the noise
level), we suggest that spontaneous perceptual transitions during MIB are caused by the
`chance' event that a sufficiently large population of OFF cells in visual cortex happened
to fire enough to fool the system into believing that a real transition occurred. This would
also explain why we found a weak (but not significant) correlation between V1 activity
and perceptual state during MIB.
We found that the responses of V1 cells to target onset or offset were reduced in
the presence of an MIB mask. So, even though the mask fell well beyond V1 classical
receptive fields due to the protection-zone, it still produced a modulatory influence on V1
responses. Primary visual cortex is likely not the only factor influencing the
Nature Precedings : hdl:10101/npre.2008.1506.1 : Posted 9 Jan 2008
disappearances, as contextual surround suppression could arise at any cortical level. But
the perceptual and physiological results presented here show that effects of the mask in
V1 likely contribute to the phenomenon.
The involvement of early visual areas in MIB has been overlooked because
several lines of evidence point away from early topographic visual areas as an important
locus for MIB. Aftereffects and adaptations, which are assumed to arise in V1, are not
affected by MIB disappearances
8,9,10,11
. Furthermore, factors assumed to be important for
MIB, such as attention, object selectivity
1
, surface completion, depth ordering
3
and
interhemispheric switch
2
are thought to arise at levels higher than V1. It has also been
shown that V1 activity does not correlate with perceptual state for other visual
disappearance illusions
12
. On the other hand, Kawabe et al. (2007)
5
and Wilke et al.
(2003)
13
provided evidence that low-level signals are involved in visual disappearance
phenomena. However, because our results implicate the transient phase of visual
responses, we can now argue that adaptations and aftereffects not being affected by MIB
is not inconsistent with an effect in early visual areas, since adaptations and aftereffects
result from prolonged sensory stimulation and are not dependent on the initial burst
response
14
.
Even if high-level effects such as object competition or attentional modulation are
the final stages responsible for target visibility, our results suggest that the mask-induced
reduction in target responses as early as V1 also play an important role. That is, when the
signal from lower levels is noisier, the detection processes in higher-level cells will also
be more error prone. In this view, we would expect activity in the whole population of V1
cells that respond to the target to correlate to some degree with the perceptual report,
Nature Precedings : hdl:10101/npre.2008.1506.1 : Posted 9 Jan 2008
although this correlation need not be as strong as during real transitions, since errors
could be initiated anywhere along the pathway, not just at the first stage. This
interpretation fits well with other single-unit studies and studies correlating fMRI signals
and local-field potentials in early visual areas to perceptual state
15
.
References
1.
Bonneh Y.S., Cooperman A. & Sagi D. Nature. 411:798-801 (2001).
2.
3.
4.
Hsu L.C., Yeh S.L. & Kramer P. Vision Res. 46:1973-81 (2006).
5.
6.
Hubel D.H. & Wiesel T.N. J. Neurophysiol. 30:1561-73 (1967).
7.
Van Essen D.C., Newsome W.T. & Bixby J.L. J. Neurosci. 2:265-83 (1982).
8.
9.
10.
Rajimehr R. Neuron. 41:663-73 (2004).
11.
Montaser-Kouhsari L., Moradi F., Zandvakili A. & Esteky H. Perception. 32:295-
305 (2003).
12.
Leopold D.A. & Logothetis N.K. Nature. 380:549-553 (1996).
13.
Wilke M., Logothetis N.K. & Leopold D.A. Neuron. 39:1043-52 (2003).
14.
15.
Wilke M., Logothetis N.K. & Leopold D.A. Proc. Natl. Acad. Sci. U.S.A.
103:17507-12 (2006)
Nature Precedings : hdl:10101/npre.2008.1506.1 : Posted 9 Jan 2008
Figure 1. (A) Full mask, a 9 x 9 field composed of 81 equally spaced blue crosses rotated
about its center-point (fixation spot) at 45°/s. A yellow target was located 2° from the
fixation spot. (B) Local mask. This was the same stimulus as the full mask (A) except
that the only part of the mask still present was a 0.5° annulus around the yellow target.
(C) Flashing bars. This was the same stimulus as the full mask (A) except that the mask
was replaced by two sets of sequentially flashing bars. The frequency of the flashes was 5
10 Hz. (D) Stimulus used in experiment 2. Here the mask was only present beyond an
imaginary line 1 degree from the target. There are 6 conditions, in all of which the target
is 2 degrees from the fixation spot. In Midline, the target is located ½ a degree to the right
of the fixation spot. In the Left 1 and Left 2 conditions the target is moved to the left by
½ and 1 degree respectively. And the opposite is true for the Right 1 and Right 2. In the
No Mask condition, no mask is present. (E) Average of normalized (to full-mask value)
number of disappearances (left) and time of invisibility (right) under Full Mask (green),
Local Mask (yellow) and Flashing Bars (grey). Error bars represent 1 standard error. (F)
Effect of having mask and target in different hemifields. Stimuli as in Figure 1D. (G)
Disappearance rate (percent of times target disappeared each cycle) of the target
(continuous lines) and luminosity of target (dotted lines) over time. Red and blue lines
represent low and high luminance respectively.
Nature Precedings : hdl:10101/npre.2008.1506.1 : Posted 9 Jan 2008
Figure 2. (A) Average response of V1 cells when the monkey report target ON (left) and
target OFF (right) under illusory (red) or real (blue) transitions. Dotted lines indicate time
of lever press. Shaded area denotes the standard error. (B) Population average of V1 cells
during passive fixation (mean + standard error) to target ON (left) and target OFF (right)
of cells with larger response to no-mask condition. Blue line represents the average firing
rate when no mask was present and red the average firing rate when the MIB mask was
present. Responses were aligned by time to peak and normalized by the maximum firing
rate for each cell.
Nature Precedings : hdl:10101/npre.2008.1506.1 : Posted 9 Jan 2008
Nature Precedings : hdl:10101/npre.2008.1506.1 : Posted 9 Jan 2008
Nature Precedings : hdl:10101/npre.2008.1506.1 : Posted 9 Jan 2008