 |
Contextual Influences on Visual Processing
Summary: Thomas Albright is interested in understanding the neuronal bases of visual perception, visually guided behavior, and visual memory in primates.
Light reflected from objects in the environment projects on to the retinal surface, resulting in intricate and dynamic patterns of brightness and color. At the earliest stages of visual processing, simple features of these retinal images, such as spots and edges, are detected and represented by the activity of individual neurons. At intermediate processing stages, image feature information is integrated to yield representations of the visual environment, with emphasis on visual surfaces and their spatial layout. At still higher processing stages, environmental attributes are evaluated in relation to the observer's memories and prior emotional experiences, which enables objects to be recognized and decisions to be made on the basis of visual input.
The operations of these intermediate- and high-level processing stages are necessarily context dependent. Just as the meaning of a single word depends on the phrase it helps compose, the percept elicited by an image feature hinges critically on the sensory and behavioral context in which the feature is embedded. Similarly, object recognition is, by definition, a visual process rooted in the context of experience.
Until recently, little was known about how sensory neurons use context to recover the environmental causes of even the simplest retinal image features. We have begun to address this problem by including a variety of contextual manipulations in quantitative studies of visual processing in the primate cerebral cortex. The neural substrates for visual motion perception have served as a convenient model system for much of this analysis of contextual effects, but our findings address general principles of sensory processing.
Our experimental methods include behavioral assessment of perceptual state (psychophysics) in combination with techniques for recording visually evoked activity from single neurons in alert monkeys. With these methods we can identify neuronal structures and events that contribute to specific perceptual phenomena. Comparison of neuronal responses seen at sequential stages of visual processing, in conjunction with knowledge of anatomical connectivity, enables us to deduce functional circuitry. We then incorporate these data into computational models.
Motion Cues for Surface Depth Ordering
Psychophysical and neurophysiological studies have demonstrated that visual motion is extracted by a "first-order" mechanism based on luminance, as well as by a "second-order" mechanism based on nonluminance cues, such as contrast and flicker. We have recently discovered that second-order stimuli convey information about the relative depth of overlapping surfaces (i.e., depth order). Beginning with discoveries from our lab in the early 1990s, several visual cortical areas have been found to respond to second-order motion. Our new findings cast that work in an exciting new light, suggesting that these areas may be involved in extracting depth order in addition to motion information. We are currently investigating this hypothesis in cortical area MT, an area that responds to second-order stimuli.
The Influence of Adaptation on Speed Perception
In a laboratory setting, one can show that extended exposure to a high speed of motion reduces perceived speed. This is a form of sensory adaptation: the visual system adapts to its (high-speed) environment. Upon first consideration, there appears to be no benefit to this phenomenon. Upon closer inspection, however, one can observe that, even though an observer underestimates the true speed after adaptation, the observer is better able to discriminate between different speeds. The latter phenomenon may be the function of adaptation-induced changes in perceived speed.
We investigated the neural basis of this phenomenon with single-cell recordings from neurons in the middle temporal area (MT) of the macaque monkey. First, we found that the cells consistently reduced their overall firing rate following adaptation. To interpret these rate changes, we used a model of the representation of speed in MT that assumes that cells vote for their preferred speed with a weight proportional to their firing rate. In other words, a cell that prefers high speeds, votes for high speeds. But if such a cell fires less after adaptation, it will have a smaller vote. As a consequence, perceived speed will be reduced. Of course, cells that prefer low speeds predict just the opposite effect, but we found that there are fewer of those cells and they show a smaller reduction in firing. Hence, overall, one would expect a reduction in perceived speed, just as we found in the human and monkey percept.
The neurons we studied did not just reduce firing rate after adaptation; we also found that the difference in firing elicited by two different speeds was larger following adaptation. When interpreted using our model of speed perception, the observed enhancement of rate differences was shown to lead to an enhancement of discrimination performance. We conclude that the changes that take place in speed perception with long exposure to a moving stimulus can be explained by specific neural changes in area MT.
The Color Tuning of Motion Detectors in Primary Visual Cortex
The motion of a colored object against an isoluminant background is difficult to see, but the addition of even a small amount of luminance contrast can greatly enhance the perceived motion. These observations suggest that the neurons that mediate motion perception are particularly sensitive to luminance contrast.
One class of neurons known to play an important role in motion perception is the class in primary visual cortex (area V1) with space/time-oriented receptive fields (RFs). To determine whether such neurons are preferentially tuned for luminance, we stimulated V1 neurons in awake, fixating monkeys with a randomly modulated, colorful stimulus. Analysis of the stimulus patterns that preceded spikes allowed us to partition cells into luminance-tuned and color-tuned categories. It also allowed us to estimate the orientation of the space/time RF of each neuron. A large proportion of the luminance-tuned cells had space/time-oriented RFs, but only one of the color-tuned cells did. The result suggests that the privileged role that luminance information has in motion perception derives in part from the establishment of space/time-oriented RFs among luminance-tuned, but not color-tuned, V1 neurons.
Associative Learning in Cortical Visual Area MT of Macaque
When two stimuli are frequently seen together, the sight of one often elicits the thought or image of the other. This associative learning is frequently thought of as a high-level process, and evidence suggests that it may be mediated by plastic changes in the stimulus selectivities of neurons in the inferior temporal cortex. We sought to determine whether associative learning-induced neuronal plasticity is limited to this high-level visual processing stage, or whether it is a more general property that is also present at earlier processing stages. To achieve this goal, we trained monkeys to behaviorally associate naturally effective stimuli for neurons in cortical visual area MT (i.e., translatory motions) with arbitrary noneffective stimuli (static two-dimensional patterns). For example, we trained animals to associate leftward and rightward moving patches of dots with stationary leftward and rightward pointing arrows. We hypothesized that learning of the behavioral association would be paralleled by the emergence of selective neuronal responses to the static stimuli. We recorded from MT neurons before training began and after training was complete to a criterion of 80 percent correct performance.
Before training, only 4 percent of the neurons (i.e., less than chance) showed selective responsiveness to the stationary stimuli. After training, this percentage increased to 17 percent. For 80 percent of these neurons, selectivity for the stationary stimulus was detectable in the neuronal response 70 ms following stimulus onset. Remarkably, the preferred static stimulus for a given neuron tended to be the one that the monkey had learned to associate with the preferred direction of motion for that neuron. This result was highly significant over the population of neurons recorded (p < 0.001). Thus, the stimulus selectivities of many neurons in area MT, a relatively early visual processing stage, are plastic and selectively modifiable by associative learning. The early onset of these “trained responses suggests that they may result from local hebbian plasticity, rather than feedback from higher visual areas.
Neural Correlates of Knowledge
Behavioral responses to a sensory stimulus are often guided by associative memories. These associations remain intact even when other factors determine behavior. The substrates of associative memory should therefore be identifiable by neuronal responses that are independent of behavioral choices. We tested this hypothesis using a paired-associates task in which monkeys learned arbitrary associations between pairs of visual stimuli. We examined the activity of neurons in inferior temporal cortex as the animals prepared to choose a remembered stimulus from a visual display. The activity of some neurons (22 percent) depended on the monkey's behavioral choice. But for a novel class of neurons (54 percent), activity reflected the stimulus the monkey was instructed to choose, regardless of the behavioral response. These neurons appear to represent memorized stimulus associations that are stable across variations in behavioral performance.
Grants from the National Institutes of Health provided support for some of these studies.
Last updated July 14, 2005
|
|
|
|
