Perceptual “Read-Out” of Conjoined Direction and Disparity Maps in Extrastriate Area MT

Gregory C. DeAngelis, William T. NewsomePLoS 2004 · DOI

Abstract

Cortical neurons are frequently tuned to several stimulus dimensions, and many cortical areas contain intercalated maps of multiple variables. Relatively little is known about how information is "read out" of these multidimensional maps. For example, how does an organism extract information relevant to the task at hand from neurons that are also tuned to other, irrelevant stimulus dimensions? We addressed this question by employing microstimulation techniques to examine the contribution of disparity-tuned neurons in the middle temporal (MT) visual area to performance on a direction discrimination task. Most MT neurons are tuned to both binocular disparity and the direction of stimulus motion, and MT contains topographic maps of both parameters. We assessed the effect of microstimulation on direction judgments after first characterizing the disparity tuning of each stimulation site. Although the disparity of the stimulus was irrelevant to the required task, we found that microstimulation effects were strongly modulated by the disparity tuning of the stimulated neurons. For two of three monkeys, microstimulation of nondisparity-selective sites produced large biases in direction judgments, whereas stimulation of disparity-selective sites had little or no effect. The binocular disparity was optimized for each stimulation site, and our result could not be explained by variations in direction tuning, response strength, or any other tuning property that we examined. When microstimulation of a disparity-tuned site did affect direction judgments, the effects tended to be stronger at the preferred disparity of a stimulation site than at the nonpreferred disparity, indicating that monkeys can selectively monitor direction columns that are best tuned to an appropriate conjunction of parameters. We conclude that the contribution of neurons to behavior can depend strongly upon tuning to stimulus dimensions that appear to be irrelevant to the current task, and we suggest that these findings are best explained in terms of the strategy used by animals to perform the task.

Introduction

Cortical neurons are rarely "experts" in just one thing; they are typically tuned to multiple stimulus dimensions simultaneously. In the visual area MT, for instance, almost all neurons are selective for both the direction of motion (task-relevant in this study) and binocular disparity (depth, which was task-irrelevant). This study reveals that the brain’s "read-out" of sensory information is not a simple hard-wired process, but a flexible strategy. By microstimulating neurons in the MT area of macaques, researchers found that the brain often ignores direction-sensitive neurons if they are also tuned to "irrelevant" dimensions like depth, suggesting that perceptual decisions are based on a selective filtering of multidimensional neural maps.

Experimental Methodology

The authors trained three rhesus monkeys to perform a classic direction discrimination task using random-dot patterns. While the monkeys judged motion, the researchers used electrical microstimulation (20 μA) to artificially bias the activity of specific "columns" of neurons in area MT.

Experimental setup showing the fixation point, receptive field (RF), and targets.

The key experimental twist was characterizing the disparity (depth) tuning of each site before testing its influence on direction judgments. Even though the monkeys only had to report direction, the researchers varied the depth of the dots to see if the microstimulation's effect depended on whether the dots' depth matched the neurons' preferred depth. They used a Disparity Tuning Index (DTI) to quantify how "picky" a site was about depth and a Microstimulation Selectivity Ratio (MSR) to measure how much the stimulation effect depended on the stimulus matching that depth preference. The study is based on data from 102 microstimulation sites across three monkeys.

Results and Insights

The study found that the brain's read-out strategy varies significantly between individuals. In two of the three monkeys, there was a strong negative correlation ($r \approx -0.6$) between a site's depth selectivity (DTI) and the magnitude of the microstimulation effect on direction.

Psychometric curves showing the proportion of preferred direction decisions as a function of motion coherence, with and without microstimulation.

As shown in the behavioral data, microstimulation (Stim) shifts the psychometric curve relative to the control (No Stim), indicating a bias toward the preferred direction of the stimulated neurons. Key findings include:

  • Selective Read-out: The brain may prefer to "listen" to neurons that respond well across all depths (nondisparity-selective) rather than switching between specialized depth-tuned neurons as the stimulus moves. This "generalist" strategy simplifies the read-out mechanism.
  • Conjoined Filtering: When depth-tuned sites did influence behavior, the effect was "tuned." The bias was significantly stronger when the visual stimulus matched the site's preferred depth, indicating the read-out mechanism can selectively monitor specific "conjunctions" of features.
  • Learned Strategies: The third monkey (Monkey R) utilized signals from both selective and non-selective sites and achieved a lower psychophysical threshold, suggesting that incorporating specialized, depth-tuned neurons into the decision process is a more difficult but more accurate strategy.

Discussion

The primary limitation is the individual variability between monkeys, which suggests that "read-out" is a learned strategic choice rather than a fixed anatomical circuit. The authors suggest that these findings are best explained in terms of the strategy used by animals to perform the task.

Future work could investigate whether monkeys can be "forced" to change their read-out strategy by making the task impossible to solve using only nondisparity-selective neurons. Additionally, exploring how these strategies adapt in real-time when task demands shift would provide deeper insight into the plasticity of the sensory-to-motor transformation.

Summary updated

Abstract

Cortical neurons are frequently tuned to several stimulus dimensions, and many cortical areas contain intercalated maps of multiple variables. Relatively little is known about how information is "read out" of these multidimensional maps. For example, how does an organism extract information relevant to the task at hand from neurons that are also tuned to other, irrelevant stimulus dimensions? We addressed this question by employing microstimulation techniques to examine the contribution of disparity-tuned neurons in the middle temporal (MT) visual area to performance on a direction discrimination task. Most MT neurons are tuned to both binocular disparity and the direction of stimulus motion, and MT contains topographic maps of both parameters. We assessed the effect of microstimulation on direction judgments after first characterizing the disparity tuning of each stimulation site. Although the disparity of the stimulus was irrelevant to the required task, we found that microstimulation effects were strongly modulated by the disparity tuning of the stimulated neurons. For two of three monkeys, microstimulation of nondisparity-selective sites produced large biases in direction judgments, whereas stimulation of disparity-selective sites had little or no effect. The binocular disparity was optimized for each stimulation site, and our result could not be explained by variations in direction tuning, response strength, or any other tuning property that we examined. When microstimulation of a disparity-tuned site did affect direction judgments, the effects tended to be stronger at the preferred disparity of a stimulation site than at the nonpreferred disparity, indicating that monkeys can selectively monitor direction columns that are best tuned to an appropriate conjunction of parameters. We conclude that the contribution of neurons to behavior can depend strongly upon tuning to stimulus dimensions that appear to be irrelevant to the current task, and we suggest that these findings are best explained in terms of the strategy used by animals to perform the task.
BibTeX
@inproceedings{deangelis2004perceptual,
  author = {Gregory C. DeAngelis and William T. Newsome},
  title = {{Perceptual “Read-Out” of Conjoined Direction and Disparity Maps in Extrastriate Area MT}},
  booktitle = {PLoS},
  year = {2004},
  doi = {10.1371/journal.pbio.0020077},
}
Figure