Intrinsic timescales of sensory integration for motion perception

Woochul Choi, Se‐Bum Paik

Abstract

A subject-specific process of accumulation of information may be responsible for variations in decision time following visual perceptions in humans. A detailed profile of this perceptual decision making, however, has not yet been verified. Using a coherence-varying motion discrimination task, we precisely measured the perceptual decision kernel of subjects. We observed that the kernel size (decision time) is consistent within subjects, independent of stimulus dynamics, and the observed kernel could accurately predict each subject’s performance. Interestingly, the performance of most subjects was optimized when stimulus duration was matched to their kernel size. We also found that the observed kernel size was strongly correlated with the perceptual alternation in bistable conditions. Our result suggests that the observed decision kernel reveals a subject-specific feature of sensory integration.

Introduction

When two people observe the same moving object, they often take different amounts of time to decide on its direction. Traditionally, this inter-individual variability has been dismissed as "noise" or mere statistical fluctuation. However, computational models, such as the "diffusion-to-boundary" process, suggest that our brains accumulate evidence over time until a specific threshold is reached.

The central question addressed in this research is whether this accumulation process is a flexible state that adapts to the environment or a fixed, structural trait of an individual's neural circuitry. By identifying a stable "perceptual decision kernel" for each person, this study seeks to explain why some individuals are naturally more adept at certain visual tasks than others.

Methods

The researchers employed a coherence-varying motion discrimination task involving 45 human subjects (23 females, 22 males, aged 20–29). Subjects viewed a circular annulus of dots where the "net motion" (coherence) fluctuated randomly over time, similar to white noise.

To identify the underlying mechanisms, the authors used several key technical approaches:

  • Reverse Correlation (Response-Triggered Average, RTA): By analyzing the stimulus patterns immediately preceding a subject's response, the researchers extracted the specific temporal window—the kernel—the brain used to make a decision.
  • System Identification: The RTA was modeled using a superposition of two alpha functions, a method typically used to describe the temporal receptive fields of retinal neurons.
  • Linear Convolution: The observed kernels were convolved with new stimulus patterns to mathematically predict when a subject would "switch" their perceived direction of motion.
  • Psychophysical Validation: A secondary experiment varied the duration of a constant motion stimulus to identify each subject's "optimal" performance point.

Key Insights and Results

The Kernel as a Biological Trait

The study found that the size of a person's decision kernel ($T_0$) remained remarkably consistent even when the stimulus dynamics changed eight-fold. This suggests that the timescale of sensory integration is "hard-wired" into the individual's brain rather than being a temporary state. The intrinsic decision window ($T_0$) varied significantly between individuals, ranging from 1 to 4 seconds.

The "Goldilocks" Effect in Perception

Counter-intuitively, more information does not always lead to better performance. Because the decision kernel has a non-linear shape—characterized by a positive peak followed by a negative "overshoot"—integrating information for too long can actually degrade performance. The researchers discovered that accuracy peaks when the stimulus duration matches the individual's kernel size (specifically, approximately 1.2 times the kernel size).

Predictive Power and Bistability

By knowing a subject's kernel size, the researchers could accurately predict their accuracy and reaction time. There was a strong negative correlation ($r = -0.71$) between kernel size and accuracy; subjects with shorter kernels were generally better at detecting fast changes in motion.

Furthermore, the kernel size was strongly correlated ($r = 0.71$) with the phase duration of illusory motion in bistable conditions (such as the spinning dancer illusion). This suggests that the same neural mechanism governing real motion perception also dictates how we experience illusory "flips" in perception.

Discussion and Future Work

These findings suggest that the observed decision kernel is a "perceptual fingerprint" unique to each individual. While the study identifies the existence of these kernels, their exact biological origin remains a hypothesis. The authors suggest that structural variations in the superior parietal lobes (SPL) or differences in neural inhibition levels might be responsible, though further neuroimaging (fMRI/EEG) is required for confirmation.

Additionally, while this study focused on motion, it raises the question of whether these intrinsic timescales are "modality-general." Future research may determine if a person with a slow motion-integration kernel also processes auditory or tactile stimuli at a similarly slow pace, potentially revealing a unified temporal framework for human perception.

Summary updated

Abstract

A subject-specific process of accumulation of information may be responsible for variations in decision time following visual perceptions in humans. A detailed profile of this perceptual decision making, however, has not yet been verified. Using a coherence-varying motion discrimination task, we precisely measured the perceptual decision kernel of subjects. We observed that the kernel size (decision time) is consistent within subjects, independent of stimulus dynamics, and the observed kernel could accurately predict each subject’s performance. Interestingly, the performance of most subjects was optimized when stimulus duration was matched to their kernel size. We also found that the observed kernel size was strongly correlated with the perceptual alternation in bistable conditions. Our result suggests that the observed decision kernel reveals a subject-specific feature of sensory integration.
BibTeX
@inproceedings{choi2019intrinsic,
  author = {Woochul Choi and Se‐Bum Paik},
  title = {{Intrinsic timescales of sensory integration for motion perception}},
  booktitle = {Scientific},
  year = {2019},
  doi = {10.1038/s41598-019-40649-9},
}
Figure