Abstract
Neurobiological investigations of perceptual decision-making have furnished the first glimpse of a flexible cognitive process at the level of single neurons. Neurons in the parietal and prefrontal cortex are thought to represent the accumulation of noisy evidence, acquired over time, leading to a decision. Neural recordings averaged over many decisions have provided support for the deterministic rise in activity to a termination bound. Critically, it is the unobserved stochastic component that is thought to confer variability in both choice and decision time. Here, we elucidate this drift-diffusion signal on individual decisions. We recorded simultaneously from hundreds of neurons in the lateral intraparietal cortex (LIP) of monkeys while they made decisions about the direction of random dot motion. We show that a single scalar quantity, derived from the weighted sum of the population activity, represents a combination of deterministic drift and stochastic diffusion. Moreover, we provide direct support for the hypothesis that this drift-diffusion signal approximates the quantity responsible for the variability in choice and reaction times. The population-derived signals rely on a small subset of neurons with response fields that overlap the choice targets. These neurons represent the integral of noisy evidence. Another subset of direction-selective neurons with response fields that overlap the motion stimulus appear to represent the integrand. This parsimonious architecture would escape detection by state-space analyses, absent a clear hypothesis.
Introduction: The Challenge of Single-Trial Decision Dynamics
For decades, the Drift-Diffusion Model (DDM) has served as the foundational framework for understanding perceptual decision-making. The model posits that the brain integrates noisy sensory evidence (drift) over time until a decision threshold (bound) is reached. While the DDM successfully accounts for behavioral metrics such as accuracy and reaction time (RT), direct neural validation has been constrained by the inherent noise of individual neurons. Traditionally, researchers have relied on trial-averaging to reveal the "ramp-to-threshold" activity, a process that inadvertently masks the stochastic "diffusion" component—the unique random walk that characterizes a single decision. This study bridges the gap between theory and observation by identifying the drift-diffusion signal within the simultaneous activity of a neural population on a single trial.
Experimental Methods
The researchers employed Neuropixels probes to record from populations of 54 to 203 neurons in the lateral intraparietal area (LIP) of two rhesus monkeys (Macaca mulatta). The subjects performed a random dot motion task, requiring them to discern the net direction of motion and indicate their choice via a saccade to one of two targets.
The analysis focused on identifying a decision variable (DV) from the population activity. Key innovations included:
- Targeted Subpopulation Analysis: Focusing on $T_{in}$ neurons, whose response fields overlap with choice targets.
- Dimensionality Reduction: Utilizing PCA and a custom "diffusion vector" ($d$) to project high-dimensional activity into a one-dimensional DV.
- Mediation Analysis: A statistical approach used to demonstrate that the neural signal at a given time point accounts for the relationship between earlier signals and the final behavioral outcome (choice and RT).
Results: Direct Observation of Drift and Diffusion
By pooling activity from a small subset of neurons (approximately 17 $T_{in}$ cells), the researchers were able to suppress individual neuronal noise and reveal clear drift-diffusion trajectories on a trial-by-trial basis.
The Decision Variable and Behavioral Prediction
The population-derived DV proved highly predictive of both choice and reaction time. A logistic decoder trained on the full population reached a peak accuracy of 72%, matching the theoretical maximum predicted by behavioral models. Crucially, the $T_{in}$ subpopulation was found to be the primary driver of this signal. Removing these 17 neurons reduced decoding accuracy by 8%, whereas removing a random set of 17 neurons resulted in only a 1% decrease.
Mediation and the "Smoking Gun"
The mediation analysis provided direct evidence that the stochastic fluctuations (diffusion) in the neural signal are responsible for behavioral variability. As shown in the mediation plots (Figure 4g-i), the neural signal at later time points "mediated" the predictive power of earlier signals. This confirms that the LIP population represents a continuous accumulation process where the current state of the DV determines the future path toward a decision bound.
Parsimonious Neural Architecture
State-space analysis revealed that the direction in neural space representing the decision ($d$) aligns closely with the spatial locations of the choice targets (Figure 4f). This suggests a biologically simple organization: the neurons responsible for planning the motor response (saccade) are the same neurons performing the evidence integration.
Discussion
This work provides a direct look at the "random walk" of a decision, confirming the long-standing DDM hypothesis at the single-trial level. The findings suggest that the LIP's role in decision-making is organized around a parsimonious architecture where integration is tied to motor preparation.
While this study focused on the first 500ms of decision formation to avoid boundary effects, future research will investigate how this signal interacts with downstream structures like the superior colliculus to terminate the decision. Furthermore, the extent to which this motor-centric integration applies to more abstract or non-sensory decisions remains a key area for exploration.
Summary updated