Overt visual attention modulates decision-related signals in the frontal cortex

Eyes Don't Just Watch—They Weight the Math of Choice

When making decisions, our eyes are closely linked to how our brain weights value. This study shows that looking at an option is associated with amplified value signals in the brain's decision-making circuits. These results suggest that visual attention may serve as a mathematical weight in the evidence-accumulation process.

The Confound of the Accumulator

Researchers often rely on Sequential Sampling Models (SSMs) to study human choice. These models suggest the brain gathers evidence for each possibility over time. This process uses two distinct components: inputs and integrators. Inputs represent the immediate value of the information being sampled (the "drift rate"). Integrators represent the running total of all evidence gathered (the "decision variable").

Identifying the neural substrates of these components is notoriously difficult. Most decision-making tasks are extremely rapid. They often finish faster than the temporal resolution of functional magnetic resonance imaging (fMRI). fMRI measures blood-oxygen-level-dependent (BOLD) signals, which are indirect markers of neural activity.

Because these tasks are so fast, researchers struggle to distinguish between overlapping signals. Is a brain region active because it is accumulating evidence? Or is it reacting to the "conflict" of two similar options? Is the activity a reflection of the actual decision variable? Or is it just a byproduct of "time-on-task"—the fact that more time has passed? Without decoupling these variables, the true architecture of the decision engine remains obscured.

Dissecting the Decision Timeline

To resolve these ambiguities, the authors used a dual-modality approach. They combined simultaneous eye-tracking and fMRI. The core innovation was an "expanded-judgment task." Instead of a split-second choice, subjects learned the value of two food lotteries. They sampled random draws every 4 to 8 seconds .

This slowed the decision process to approximately one minute. This length provided a window to observe how neural signals evolve.

The methodology relies on a mathematical dissociation between two signals: 1. Sampled Value (SV): The immediate difference in value between the two items currently being viewed. This serves as the "input" signal. 2. Accumulated Value (AV): The running sum of all previous sampled differences throughout the trial. This serves as the "integrator" signal.

Separating these is a methodological breakthrough. It allows researchers to tell if a region is merely sensing new information (input) or performing the math to combine it (integration). By tracking BOLD activity, the researchers could pinpoint which regions act as "input" sensors and which act as "integrators." They also measured gaze. This allowed them to calculate "gaze-weighted" values. They tested whether visual attention correlates with the weighting of incoming evidence.

Gaze as a Mathematical Multiplier

Behavioral results confirmed that subjects followed the rules of an SSM. They made choices based on both the immediate sample and the total accumulated evidence [Figure 3A]. The authors also found a significant "recency bias." This means information sampled toward the end of a trial influenced the final choice more than earlier information .

The neuroimaging data provided evidence for the role of attention. The authors report that the ventromedial prefrontal cortex (vmPFC) and the ventral striatum primarily process immediate inputs (SV). However, these regions also showed sensitivity to gaze. Specifically, the striatum showed a significant negative correlation with gaze-weighted sampled values [Figure 4C].

The most notable findings appeared in candidate integrator regions. The researchers found that the pre-supplementary motor area (pre-SMA) and the dorsolateral prefrontal cortex (dlPFC) responded to the accumulated value (AV) [Figure 4B]. More importantly, these regions showed significant correlations with gaze-weighted accumulated values [Figure 4D]. This suggests that the pre-SMA and dlPFC track the value as it is modulated by gaze. In this model, the effective value is represented as a product of the stimulus value and the gaze allocation.

Limits of the Integration Map

The study provides a compelling map, but it is not exhaustive. The researchers note that pre-SMA recruitment might be partially confounded by motor actions. Since subjects used a button press to finalize choices, the pre-SMA might reflect motor preparation. This region is closely tied to the intention to move.

Findings regarding the intraparietal sulcus (IPS) and the dlPFC were also somewhat inconsistent. While the dlPFC showed signs of gaze-modulated integration, the IPS did not show significant correlations with gaze-weighted accumulated values in all analyses. This suggests that integration might be distributed across a complex network. Finally, the sample size ($N=23$ for the fMRI analysis) is modest. While the high number of observations per subject helps, larger cohorts are needed to characterize marginal effects in the vmPFC.

The Verdict: A Weighted Reality

The evidence suggests that visual attention is a fundamental component of the decision process. By showing that gaze-modulation persists in the signals of the pre-SMA and dlPFC, the authors suggest that our eyes help weight the evidence. This evidence is then processed by our internal accumulators.

This research moves us closer to understanding how perception and cognition interact. It suggests that the brain does not just "see" and then "think." Instead, it may use the act of seeing to tune the parameters of its internal logic. Future work must determine if this gaze-weighting mechanism is universal across different types of decisions.

Figures from the paper