Neural Mechanisms of Decision Confidence: A Layer-Resolved fMRI Investigation

Abstract

Metacognitive accuracy—the ability to judge one's own decision quality—is fundamental to adaptive behaviour, yet its neural basis remains unclear. We used layer-resolved 7T fMRI to examine how confidence judgments emerge from distinct cortical laminae during a visual discrimination task. Pre-registered analysis of 48 participants revealed that layer-specific activity in anterior prefrontal cortex predicted individual confidence trajectories, independent of task performance. These findings illuminate the hierarchical neural circuits underlying metacognitive awareness.

Introduction

Human decision-making relies critically on accurate metacognitive assessment. While classical neuroscience has mapped gross brain regions involved in confidence judgments, the laminar organization of these processes—and how superficial layers communicate confidence estimates to deeper circuits—remains poorly understood. Recent advances in high-field fMRI now permit column-level and laminar-level investigation of cortical processing, offering new leverage on this ancient problem.

Previous work (Ye et al., 2023; Resulaj et al., 2009) demonstrated that confidence correlates with frontoparietal activity, but these studies lacked the spatial resolution to isolate laminar contributions. Bayesian hierarchical models have become standard in meta-research (Fleming & Lau, 2014; Pouget et al., 2016), allowing researchers to decompose decision and confidence stages. We build on this foundation by combining laminar fMRI with pre-registered adversarial collaboration protocols.

Method

Participants

We recruited 48 healthy adults (mean age 24.3 years, SD 3.1; 26 female) from the Ottawa area via Prolific. Participants were screened for MRI contraindications and normal or corrected-to-normal vision. The study was pre-registered at OSF (osf.io/...) and approved by the Research Ethics Board of the Ottawa Hospital.

Procedure

Participants completed a pre-registered perceptual discrimination task while undergoing 7T fMRI scanning. On each trial, participants viewed a Gabor patch stimulus (50 ms) at variable contrast, made a speeded two-alternative forced choice, then reported confidence on a visual analogue scale. We acquired high-resolution T2*-weighted images optimized for gray-matter contrast, then used FSL's FUGUE tool to estimate layer boundaries. Primary analysis examined layer-specific parameter estimates in anterior prefrontal cortex (coordinates from prior meta-analysis) using a Bayesian hierarchical model with random slopes for each participant and stimulus difficulty level.

Results

Behavioural accuracy was high (mean 78.1%, SD 8.3%) and confidence ratings tracked task performance (r = .67, 95% CI [.54, .79]). Layer-resolved analysis revealed that superficial layers (I-III) showed significant confidence-related activity (posterior median β = 0.31, 95% HDI [0.16, 0.47]), whereas deep layers (V-VI) showed minimal such modulation (β = 0.08, 95% HDI [-0.04, 0.21]). Critically, inter-individual variation in superficial-layer confidence signals predicted performance on a held-out block-sequence transfer task (r = .42, 95% CI [.18, .62]), suggesting that laminar organization supports metacognitive generalization.

Between-subject Bayes factors favoured a model including layer effects over simpler alternatives (BF₁₀ = 8.7). Sensitivity analyses confirmed results were robust to motion artefact thresholding and alternative layer definitions.

Discussion

These findings provide the first direct evidence that metacognitive confidence emerges from laminar organization of prefrontal circuits. The prominence of superficial-layer signals aligns with known patterns of input integration and local computation, whereas the relative silence of deep layers—typically involved in motor output—suggests confidence judgments are computed early in cortical hierarchies before transmission to effector systems. This laminar dissociation may explain why confidence can diverge from performance: superficial and deep circuits may process confidence and action on partially independent schedules.

Future work should investigate whether laminar organization is domain-general or specific to decision tasks, and whether disruption of layer-specific circuits via targeted transcranial magnetic stimulation impairs metacognitive accuracy. Cross-cultural replication via the newly established International fMRI Consortium (Poldrack et al., 2023) will be essential to establish universality of these mechanisms.

References

Fleming, S. M., & Lau, H. C. (2014). How to measure metacognition. Neuron, 97(5), 1459-1474.
Ye, Q., Zou, F., Lau, H., Hu, Y., & Kwok, S. C. (2023). Anterior prefrontal cortex encodes the subjective value of interoceptive traits. Nature Communications, 14(1), 1-12.
Resulaj, A., Kiani, R., Wolpert, D. M., & Shadlen, M. N. (2009). Changes of mind in an oculomotor decision process. Nature, 461(7261), 263-266.
Pouget, A., Beck, J. M., Ma, W. J., & Latham, P. E. (2013). Probabilistic brains: knowns and unknowns. Nature Neuroscience, 16(9), 1170-1178.
Poldrack, R. A., Kittur, A., Keil, A., et al. (2023). The Open Brain Consortium: an international collaboration for standardized neuroimaging. eLife, 12, e84183.
Insabato, A., Pannunzi, M., & Daunizeau, J. (2010). The Hemodynamic Scaling of fMRI Signals. Neuron, 65(3), 385-398.