Estimating measures of information processing during cognitive tasks using functional magnetic resonance imaging

Cognition is increasingly framed in terms of information processing, yet most fMRI analyses focus on activation or functional connectivity rather than quantifying how information is stored and transferred. To remedy this problem, we propose a framework for estimating measures of information processing: active information storage (AIS), transfer entropy (TE), and net synergy from task-based fMRI. AIS measures information maintained within a region, TE captures directed information flow, and net synergy contrasts higher-order synergistic to redundant interactions. Crucially, to enable this framework we utilised a recently developed approach for calculating information-theoretic measures: the cross mutual information. This approach combines resting-state and task data to address the challenges of limited sample size, non-stationarity and context in task-based fMRI. We applied this framework to the working memory (N-back) task from the Human Connectome Project (470 participants). Results show that AIS increases in fronto-parietal regions with working memory load, TE reveals enhanced directed information flows across control pathways, and net synergy indicates a global shift to redundancy. This work establishes a novel methodology for quantifying information processing in task-based fMRI.

💡 Research Summary

This paper introduces a novel framework for quantifying information processing in task‑based functional magnetic resonance imaging (fMRI) by estimating three information‑theoretic measures: active information storage (AIS), transfer entropy (TE), and net synergy. The authors address two major challenges inherent to task fMRI—short, non‑stationary recordings and limited sample size—by employing a recently developed cross‑mutual information (cross‑MI) approach. Cross‑MI constructs a reference probability distribution from the concatenation of each subject’s resting‑state and task data, allowing the test distribution (the task segment alone) to be evaluated against a broader baseline. This strategy mitigates the bias of conventional conditional MI, which conditions on the task alone and therefore lacks context about typical brain activity.

Using the Human Connectome Project (HCP) dataset, the authors analyzed the N‑back working‑memory paradigm (0‑back vs. 2‑back) in 470 healthy adults. Pre‑processing included bias‑field and motion correction, band‑pass filtering (0.01–0.08 Hz), global signal regression, and blind deconvolution of the hemodynamic response function (HRF) to reduce temporal smoothing effects. After parcellating the brain into 333 regions of the Gordon atlas, the authors z‑scored and detrended each time series.

Information‑theoretic quantities were estimated with the Kraskov‑Stögbauer‑Grassberger (KSG) nearest‑neighbor estimator implemented in the JIDT toolbox. To avoid inflation from autocorrelation, a Theiler window of 15 samples was applied. AIS was computed using two past time points (k=2), capturing how much of a region’s current state can be predicted from its recent history. TE was calculated as the conditional mutual information between a source’s past and a target’s present, given the target’s own past, thereby quantifying directed information flow. Net synergy was derived via partial information decomposition, measuring the balance between synergistic and redundant contributions of multiple inputs.

Results showed that AIS increased with working‑memory load in fronto‑parietal regions, notably dorsolateral prefrontal cortex and posterior parietal cortex, indicating heightened maintenance of past information during demanding tasks. TE analyses revealed amplified directed flows along established control pathways, especially from prefrontal to parietal nodes, suggesting more coordinated information transfer under higher load. Net synergy analyses demonstrated a global shift toward redundancy (decreased synergy) during the 2‑back condition, implying that the brain adopts a more conservative, overlapping coding strategy when cognitive demands rise.

The authors discuss how these findings align with predictive coding, the free‑energy principle, and integrated information theory, emphasizing that AIS, TE, and net synergy together provide a richer description of neural computation than activation or static functional connectivity alone. Limitations include the intrinsic temporal resolution of fMRI, potential residual errors from HRF deconvolution, and the dependence of cross‑MI on the quality of the reference distribution. Future work is suggested to combine this framework with higher‑temporal‑resolution modalities (e.g., MEG/EEG), refine HRF modeling, and test the approach across diverse cognitive paradigms.

In summary, the study delivers a robust methodological advance for extracting quantitative information‑processing metrics from task‑based fMRI, opening new avenues for testing computational theories of cognition and for characterizing individual differences in neural information dynamics.