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

TL;DR

This study addresses the gap in task-based fMRI analyses by quantifying information processing rather than solely activations or functional connectivity. It introduces active information storage ($A_X = I(X_t; X_{<t})$), transfer entropy ($T_{X\rightarrow Y} = I(Y_t; X_{<t} | Y_{<t})$), and net synergy ($S_{X\rightarrow Y} = T_{X\rightarrow Y} - I(X_{<t}; Y_t)$) within a cross mutual information framework, using a reference distribution that concatenates resting-state and task data. Applying this to the Human Connectome Project's $N$-back task, the authors show global increases in information storage and transfer with memory load, a general shift toward redundant information exchange, and selective regional increases in synergy, particularly within control and visual modules. Importantly, individual working-memory performance correlates with greater AIS changes in frontal regions and a stronger redundancy shift, highlighting the behavioral relevance of information-theoretic dynamics. The framework demonstrates a principled method to quantify cognitive information processing in task-based fMRI with potential implications for understanding neural coding robustness and flexible network interactions.

Abstract

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.

Figures (7)

• Figure 1: Calculation of condition-specific measures. A) The time series for some local (instantaneous) measure i. Examples include the BOLD signal, local AIS (for a region), and local MI (for an edge). B) Visualisation of the task design matrix where the blue line indicates the time points corresponding to the rest, 0-back and 2-back condition.
• Figure 2: Conventional measures for studying task fMRI data. For the 2-back vs rest condition (left) and 2-back vs 0-back condition (right): A) Change in mean activity, averaging over subjects. B) Change in conditional MI, averaged over subjects. This was calculated for using a conventional approach with the task data defining the reference distribution. C) Change in cross MI, averaged over subjects. This was calculated using the new approach using the resting-state and task data for the reference distribution. All plots have been thresholded to only show contrasts with $p$-values$\,<0.05$. $P$-values were calculated using group-level GLM sign-flip permutations, taking the maximum $t$-statistic over regions/edges to control for multiple comparisons.
• Figure 3: Information-theoretic analysis reveals modular reorganisation during the 2-back task (2-back vs rest). A) Change in AIS (calculated using Equation \ref{['eq:ais']}) averaged over subjects. We see that AIS increases globally, especially in visual modules. B) Change in TE (calculated using Equation \ref{['eq:te']}) averaged over subjects and edges that correspond to Yeo modules. We see that TE decreases across most modules but increases in VIS A, DAN A and CON B. C) Change in net synergy (calculated using Equation \ref{['eq:net-synergy']}) averaged over subjects and edges that correspond to Yeo modules. We see that net synergy shifts towards redundancy across modules, except for VIS B and CON A. Only contrasts with $p<0.05$ are shown. $P$-values were calculated using group-level GLM sign-flip permutations, taking the maximum $t$-statistic over regions/edges to control for multiple comparisons.
• Figure 4: Working memory enhances storage and transfer but induces a global redundancy shift (2-back vs 0-back). A) Change in AIS (calculated using Equation \ref{['eq:ais']}) averaged over subjects.. We see that AIS increases globally. B) Change in TE (calculated using Equation \ref{['eq:te']}) averaged over subjects and edges that correspond to Yeo modules. We see that TE increases broadly across modules except notably VIS B and CON A, which show decreases. C) Change in net synergy (calculated using Equation \ref{['eq:net-synergy']}) averaged over subjects and edges that correspond to Yeo modules. We see that net synergy shifts to redundancy except notably within CON A, which shows increased synergy. Only contrasts with $p<0.05$ are shown. $P$-values were calculated using group-level GLM sign-flip permutations, taking the maximum $t$-statistic over regions/edges to control for multiple comparisons.
• Figure 5: Within-module information processing changes predicts task performance. Regression coefficients ($\beta$) for predicting response accuracy with: A) Change in AIS (2-back vs rest) for each subject. We see greater change in AIS, particularly in frontal regions is linked to higher accuracy. B) Change in net synergy (2-back vs rest) for each subject. We see a stronger shift towards redundancy is linked to better performance. Only coefficients with $p<0.05$ are shown. $P$-values were calculated using group-level GLM sign-flip permutations, taking the maximum $t$-statistic over regions/edges to control for multiple comparisons.