An empirical assessment of neural-oscillatory mechanisms underlying flexible communication
Poster No:
1844
Submission Type:
Abstract Submission
Authors:
Varun Madan Mohan1, Caio Seguin2, Thomas Varley3, Andrew Zalesky1
Institutions:
1The University of Melbourne, Melbourne, Victoria, 2Department of Psychological and Brain Sciences, Indiana University, Bloomington, IN, 3Indiana University, Bloomington, IN
First Author:
Co-Author(s):
Introduction:
The brain relies on fast and accurate communication, or information routing between its various components, for healthy function [1]. However, despite its importance, a lack of methods to systematically investigate empirical communication patterns have led to a notable lack of consensus regarding routing principles in the brain, even with multiple models of communication having been proposed [2,3,4].
In this work, we develop an information-theoretic method to study directional communication patterns in neural recordings. First, we demonstrate the method on a network of three nodes with simple stochastic dynamics. We then apply it to MEG recordings, and exemplify its use in studying principles of inter-regional communication. Specifically, we correlate our measure of information flow with coherence in the alpha and gamma bands: neural-oscillatory metrics previously theorised to shape communication patterns [3,4].
In this work, we develop an information-theoretic method to study directional communication patterns in neural recordings. First, we demonstrate the method on a network of three nodes with simple stochastic dynamics. We then apply it to MEG recordings, and exemplify its use in studying principles of inter-regional communication. Specifically, we correlate our measure of information flow with coherence in the alpha and gamma bands: neural-oscillatory metrics previously theorised to shape communication patterns [3,4].
Methods:
NETWORK MODEL
A system comprising three nodes, with Linear Stochastic Model (LSM) dynamics [5] was defined [Fig2A]. A Poisson process caused two of the regions to "pulse" randomly, with an average frequency of 0.2Hz (source). The noise level of the LSM was systematically varied relative to the pulse amplitude. The system's activity was recorded for 200 seconds.
EMPIRICAL DATA
Resting-state MEG scans of 30 healthy subjects (13 males, age range 22-35), were obtained from the Human Connectome Project (HCP) [6,7], source localised using the dynamical-SPM method in Brainstorm [8] as per Brainstorm's tutorial on HCP data, and parcellated using the Schaefer 7-Network 100 region atlas [8].
ESTIMATING COMMUNICATION PATTERNS
The rationale behind the developed method is that deviations from mean activity of a region should cause a measurable delayed downstream effect on the activities of the rest of the network. Additionally, focusing the information flow estimation around the significant deviations (communication events) excludes the contribution of noisy segments that can cloud true communication patterns.
The pipeline is [Fig1]:
1) Regional activities are epoched into 10 second segments; 2) a region is chosen as a source, its activity is z-scored, and timepoints where |z|>3 are marked as events; 3) a "communication window" of 1-second is placed starting at the event. A similar window, delayed in proportion to inter-regional distance, is placed at all other brain regions (targets); 4) The mutual information (MI) between the source and target within the defined windows, conditional upon the target's past, is computed. This captures the directional information flow.
In addition to these steps, to test the communication principles:
5) Within the communication window, alpha and gamma band coherence is computed; 6) The Pearson-r values between the MI and coherences is computed, quantifying the dependence of information flow on neural-oscillatory relationships.
A system comprising three nodes, with Linear Stochastic Model (LSM) dynamics [5] was defined [Fig2A]. A Poisson process caused two of the regions to "pulse" randomly, with an average frequency of 0.2Hz (source). The noise level of the LSM was systematically varied relative to the pulse amplitude. The system's activity was recorded for 200 seconds.
EMPIRICAL DATA
Resting-state MEG scans of 30 healthy subjects (13 males, age range 22-35), were obtained from the Human Connectome Project (HCP) [6,7], source localised using the dynamical-SPM method in Brainstorm [8] as per Brainstorm's tutorial on HCP data, and parcellated using the Schaefer 7-Network 100 region atlas [8].
ESTIMATING COMMUNICATION PATTERNS
The rationale behind the developed method is that deviations from mean activity of a region should cause a measurable delayed downstream effect on the activities of the rest of the network. Additionally, focusing the information flow estimation around the significant deviations (communication events) excludes the contribution of noisy segments that can cloud true communication patterns.
The pipeline is [Fig1]:
1) Regional activities are epoched into 10 second segments; 2) a region is chosen as a source, its activity is z-scored, and timepoints where |z|>3 are marked as events; 3) a "communication window" of 1-second is placed starting at the event. A similar window, delayed in proportion to inter-regional distance, is placed at all other brain regions (targets); 4) The mutual information (MI) between the source and target within the defined windows, conditional upon the target's past, is computed. This captures the directional information flow.
In addition to these steps, to test the communication principles:
5) Within the communication window, alpha and gamma band coherence is computed; 6) The Pearson-r values between the MI and coherences is computed, quantifying the dependence of information flow on neural-oscillatory relationships.
Results:
In the network model, region 2's activity is a mixture of internal dynamics, and inputs from regions 1 and 3. Our method accurately identifies information outflow from both the sources (1 and 3), even at high noise levels. Additionally, estimating the MI around the significant events shows stronger information flow from regions 1 and 3 to 2, compared to when the MI is computed over the entire regional timeseries [Fig2B,C].
Application of the method on source-level MEG and quantifying the dependence of information flow with the alpha and gamma coherence reveals a clear regional variation in the spectral metrics that best correlate with communication. Information flow correlates maximally with alpha coherence in posterior regions, and with gamma coherence in temporal regions [Fig2D,E].
Application of the method on source-level MEG and quantifying the dependence of information flow with the alpha and gamma coherence reveals a clear regional variation in the spectral metrics that best correlate with communication. Information flow correlates maximally with alpha coherence in posterior regions, and with gamma coherence in temporal regions [Fig2D,E].
Conclusions:
In this study, we present an information-theoretic method of estimating directed empirical communication patterns, and showcase its applicability in studying inter-regional communication principles, by revealing a heterogeneous dependence of information flow on neural-oscillatory relationships.
Modeling and Analysis Methods:
Connectivity (eg. functional, effective, structural)
EEG/MEG Modeling and Analysis 2
Methods Development 1
Task-Independent and Resting-State Analysis
Other Methods
Keywords:
Computational Neuroscience
Data analysis
MEG
Modeling
Other - Communication
1|2Indicates the priority used for review
Provide references using author date format
[2] Seguin, C. (2023), ‘Brain network communication: concepts, models and applications’, Nature Reviews Neuroscience, vol. 24, no. 9, pp. 557-574.
[3] Fries, P. (2015), ‘Rhythms for cognition: communication through coherence’, Neuron, vol. 88, no. 1, pp. 220-235.
[4] Jensen, O. (2010), ‘Shaping functional architecture by oscillatory alpha activity: gating by inhibition’, Frontiers in human neuroscience, vol. 4, 186.
[5] Van Essen, D. C. (2013), ‘The WU-Minn human connectome project: an overview’, Neuroimage, vol. 80, pp. 62-79.
[6] Larson-Prior, L. J. (2013), ‘Adding dynamics to the Human Connectome Project with MEG’, Neuroimage, vol. 80, pp. 190-201.
[7] Tadel, F. (2011), ‘Brainstorm: a user-friendly application for MEG/EEG analysis’, Computational intelligence and neuroscience, vol. 2011, pp. 1-13.
[8] Schaefer, A. (2018), ‘Local-global parcellation of the human cerebral cortex from intrinsic functional connectivity MRI’, Cerebral cortex, vol. 28, no. 9, pp. 3095-3114.