Constructing EEG Networks to Detect Changes in Resting and Cognitive Brain Dynamics After Anesthesia
Poster No:
2371
Submission Type:
Abstract Submission
Authors:
Elizabeth Teel1, Kira Dolhan2, George Mashour3, Stefanie Blain-Moraes2
Institutions:
1Concordia University, Montréal, Quebec, 2McGill University, Montreal, Quebec, 3University of Michigan, Ann Arbor, MI
First Author:
Co-Author(s):
Introduction:
Constructing graphs from neurophysiological data requires several input parameters1-3 but the optimal parameters to study changes in cognitive function are unknown. Our objective was to compare EEG-derived brain network outcomes from graphs constructed using different input parameters from data recorded at rest and during cognitive testing as healthy adults recovered from anesthetic-induced unconsciousness.
Methods:
Healthy adults (age= 25.3 ± 2.2 years) were randomized into anesthesia (n=9) or control (n=6) groups. Anesthesia subjects were rendered deeply unconscious during a 3-hour intervention period, while controls were awake and minimally active (reading, watching TV, etc.). Subjects completed a cognitive testing battery 30mins before and 0-, 30-, 60-, 90-, and 120-mins post-intervention.4
Subjects wore a 128-channel EEG headset (Electrical Geodesics, Inc.) throughout the study. EEG data were source localized using a forward model and mean source activity was calculated across each of the 82 regions of interest (ROI) defined by the Automated Anatomical Labelling brain atlas.3 Functional connectivity (FC) matrices were derived from sensor- and source-space data. Pearson's correlations were calculated between the amplitude envelope (amplitude envelope coupling; AEC) or instantaneous phase (weighted phase lag index; wPLI) of each combination of pair-wise electrodes or ROIs.3 Graphs were constructed on all FC matrices using weighted and binary methods. For binary graphs, the FC values were set to 0 or 1 based on an orthogonal minimally spanning tree threshold.5 Path length, global efficiency, clustering coefficient (CC), small-world architecture (SWA), modularity, node strength, and betweenness centrality (BC) outcomes were calculated for each graph.
Mixed linear effect models evaluated the effect of parameter, time, and their interaction, with random terms for participant and time. Significance was set at p=0.01.
Subjects wore a 128-channel EEG headset (Electrical Geodesics, Inc.) throughout the study. EEG data were source localized using a forward model and mean source activity was calculated across each of the 82 regions of interest (ROI) defined by the Automated Anatomical Labelling brain atlas.3 Functional connectivity (FC) matrices were derived from sensor- and source-space data. Pearson's correlations were calculated between the amplitude envelope (amplitude envelope coupling; AEC) or instantaneous phase (weighted phase lag index; wPLI) of each combination of pair-wise electrodes or ROIs.3 Graphs were constructed on all FC matrices using weighted and binary methods. For binary graphs, the FC values were set to 0 or 1 based on an orthogonal minimally spanning tree threshold.5 Path length, global efficiency, clustering coefficient (CC), small-world architecture (SWA), modularity, node strength, and betweenness centrality (BC) outcomes were calculated for each graph.
Mixed linear effect models evaluated the effect of parameter, time, and their interaction, with random terms for participant and time. Significance was set at p=0.01.
Results:
No significant interaction effects were observed for signal source and FC type. Main effects of time were observed, with more significant differences for AEC-derived network measures (CC: p=0.0007; SWA: p=0.0009; strength: p=0.0005). A significant interaction effect for modularity was observed (p=0.004) when comparing weighted and binary graphs, with the rate of change greater for binary graphs. There were also five significant main effects of time; the regression line was significantly steeper for binary graphs for four of the five variables (Figure 1). Binary graphs generated using sensor space data and AEC FC matrices best captured the effects of anesthetic-induced unconsciousness in resting data; cognitive graphs were constructed using this combination of input parameters.
Five significant interaction effects (time by task-condition) were observed during the Digit Symbol Substitution Test (CC: p<0.0001, SWA: p<0.0001, modularity: p=0.0007, BC: p=0.002, & strength: p<0.0001). Significant interaction effects were also observed during administrations of the Visual Object Learning Test (modularity: p=0.005, strength: p=0.0009) and Fractal N-Back Test (strength: p=0.0009). Main effects of time were observed for node strength, with outcomes increasing across recovery from unconsciousness when comparing all cognitive tests to resting conditions (p≤0.004, Figure 2).
Five significant interaction effects (time by task-condition) were observed during the Digit Symbol Substitution Test (CC: p<0.0001, SWA: p<0.0001, modularity: p=0.0007, BC: p=0.002, & strength: p<0.0001). Significant interaction effects were also observed during administrations of the Visual Object Learning Test (modularity: p=0.005, strength: p=0.0009) and Fractal N-Back Test (strength: p=0.0009). Main effects of time were observed for node strength, with outcomes increasing across recovery from unconsciousness when comparing all cognitive tests to resting conditions (p≤0.004, Figure 2).
Conclusions:
Under resting conditions, graphs were robust to input parameter selection with few interaction effects observed. Conversely, several main effects of time were found, suggesting that most input parameter selections can identify anesthetic-induced changes in functional brain networks. Network organization differed between cognitive and task-free conditions, with the strongest effects observed during memory tests. These findings add to the growing literature highlighting the potential for network measures to elucidate neural correlates of consciousness and cognitive function.6,7
Higher Cognitive Functions:
Executive Function, Cognitive Control and Decision Making
Modeling and Analysis Methods:
Connectivity (eg. functional, effective, structural)
Novel Imaging Acquisition Methods:
EEG 1
Perception, Attention and Motor Behavior:
Consciousness and Awareness 2
Keywords:
Cognition
Consciousness
Electroencephaolography (EEG)
Memory
Source Localization
Other - Graph Theory
1|2Indicates the priority used for review
·Summary of all significant findings and their directionality for input parameter comparisons (white box= not significant; red box= lower than the reference; green box= greater than the reference).
·Change in node strength across time during rest and for each cognitive test for a single anesthesia participant (A), the anesthesia group (B) and the control group (C).
Provide references using author date format
2. Bullmore, E. and O. Sporns (2009). "Complex brain networks: graph theoretical analysis of structural and functional systems." Nature reviews neuroscience 10(3): 186-198.
3. Duclos, C., C. Maschke, Y. Mahdid, K. Berkun, J. da Silva Castanheira, V. Tarnal, P. Picton, G. Vanini, G. Golmirzaie and E. Janke (2021). "Differential classification of states of consciousness using envelope-and phase-based functional connectivity." NeuroImage 237: 118171.
4. Maier, K. L., A. R. McKinstry-Wu, B. J. A. Palanca, V. Tarnal, S. Blain-Moraes, M. Basner, M. S. Avidan, G. A. Mashour and M. B. Kelz (2017). "Protocol for the Reconstructing Consciousness and Cognition (ReCCognition) Study." Front Hum Neurosci 11: 284.
5. Dimitriadis, S. I., M. Antonakakis, P. Simos, J. M. Fletcher and A. C. Papanicolaou (2017). "Data-driven topological filtering based on orthogonal minimal spanning trees: application to multigroup magnetoencephalography resting-state connectivity." Brain connectivity 7(10): 661-670.
6. Blain-Moraes, S., V. Tarnal, G. Vanini, T. Bel-Behar, E. Janke, P. Picton, G. Golmirzaie, B. J. Palanca, M. S. Avidan and M. B. Kelz (2017). "Network efficiency and posterior alpha patterns are markers of recovery from general anesthesia: a high-density electroencephalography study in healthy volunteers." Frontiers in human neuroscience 11: 328.
7. Wright, L. M., M. De Marco and A. Venneri (2021). "A graph theory approach to clarifying aging and disease related changes in cognitive networks." Frontiers in aging neuroscience 13: 676618.