Consistency of resting-state correlations between fMRI networks and EEG features
Poster No:
2413
Submission Type:
Abstract Submission
Authors:
Marta Xavier1, Inês Esteves1, João Jorge2, Rodolfo Abreu3, Anne-Lise Giraud4, Sepideh Sadaghiani5, Jonathan Wirsich6, Patricia Figueiredo1
Institutions:
1ISR-Lisboa and Department of Bioengineering, Instituto Superior Técnico, Universidade de Lisboa, Lisboa, Portugal, 2CSEM, Swiss Center for Electronics and Microtechnology, Neuchâtel, Switzerland, 3Coimbra Institute for Biomedical Imaging and Translational Research (CIBIT), University of Coimbra, Coimbra, Portugal, 4Department of Neuroscience, University of Geneva, Genève, Switzerland, 5Beckman Institute, Department of Psychology, University of Illinois at Urbana-Champaign, Urbana, IL, 6University of Geneva, Genève, Switzerland
First Author:
Marta Xavier
ISR-Lisboa and Department of Bioengineering, Instituto Superior Técnico, Universidade de Lisboa
Lisboa, Portugal
ISR-Lisboa and Department of Bioengineering, Instituto Superior Técnico, Universidade de Lisboa
Lisboa, Portugal
Co-Author(s):
Inês Esteves
ISR-Lisboa and Department of Bioengineering, Instituto Superior Técnico, Universidade de Lisboa
Lisboa, Portugal
ISR-Lisboa and Department of Bioengineering, Instituto Superior Técnico, Universidade de Lisboa
Lisboa, Portugal
Rodolfo Abreu
Coimbra Institute for Biomedical Imaging and Translational Research (CIBIT), University of Coimbra
Coimbra, Portugal
Coimbra Institute for Biomedical Imaging and Translational Research (CIBIT), University of Coimbra
Coimbra, Portugal
Sepideh Sadaghiani
Beckman Institute, Department of Psychology, University of Illinois at Urbana-Champaign
Urbana, IL
Beckman Institute, Department of Psychology, University of Illinois at Urbana-Champaign
Urbana, IL
Patricia Figueiredo
ISR-Lisboa and Department of Bioengineering, Instituto Superior Técnico, Universidade de Lisboa
Lisboa, Portugal
ISR-Lisboa and Department of Bioengineering, Instituto Superior Técnico, Universidade de Lisboa
Lisboa, Portugal
Introduction:
Simultaneous EEG-fMRI acquisitions have been leveraged to investigate the electrophysiological correlates of fMRI resting-state networks (RSNs). Studies have reported temporal correlations between spontaneous activity of fMRI RSNs and concurrent fluctuations in EEG band-power[1],[2]. However, inconsistencies in results exist, possibly due to varying acquisition conditions, pre-processing, and analysis methodologies. We investigated the consistency of correlations between EEG band-power and fMRI RSN activity across subjects and acquisition conditions, using 3 EEG-fMRI datasets at 1.5, 3, and 7T.
Methods:
Data acquisition. 1.5T: 10 subjects, Siemens 1.5T MR-scanner, 10min eyes-open rest[3]; fMRI with GRE-EPI (TR/TE=2160/30ms, 3.3x3.3x4.0mm3); EEG with two 32-channel MR-compatible amplifiers (BrainAmpMR) and 63 electrodes (BrainCap). 3T: 23 subjects, Siemens 3T MR-Scanner, 30min eyes-closed rest [4]; fMRI with GRE-EPI (TR/TE=2000/50ms, 3mm isotropic); EEG with a 62-channels BrainAmpMRs. 7T: 9 subjects, Siemens 7T MR-scanner, 8min eyes-open rest[5]; fMRI with gradient-echo 2D-EPI sequence (TR/TE=1000/25ms, 2.2mm isotropic); EEG with two 32-channel BrainAmpMRs and 63 electrodes (EasyCap).
Data analysis. fMRI underwent motion correction, nuisance regression of motion and physiological signals, high-pass temporal filtering (0.01Hz) and spatial smoothing. 7 canonical RSNs were identified by group-level ICA and template matching[6], and their fMRI time-series extracted in each subject by regression. EEG was corrected for scanner gradient and pulse artifacts, band-pass filtered (0.3-60Hz), re-referenced, and ICA-denoised. Source reconstructed data[7] was parcellated into the 68 ROIs of the Desikan-Killiany atlas[8]. EEG band-power was estimated with 4s Morlet wavelets for the delta(2-4Hz), theta(5-7Hz), alpha(8-12Hz), beta(15-29Hz), and gamma(3-60Hz) bands. EEG was down-sampled to the fMRI TR and convolved with a hemodynamic response function (HRF), with overshoot delays of 2,4,5,6,8 and 10s. EEG TRs contaminated with motion were excluded[9]. Pearson's correlation was computed between EEG band-power and fMRI RSN time-series for each HRF delay and in each channel/ROI. Grand-mean scalp- and source-space correlation maps were estimated for each dataset. T-tests assessed the significance of the correlations across subjects. Multi-way ANOVA was performed to evaluate the impact on the channel/ROI-averaged EEG-fMRI correlations of factors dataset, EEG space (scalp/source), RSN, frequency-band, and HRF delay.
Data analysis. fMRI underwent motion correction, nuisance regression of motion and physiological signals, high-pass temporal filtering (0.01Hz) and spatial smoothing. 7 canonical RSNs were identified by group-level ICA and template matching[6], and their fMRI time-series extracted in each subject by regression. EEG was corrected for scanner gradient and pulse artifacts, band-pass filtered (0.3-60Hz), re-referenced, and ICA-denoised. Source reconstructed data[7] was parcellated into the 68 ROIs of the Desikan-Killiany atlas[8]. EEG band-power was estimated with 4s Morlet wavelets for the delta(2-4Hz), theta(5-7Hz), alpha(8-12Hz), beta(15-29Hz), and gamma(3-60Hz) bands. EEG was down-sampled to the fMRI TR and convolved with a hemodynamic response function (HRF), with overshoot delays of 2,4,5,6,8 and 10s. EEG TRs contaminated with motion were excluded[9]. Pearson's correlation was computed between EEG band-power and fMRI RSN time-series for each HRF delay and in each channel/ROI. Grand-mean scalp- and source-space correlation maps were estimated for each dataset. T-tests assessed the significance of the correlations across subjects. Multi-way ANOVA was performed to evaluate the impact on the channel/ROI-averaged EEG-fMRI correlations of factors dataset, EEG space (scalp/source), RSN, frequency-band, and HRF delay.
Results:
Each RSN showed a distinct, frequency- and delay-dependent spatial distribution of EEG power correlations, without strong polarization, as illustrated in Fig.1 for the dorsal attention and default mode networks (DAN and DMN). Significant (p<0.05,FDR-corr.) correlations were found (Fig.2) for each dataset, mostly sharing the same direction (positive/negative) across datasets. Consistent observations across datasets included: visual network's positive delta and theta and negative alpha correlations; somatomotor network's positive delta and theta and negative beta and gamma correlations; DAN's negative alpha correlations; frontoparietal network's negative theta and positive gamma correlations; DMN's negative delta and theta and positive alpha and beta correlations. ANOVA showed significant main effects for dataset, RSN, frequency-band, and HRF delay, and various interactions, including a significant triple interaction among dataset, RSN, and frequency-band.
·Fig1.Spatial maps of temporal correlations between EEG band-power (delta, theta, alpha, beta and gamma) and fMRI dorsal attention and default mode networks, for three independent datasets (1.5, 3, 7T)
·Fig2.Spatially averaged temporal correlations between EEG band-power (delta, theta, alpha, beta, and gamma) and seven canonical fMRI resting-state networks, for three independent datasets (1.5, 3, 7T)
Conclusions:
We found consistent temporal correlations between fMRI RSN activity and EEG band-power across subjects in 3 datasets, with variations linked to RSN, EEG frequency-band, and HRF delay, but not EEG space (scalp/source). Despite differences in field strength, subject number, and resting-state conditions, most correlations were qualitatively similar across datasets. Our findings support the consistency of specific EEG-fMRI correlations and highlight the importance of methodological considerations.
Modeling and Analysis Methods:
Connectivity (eg. functional, effective, structural)
Task-Independent and Resting-State Analysis 2
Novel Imaging Acquisition Methods:
BOLD fMRI
EEG
Multi-Modal Imaging 1
Keywords:
Data analysis
Electroencephaolography (EEG)
FUNCTIONAL MRI
1|2Indicates the priority used for review