^{1}

^{*}

^{1}

^{1}

^{2}

^{1}

^{3}

^{4}

^{1}

^{1}

^{4}

^{5}

^{1}

^{6}

^{7}

^{8}

^{9}

^{1}

^{4}

^{8}

^{*}

^{1}

^{2}

^{3}

^{4}

^{5}

^{6}

^{7}

^{8}

^{9}

Edited by: Tetsuo Kida, National Institute for Physiological Sciences, Japan

Reviewed by: Tamer Demiralp, Istanbul University, Turkey; Marco Leite, University College London, UK; David Reutens, University College London (UCL), UK

*Correspondence: Qingbao Yu

Vince D. Calhoun

This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

The topological architecture of brain connectivity has been well-characterized by graph theory based analysis. However, previous studies have primarily built brain graphs based on a single modality of brain imaging data. Here we develop a framework to construct multi-modal brain graphs using concurrent EEG-fMRI data which are simultaneously collected during eyes open (EO) and eyes closed (EC) resting states. FMRI data are decomposed into independent components with associated time courses by group independent component analysis (ICA). EEG time series are segmented, and then spectral power time courses are computed and averaged within 5 frequency bands (delta; theta; alpha; beta; low gamma). EEG-fMRI brain graphs, with EEG electrodes and fMRI brain components serving as nodes, are built by computing correlations within and between fMRI ICA time courses and EEG spectral power time courses. Dynamic EEG-fMRI graphs are built using a sliding window method, versus static ones treating the entire time course as stationary. In global level, static graph measures and properties of dynamic graph measures are different across frequency bands and are mainly showing higher values in eyes closed than eyes open. Nodal level graph measures of a few brain components are also showing higher values during eyes closed in specific frequency bands. Overall, these findings incorporate fMRI spatial localization and EEG frequency information which could not be obtained by examining only one modality. This work provides a new approach to examine EEG-fMRI associations within a graph theoretic framework with potential application to many topics.

Graph theory-based analysis is a powerful technique to characterize the architecture of human brain networks (Avena-Koenigsberger et al.,

Different imaging techniques are sensitive to different aspects of brain dynamics. For example, functional magnetic resonance imaging (fMRI) measures the highly localized hemodynamic response throughout the brain, with a good spatial resolution (about 2–3 mm) but relatively poor temporal resolution. Electroencephalography (EEG) measures cortical electrical activity with a much higher temporal resolution, but its poor spatial resolution precludes precise anatomical identification of underlying neural sources. FMRI and EEG therefore represent complementary imaging signals, and combining concurrently collected data is a particularly useful way to examine brain dynamics over a broad range of spatial and temporal scales (Menon and Crottaz-Herbette,

For coupling concurrent EEG-fMRI data, a popular approach is to analyze correlations between fMRI voxel time-series and EEG spectral power fluctuations (Valdes-Sosa et al.,

Dynamic connectivity over time is an important feature in functional brain networks (Hutchison et al.,

The aim of this study is to explore graph properties of concurrent EEG-fMRI multi-modal brain connectivity. Static and dynamic EEG-fMRI brain graphs are built using concurrently collected data from 25 healthy subjects during eyes open and eyes closed. Graph nodes are represented by EEG electrodes and fMRI components identified using group independent component analysis (ICA; Calhoun et al.,

Twenty-five healthy subjects (age: 29 ± 8; 8 females) were recruited via advertisements at the University of New Mexico and by word-of-mouth. Each individual had normal or corrected to normal vision and hearing. Prior to inclusion in the study, participants were screened to ensure they were free from DSM-IV Axis I or Axis II psychopathology [assessed using the SCID (First et al.,

Simultaneous EEG-fMRI data were recorded while individuals rested first with their eyes closed (8.5 min), and then with their eyes open (8.5 min). Individuals were instructed to relax, lie still, and remain awake for the duration of each recording.

EEG was recorded with a 32-channel BrainAmp MR-compatible system (Brainproducts, Munich, Germany) and a BrainCap electrode cap (Falk Minow Services, Herrsching-Breitbrunn, Germany). The Ag/AgCI electrodes were placed according to the international 10–20 system. Electrocardiogram (ECG) and eye movement (EOG) signals were recorded in separate channels, reducing the number of scalp electrodes to 30. The reference channel was placed at FCz. The impedance of each electrode was kept lower than 5 KΩ using conductive and abrasive electrode paste. The EEG signals were sampled at 5 KHz. To avoid temporal jitter, the EEG amplifier and fMRI were synchronized using an in-house device.

Functional MRI brain images were acquired with a Siemens Sonata (Siemens, Malvern PA) scanner at 1.5 T by means of a T2^{*}-weighted echo planar imaging sequence with the following parameters: repeat time (TR) = 2 s, echo time (TE) = 39 ms, field of view = 224 mm, acquisition matrix = 64 × 64, flip angle = 80⋅, voxel size = 3.5 × 3.5 × 3 mm, gap = 1 mm, 27 slices, ascending acquisition. FMRI scans consisted of 256 volumes for each condition (eyes open and eyes close).

EEG data were preprocessed in Matlab (

The preprocessed EEG data were variance normalized, segmented into 2 s epochs (resulting in 256 epochs, i.e., an epoch that corresponds to each concurrently recorded fMRI volume) and converted to the frequency domain by the fast Fourier transform (FFT). The spectral power was averaged within 5 frequency bands (delta: 1–4 Hz; theta: 4–8 Hz; alpha: 8–13 Hz; beta: 13–30 Hz; low gamma: 30–35 Hz) for each epoch, resulting in 10 matrices (time [256] × electrodes [30]) of time series of spectral power for each subject (

FMRI data were preprocessed using SPM5 (

One spatial group ICA (Calhoun et al.,

A correlation matrix R was constructed with elements (r_{ij}) representing Pearson correlation coefficients computed using the 30 EEG electrodes' spectral time-courses and the 54 fMRI ICs' time-courses. This process was repeated for the five EEG frequency bands and the two conditions (EO and EC). When computing the correlation between EEG and fMRI signals, following previous studies (Goldman et al.,

Consequently, undirected static connectivity EEG-fMRI graphs were built from each of the ^{+}) and negative (^{−}) connection graphs were built based on

Dynamic EEG-fMRI graph analysis was performed by calculating correlation matrices along successive sliding windows of the matrix EF (256 × 84; width, L = 20 TRs, in steps of 1 TR; Allen et al.,

Here we use a window width of 20 TRs (40 s) based on a previous study indicating that cognitive states may be correctly identified with as little as 30–60 s of data (Shirer et al.,

Connectivity strength (CS), clustering coefficient (CC), and global efficiency (GE) are three basic and important graph metrics which measure the functional segregation and integration of brain networks (Rubinov and Sporns,

We denote

Recent fMRI studies showed that fluctuations of time-varying functional brain connectivity gives rise to discrete highly-organized patterns that may emerge or dissolve over time, which are called connectivity states (Cribben et al.,

Figure

Figure

For the global level graph metrics of positive connection networks, a five (frequency band: delta, theta, alpha, beta, low gamma) × two (eyes condition: open, closed) compound symmetry repeated measures ANOVA shows that the main effect of frequency band is significant (

For the global level graph metrics of negative connection networks, a five (frequency band: delta, theta, alpha, beta, low gamma) × two (eyes condition: open, closed) compound symmetry repeated measures ANOVA shows that the main effects of frequency band and eyes condition are significant (

Figures

For nodal level graph metrics of positive connection graphs, the main effect of eyes condition is significant (FDR correction,

For nodal level graph metrics of negative connection graphs, the main effect of eyes condition is significant (FDR correction,

To visually display the difference between eyes conditions of the brain network, as an example, we show the values of the nodal graph measures and the pattern of the connections from a visual component node to all of the other graph nodes in the five frequency bands during eyes open and eyes closed conditions in positive and negative connection graphs in Figures

Figures

Variance (VAR) and amplitude of low frequency (LFA) [0–0.025 Hz] oscillations of the time varying global level graph metrics are computed. For time-varying positive connection graphs, five (frequency band: delta, theta, alpha, beta, low gamma) × two (eyes condition: open, close) compound symmetry repeated measure ANOVA shows that the main effect of eyes condition is not significant on VAR and LFA of all dynamic graph measures. The main effect of frequency band is significant (

For nodal level dynamic graph metrics of positive connection networks, the main effect of eyes condition on the VAR and LFA of all three dynamic measures is significant (FDR correction,

Consistent with previous dynamic fMRI connectivity studies (Allen et al.,

In the present study, concurrent EEG-fMRI resting state data collected during eyes open and eyes closed conditions are used to build multi-modal brain graphs. FMRI data are decomposed with group ICA into ICNs and corresponding time courses. EEG signals are segmented into 2 s-epochs and the spectral power is computed and averaged within five frequency bands (delta, theta, alpha, beta, and low gamma) for each segment. EEG-fMRI brain graphs are built by computing the correlations between and among fMRI ICA time courses and EEG spectral power time courses. Connectivity strength, local efficiency, and global efficiency are calculated for both static graphs, which are estimated using the full length of time courses, and dynamic graphs, which are estimated using a sliding window method. Five (frequency band: delta, theta, alpha, beta, low gamma) × two (eyes condition: open, close) compound symmetry repeated measure ANOVA and paired

In early studies which combine EEG and fMRI, EEG signals are traditionally separated into five frequency bands: delta, theta, alpha, beta, and gamma (Laufs et al.,

The majority of previous studies which combine EEG and fMRI data use correlation or general linear modeling (GLM) to link fluctuations between multiple EEG frequency bands and fMRI voxels (Bridwell and Calhoun,

In this work, we separate the EEG data into five frequency bands as in previous studies. However, in addition to computing the correlations between EEG signals and fMRI BOLD signals, we compute EEG-fMRI multi-modal brain graphs in which EEG nodes provide high temporal resolution information and fMRI nodes provide high spatial resolution. The finding that graph metrics show differences across frequency bands (the main effect of frequency band is significant) is consistent with the hypothesis that different EEG frequencies are associated with different BOLD activities.

Within this study, we characterize different graph properties between eyes open and eyes closed conditions. In static positive connection EEG-fMRI graphs, nodal level graph metrics are higher during the eyes closed condition in three brain components which belong to somatomotor, visual, and auditory areas. In negative connection graphs, a visual component shows different nodal level graph measures (for all three metrics) between eyes conditions. The LFA and VAR of dynamic nodal graph measures of two cognitive control components are higher during eyes closed than eyes open for positive connection networks. In general, these findings are consistent with and add to previous studies demonstrating differences in BOLD amplitudes and functional connectivity across the two conditions (McAvoy et al.,

Multiple recent brain imaging studies suggest that the functional brain connectivity is not stationary but changes over minute-to-minute intervals (Hutchison et al.,

Notably, the findings that alteration of graph measures of specific fMRI nodes across eyes conditions occurs in particular EEG frequency bands provide new electrophysiological signatures of functional brain connectivity examined in fMRI data, and imply that the graph-theory based analysis is powerful to assess the associations between EEG and fMRI. However, a few potential methodological limitations need to be discussed. Graph metrics may depend in part on the methods used to identify nodes. Thus, it is worth considering the difference between ICA-based and anatomically-based approaches. Within fMRI, brain graph nodes are often formed using predefined anatomical templates such as automated anatomical labeling (AAL; Tzourio-Mazoyer et al.,

It's important to note that graph measures were computed based on the formula defined for a single-modal (classical) graph. The graph metrics may be interpreted in the same way as traditional graphs. But it is unclear how global level graph measures within a multi-modal graph would be affected by the distribution of edges and nodes from different modalities. This limitation is shared by each of the two conditions examined within the present study (eyes open and eyes closed). Thus, the observation of graph metric differences here motivates further studies which extend the single-modal (classical) graph formula for EEG-fMRI (Zhang et al.,

We believe that this work provides an important beginning step in characterizing EEG-fMRI associations within a graph theoretical framework. Both static and dynamic EEG-fMRI graphs are built in five EEG frequency bands on concurrently collected EEG-fMRI data while individuals rested with eyes open and eyes closed. Differences in global and nodal level static graph metrics including connectivity strength, local efficiency, and global efficiency, are revealed among frequency bands and between eyes conditions. Dynamic properties of the graph metrics also show differences between eyes conditions. These findings incorporate spatial location (provided by fMRI) information and frequency (delta, theta, alpha, beta, and gamma bands provided by EEG) information in identifying graph properties that differ between brain states (i.e., eyes open vs. eyes closed) by linking electro-hemodynamic responses. This paper proposes a novel approach for assessing associations among concurrent EEG and fMRI measures which couples electoral and hemodynamic BOLD signals in the brain at a network level.

QY designed the study; analyzed and interpreted the data; drafted and revised the manuscript; gave final approval. LW designed the study; collected, analyzed, and interpreted the data; revised the manuscript and gave final approval. DB analyzed and interpreted the data; revised the manuscript and gave final approval. EE analyzed and interpreted the data; revised the manuscript and gave final approval. YD analyzed and interpreted the data; revised the manuscript and gave final approval. HH analyzed and interpreted the data; revised the manuscript and gave final approval. JC collected, analyzed, and interpreted the data; revised the manuscript. PL analyzed and interpreted the data; revised the manuscript and gave final approval. JS collected, analyzed, and interpreted the data; revised the manuscript. GP designed the study; interpreted the data; revised the manuscript and gave final approval. VC designed the study; interpreted the data; revised the manuscript and gave final approval.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

This work is supported by the National Institutes of Health (NIH) grants including a COBRE grant (P20GM103472), R01 grants (R01EB005846, 1R01EB006841, 1R01DA040487, REB020407, EB000840), and other grants (5P20RR021938, R37 MH43775 PI: Pearlson). This work is partly supported by the “100 Talents Plan” of Chinese Academy of Sciences, the state high-tech development plan of China (863) 2015AA020513 (PI: Sui J), the Strategic Priority Research Program of the Chinese Academy of Sciences (XDB02060005), Chinese NSF (81471367, PI: Sui J; 81471738, PI: Liu P), and natural science foundation of Shanxi (2016021077, PI: YHD).

The Supplementary Material for this article can be found online at: