Skip to main page content
U.S. flag

An official website of the United States government

Dot gov

The .gov means it’s official.
Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

Https

The site is secure.
The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
. 2025 Mar 20;9(1):421-446.
doi: 10.1162/netn_a_00441. eCollection 2025.

Validating MEG estimated resting-state connectome with intracranial EEG

Jawata Afnan  1   2   3 , Zhengchen Cai  3 , Jean-Marc Lina  4   5   6 , Chifaou Abdallah  1   2   3   7 , Giovanni Pellegrino  8 , Giorgio Arcara  9 , Hassan Khajehpour  7 , Birgit Frauscher  7 , Jean Gotman  3 , Christophe Grova  1   3   4   10
Affiliations

Validating MEG estimated resting-state connectome with intracranial EEG

Jawata Afnan et al. Netw Neurosci. .

Abstract

Magnetoencephalography (MEG) is widely used for studying resting-state brain connectivity. However, MEG source imaging is ill posed and has limited spatial resolution. This introduces source-leakage issues, making it challenging to interpret MEG-derived connectivity in resting states. To address this, we validated MEG-derived connectivity from 45 healthy participants using a normative intracranial EEG (iEEG) atlas. The MEG inverse problem was solved using the wavelet-maximum entropy on the mean method. We computed four connectivity metrics: amplitude envelope correlation (AEC), orthogonalized AEC (OAEC), phase locking value (PLV), and weighted-phase lag index (wPLI). We compared spatial correlation between MEG and iEEG connectomes across standard canonical frequency bands. We found moderate spatial correlations between MEG and iEEG connectomes for AEC and PLV. However, when considering metrics that correct/remove zero-lag connectivity (OAEC/wPLI), the spatial correlation between MEG and iEEG connectomes decreased. MEG exhibited higher zero-lag connectivity compared with iEEG. The correlations between MEG and iEEG connectomes suggest that relevant connectivity patterns can be recovered from MEG. However, since these correlations are moderate/low, MEG connectivity results should be interpreted with caution. Metrics that correct for zero-lag connectivity show decreased correlations, highlighting a trade-off; while MEG may capture more connectivity due to source-leakage, removing zero-lag connectivity can eliminate true connections.

Keywords: Connectivity; Intracranial EEG; MEG source imaging; Resting state connectome; Source leakage.

Plain language summary

The ill-posed nature and low spatial resolution of EEG/magnetoencephalography (MEG) source imaging affects functional connectivity estimates, which become more complicated in the resting state due to the low signal-to-noise ratio. Several connectivity metrics have been proposed to address source leakage by removing zero-lag connectivity, although this can eliminate true neuronal zero-lag connections. Intracranial EEG (iEEG) is the gold standard for validating noninvasive measurements. In this study, we validated MEG-estimated connectivity for healthy subjects using the iEEG atlas of normal brain activity (Frauscher et al., 2018) as ground truth at a group level. We employed two amplitude-based metrics and two phase-based metrics. Our findings highlight how MEG connectivity compares with the iEEG atlas and provide valuable insights for resting-state EEG/MEG connectomic studies, particularly in the choice of connectivity metrics.

PubMed Disclaimer

Conflict of interest statement

Competing Interests: The authors have declared that no competing interests exist.

Figures

<b>Figure 1.</b>
Figure 1.
The iEEG connectome consists of connectivity metrics between pairs of channels, obtained from a total of 110 patients. In each of 45 MEG connectomes, all connections originated from a single subject. To generate a new MEG connectome comparable with the original iEEG connectome, MEG subjects were randomly chosen to contribute connections between ROIs while preserving spatial information. This process was repeated for all ROI pairs, resulting in a bootstrap-resampled MEG connectome, mimicking the same subjects’ group distribution as our original iEEG connectome. The spatial Pearson correlation between the original iEEG connectome and the bootstrap-resampled MEG connectome was computed. This overall process was iterated 5,000 times, yielding 5,000 correlation values.
<b>Figure 2.</b>
Figure 2.
Connectivity averaged across frequency bands estimated by MEG and iEEG calculated using (A) AEC, (B) OAEC, (C) PLV, and (D) wPLI*. For iEEG, we considered all 1,278 iEEG ROI pairs available from all patients of the iEEG atlas and showed the averaged connectivity across six frequency bands. For MEG, we considered 1,278 virtual iEEG ROI pairs from each of the 45 subjects and showed the averaged connectivity across 45 subjects and six frequency bands. The median value of each distribution is displayed.
<b>Figure 3.</b>
Figure 3.
(A) Distribution of cross-modal correlations as well as the null distribution (red) between MEG and iEEG for six frequency bands calculated for AEC (blue) and OAEC (green). (B) The medians of the distribution of cross-modal correlations are shown in the bar plot. The correlation was considered significant if its overlap with the null range was less than 2.5% (equivalent to a 5% two-tailed threshold, with 2.5% in each tail). The frequency bands that showed significantly higher correlations than the null distribution are marked with an asterisk (*).
<b>Figure 4.</b>
Figure 4.
The median of the distribution of cross-modal correlations is depicted, considering all connections, intrahemispheric connections, and interhemispheric connections for (A) AEC and (B) OAEC. The correlation was considered significant if its overlap with the null range was less than 2.5% (equivalent to a 5% two-tailed threshold, with 2.5% in each tail). Frequency bands with significantly higher correlations than the null distribution are marked with an asterisk (*).
<b>Figure 5.</b>
Figure 5.
Distribution of cross-modal correlations and the null distribution (red) between MEG and iEEG for six frequency bands calculated for PLV (blue) and wPLI* (green). (A) The medians of the distribution of cross-modal correlations were shown as a bar plot. (B) The correlation was considered significant if its overlap with the null range was less than 2.5% (equivalent to a 5% two-tailed threshold, with 2.5% in each tail). The frequency bands that showed significantly higher correlations than the null distribution were marked with an asterisk (*).
<b>Figure 6.</b>
Figure 6.
(A) Eccentricity of iEEG channels shown on the brain cortex with 80% transparency to ensure all deep iEEG channels are visible. (B) Distribution of the distances between ROI pairs for all pairs exhibiting either an eccentricity > 85 mm (top) or < 85 mm (bottom). The cross-modal correlation between MEG and iEEG for two groups (both eccentricity values > 85 mm in blue and both eccentricity values < 85 mm in red) for AEC (C), OAEC (D), PLV (E), and wPLI* (F). The correlation was considered significant if its overlap with the null range was less than 2.5% (equivalent to a 5% two-tailed threshold, with 2.5% in each tail). The frequency bands that showed significantly higher correlations than the null distribution were shown with an *.
<b>Figure 7.</b>
Figure 7.
(A) AEC and OAEC as a function of distance between two ROIs plotted for iEEG and MEG in the beta band. (B) The distribution of differences between OAEC and AEC (OAEC minus AEC) for MEG and iEEG.
<b>Figure 8.</b>
Figure 8.
Distribution of cross-modal spatial correlations between MEG and iEEG connectomes in the beta band obtained using AEC and OAEC (obtained from 5,000 bootstrap MEG samples), as we increase the minimum number of subjects from one to five in each ROI pair. Increasing the minimum number of subjects in each ROI pair (as shown on the left) decreases the available coverage of the iEEG connectome from 44% to 3% (as shown on the right). For example, the bottom row displays histograms of the correlations between MEG and iEEG connectomes when the iEEG connectome was created with ROI pairs that include at least one patient, covering 44% of the connectome.
See this image and copyright information in PMC

References

    1. Abdallah, C., Hedrich, T., Koupparis, A., Afnan, J., Hall, J. A., Gotman, J., … Grova, C. (2022). Clinical yield of electromagnetic source imaging and hemodynamic responses in epilepsy: Validation with intracerebral data. Neurology, 98(24), e2499–e2511. 10.1212/WNL.0000000000200337, - DOI - PMC - PubMed
    1. Afnan, J., Cai, Z., Lina, J.-M., Abdallah, C., Delaire, E., Avigdor, T., … Grova, C. (2024). EEG/MEG source imaging of deep brain activity within the maximum entropy on the mean framework: Simulations and validation in epilepsy. Human Brain Mapping, 45(10), e26720. 10.1002/hbm.26720, - DOI - PMC - PubMed
    1. Afnan, J., von Ellenrieder, N., Lina, J.-M., Pellegrino, G., Arcara, G., Cai, Z., … Grova, C. (2023). Validating MEG source imaging of resting state oscillatory patterns with an intracranial EEG atlas. NeuroImage, 274, 120158. 10.1016/j.neuroimage.2023.120158, - DOI - PubMed
    1. Amblard, C., Lapalme, E., & Lina, J.-M. (2004). Biomagnetic source detection by maximum entropy and graphical models. IEEE Transactions on Biomedical Engineering, 51(3), 427–442. 10.1109/TBME.2003.820999, - DOI - PubMed
    1. Arnulfo, G., Hirvonen, J., Nobili, L., Palva, S., & Palva, J. M. (2015). Phase and amplitude correlations in resting-state activity in human stereotactical EEG recordings. NeuroImage, 112, 114–127. 10.1016/j.neuroimage.2015.02.031, - DOI - PubMed