How reliable are MEG resting-state connectivity metrics?

Area of Science:

• Neuroscience
• Brain Imaging
• Network Science

Background:

• Magnetoencephalography (MEG) offers superior dynamic and spectral resolution for resting-state connectivity compared to fMRI.
• A lack of guidance exists for selecting appropriate network estimation methods in MEG.
• Spatial leakage artifacts significantly impact connectivity measures in MEG.

Purpose of the Study:

• To assess the suitability of popular stationary connectivity measures for resting-state MEG.
• To identify reliable network estimation techniques for MEG data analysis.
• To evaluate the impact of spatial leakage on connectivity measures.

Main Methods:

• Investigated 12 stationary network estimation techniques using resting-state MEG data from the Human Connectome Project.
• Employed empirical criteria of repeatability for individual subjects and reproducibility at the group level.
• Localized magnetic sources using a scalar beamformer.

Main Results:

• Spatial leakage profoundly confounds many connectivity measures, artificially inflating consistency.
• Phase- and coherence-based metrics (e.g., phase lag index, imaginary coherency) exhibited poor test-retest reliability.
• Amplitude envelope correlation and partial correlation demonstrated the highest consistency and robustness.

Conclusions:

• Amplitude envelope correlation and partial correlation are recommended for stationary connectivity estimation in resting-state MEG.
• Researchers must be cautious of spatial leakage artifacts when interpreting MEG connectivity data.
• Careful selection of network estimation methods is crucial for reliable MEG-based brain connectivity research.