How reliable are MEG resting-state connectivity metrics?

Abstract

MEG offers dynamic and spectral resolution for resting-state connectivity which is unavailable in fMRI. However, there are a wide range of available network estimation methods for MEG, and little in the way of existing guidance on which ones to employ. In this technical note, we investigate the extent to which many popular measures of stationary connectivity are suitable for use in resting-state MEG, localising magnetic sources with a scalar beamformer. We use as empirical criteria that network measures for individual subjects should be repeatable, and that group-level connectivity estimation shows good reproducibility. Using publically-available data from the Human Connectome Project, we test the reliability of 12 network estimation techniques against these criteria. We find that the impact of magnetic field spread or spatial leakage artefact is profound, creates a major confound for many connectivity measures, and can artificially inflate measures of consistency. Among those robust to this effect, we find poor test-retest reliability in phase- or coherence-based metrics such as the phase lag index or the imaginary part of coherency. The most consistent methods for stationary connectivity estimation over all of our tests are simple amplitude envelope correlation and partial correlation measures.

Keywords: Connectome; Functional connectivity; MEG; Magnetic field spread; Network analysis; Source leakage.

Fig. 1

Mean alpha-band network matrices at group level. (A) Views of single rows from the network matrices in B. Left, connection strength to the posterior cingulate cortex (seed ROI number 38, shown in green). Upper right, connection strengths to the right motor cortex (seed ROI number 24). Lower right, connection strengths to right occipital cortex (seed ROI number 24). (B) Mean network matrices, and the relative edge-level consistency, for a subset of four of the metrics under test: the absolute value of coherency between ROIs, the imaginary part of coherency, and the amplitude envelope correlation (AEC), both with and without a multivariate correction for source leakage. Lower triangles, in red, show the mean network matrix over all 183 sessions in the dataset. The colourbar to the side indicates the scale. Correlations have been converted to Z-values before averaging. The upper triangles, in blue, indicate the relative contribution of each edge to the overall consistency of group-level network estimation, computed with Eq. (13) (c.f. Fig. 2A). The two network matrices which are susceptible to source leakage (AEC and Coherence, top row) are dominated, even at group level, by a few strong edges (highlighted with arrows), and these drive the high consistency of network estimation. (C) The parcellation of 39 fMRI-derived ROIs used for this connectivity analysis (reproduced with permission from Colclough et al. (2015)). (D) Similarity between group-level network matrices. The heat map shows the high correlations between network matrices inferred at the group-level using the undirected network measures that are either immune to source leakage bias, or have been corrected for source leakage bias (starred). AEC - amplitude envelope correlation; PAEC - partial amplitude envelope partial correlation; PLI - phase lag index; wPLI - weighted phase lag index; PLV - phase locking value; MI - mutual information of phases. Coh - band-averaged coherence; IMC - band-averaged imaginary component of coherency; PCoh - partial coherence; IMPC - partial imaginary coherence.

Fig. 2

Consistency of network matrix estimation. (A) Stability of group-level inference. Correlations between network matrices inferred from separate halves of the HCP dataset, using resting-state recordings in the alpha band. The dataset was randomly partitioned in half 100 times, and the distribution of correlations between the network edge strengths in each half was produced over the bootstrapped samples. (B) Within-subject consistency of network inference. Correlations between alpha-band network matrices inferred from each of three resting-state sessions from 61 subjects of the HCP dataset. (C) Between-subject consistency of network inference. Correlations between alpha-band network matrices inferred from 61 subjects of the HCP dataset. In all cases, the violin plots show smoothed histograms of these distributions, the median marked by a white cross. Correlations have been converted to Z-values using Fisher's transformation, but labelled with the original rho value, for clearer display. Darker distributions with background shading identify metrics which are known to suffer from source-leakage confounds. Metrics with a star indicate that a multivariate source leakage correction has been applied before computation of the network measure. AEC - amplitude envelope correlation; PAEC - partial amplitude envelope partial correlation; PLI - phase lag index; wPLI - weighted phase lag index; PLV - phase locking value; PSI - phase slope index; MI - mutual information of phases. Coh - band-averaged coherence; IMC - band-averaged imaginary component of coherency; PCoh - partial coherence; IMPC - partial imaginary coherence; PDC - partial directed coherence.