Measuring functional connectivity in MEG: a multivariate approach insensitive to linear source leakage.
Brookes MJ, Woolrich MW, Barnes GR. · 2012
Read Measuring functional connectivity in MEG: a multivariate approach insensitive to linear source leakage. by Brookes MJ, Woolrich MW, Barnes GR. (2012) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/europmc-pmc3459100.txt.
Introduction
Brookes, Woolrich, and Barnes presented a method in 2012 for measuring true connections between brain areas in magnetoencephalography data without being fooled by signal leakage. The technique promises a window on the brain's network traffic, but signals smear between nearby reconstructed spots, faking connections. Their approach is multivariate and looks for interactions where one area's rhythm genuinely lags another's.
The method ignores the zero-lag leakage that contaminates other measures and stays spatially unbiased, with a modest false-alarm rate. Simulations and real recordings both behaved as hoped. Once an interaction is found, standard tests can rank which features of the signals drive it. The authors pitch it as a way to capture the richness of these recordings in a single trustworthy statistic.
Catalog record
| CREATOR | Brookes MJ, Woolrich MW, Barnes GR. |
|---|---|
| YEAR | 2012 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 64 KB |
| LICENSE | CC BY 4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Measuring functional connectivity in MEG: A multivariate approach insensitive to linear source leakage By MJ Brookes, MW Woolrich, GR Barnes Published in: Neuroimage DOI: 10.1016/j.neuroimage.2012.03.048 ABSTRACT Abstract A number of recent studies have begun to show the promise of magnetoencephalography (MEG) as a means to non-invasively measure functional connectivity within distributed networks in the human brain. However, a number of problems with the methodology still remain — the biggest of these being how to deal with the non-independence of voxels in source space, often termed signal leakage. In this paper we demonstrate a method by which non-zero lag cortico-cortical interactions between the power envelopes of neural oscillatory processes can be reliably identified within a multivariate statistical framework. The method is spatially unbiased, moderately conservative in false positive rate and removes linear signal leakage between seed and target voxels. We demonstrate this methodology in simulation and in real MEG data. The multivariate method offers a powerful means to capture the high dimensionality and rich information content of MEG signals in a single imaging statistic. Given a significant interaction between two areas, we go on to show how classical statistical tests can be used to quantify the importance of the data features driving the interaction. Keywords: MEG, Neural oscillations, Functional connectivity, Multivariate analysis INTRODUCTION In recent years, neuroscience and neuroimaging have been revolutionised by the discovery of large distributed brain networks, some associated with simple sensory processing (e.g. the motor network) and others associated with higher level function (e.g. the default mode network). Functional magnetic resonance imaging (fMRI) has become a modality of choice for characterisation of functional connectivity within these networks, both at rest and during tasks (see e.g. Smith et al., 2009). However, fMRI is an indirect measure of brain activity based on haemodynamic responses to underlying electrodynamics. Its indirect nature and limited temporal resolution mean that fMRI cannot probe the electrical processes that mediate connectivity or the most rapid temporal fluctuations in network activity. Magnetoencephalography (MEG) detects extra-cranial magnetic fields induced by synchronised neuronal current flow in the brain. In this way it offers advantages over indirect techniques such as fMRI as it can bypass haemodynamics and probe phenomena such as neural oscillations which are thought to be a more direct manifestation of cortical connectivity (Schnitzler and Gross, 2005; Singer, 1999). The utility of MEG (and electroencephalography (EEG)) to measure connectivity has been shown by a large number of studies (Gow et al., 2008; Gross et al., 2001, 2002; Ioannides et al., 2000; Jerbi et al., 2007; Nolte et al., 2004, 2008; Ramnania et al., 2004; Schlögl and Supp, 2006; Schoffelen and Gross, 2009; Tass et al., 1998). Recent work has also begun to show that networks observable in fMRI can also be detected using MEG (Brookes et al., 2011a,b; de Pasquale et al., 2010; Liu et al., 2010). These later studies have measured temporal correlation between power envelopes of neural oscillatory signals at separate brain space voxels. In this way, network maps of correlated signals have been generated and shown to match spatially network maps in fMRI, providing electrophysiological validation of a number of fMRI resting state networks. MEG however has the potential to move beyond fMRI, and elucidate the electrodynamic processes that underpin network formation, without confounds associated with haemodynamic responses. Despite the promise of MEG as a means to measure connectivity, a number of methodological difficulties remain (Schoffelen and Gross, 2009). Using source space projected measurements, the principal limitation is that signals reconstructed at spatially separate brain locations are not necessarily independent; an artifact of the ill posed MEG inverse problem (reconstructing brain space current density measures from scalp based magnetic recordings). This non-independence means that spurious connectivity can exist between two projected MEG time series that is due entirely to signal leakage between projections. A number of techniques to correct for this leakage have been tried; one of the most successful approaches being the use of the imaginary parts of the linear interaction to quantify coherence (Guggisberg et al., 2008; Nolte et al., 2004). However, coherence assumes that signals from two connected brain areas have a fixed phase lag between them, which is not necessarily the case, even for short time windows. Recent studies using envelope correlation methods (Brookes et al., 2011a) have presented techniques to test for leakage via simulation, however such methods are highly computationally intensive and, whilst offering a means to identify spurious connectivity, do not offer a means to eliminate it. A further difficulty in MEG connectivity measurement lies in statistical thresholding of connectivity maps. Using a seed based correlation approach, Pearson correlation between the signal from a seed voxel, and signals from all other voxels is computed. Whilst these correlation values can be converted to statistical quantities (assuming accurate knowledge of temporal degrees of freedom) correction for multiple comparisons across voxels is non trivial due to the inherent spatial smoothness and inhomogeneity in MEG source space images. Finally, the richness of MEG data means that connectivity maps are generally constructed using data that have been frequency filtered into multiple frequency bands of interest (classical alpha, beta and gamma bands for example). This means that a single experiment yields multiple connectivity maps, making results difficult to interpret and leading to a further multiple comparison problem. In this paper, we address some of the limitations currently associated with MEG source space functional connectivity analysis by application of a multivariate statistical framework. Previous work has shown the utility of a multivariate approach both for identification of activity in brain areas elicited by a task (Barnes et al., 2011; Soto et al., 2009) and in connectivity analysis (Soto et al., 2010). In Soto et al. the authors put forward a scheme which eliminates leakage by suppressing any interactions in which the two (canonical) vectors describing power–power coupling between two regions are co-linear. In this work, we employ a similar, but less conservative, multivariate framework to investigate power–power coupling within or between frequency bands, across spatially separate brain areas. Here we address the signal leakage problem by regressing out the linear projection of the seed voxel from the signals at the test voxel. We subsequently assess power–power coupling both within and across multiple frequency bands, using a multivariate statistical framework and a previously described technique (Barnes et al., 2011) to correct for multiple comparisons across voxels. Simulation results show that our statistical analyses yield slightly conservative estimates of false positive rate, invariant of source space, or the number of features used in the connectivity analysis. We also apply our approach to real resting state MEG data showing that our method can identify significant cross hemisphere power–power coupling in the motor cortices. THEORY Our method, comprises the following steps: 1) source space projection of MEG data using beamforming; 2) correction for signal leakage effects by regressing out (orthogonalising with respect to) a linear projection of the seed voxel time course; 3) multivariate statistical analysis of interdependencies between timecourses; 4) correction for multiple comparisons across voxels; and 5) further quantification of the nature of the significant interaction. These five steps are described in detail below. Beamforming Using a beamformer (Barnes et al., 2011; Gross et al., 2001; Robinson and Vrba, 1998; Van Drongelen et al., 1996; Van