Magnetoencephalography (MEG) is increasingly being used to study brain function because of its excellent temporal resolution and its direct association with brain activity at the neuronal level. One possible cause of error in the analysis of MEG data comes from the fact that participants, even MEG-experienced ones, move their head in the MEG system. Head movement can cause source localization errors during the analysis of MEG data, which can result in the appearance of source variability that does not reflect brain activity. The MEG community places great importance in eliminating this source of possible errors as is evident, for example, by recent efforts to develop head casts that limit head movement in the MEG system. In this work we use software tools to identify, assess and eliminate from the analysis of MEG data any possible correlations between head movement in the MEG system and widely-used measures of brain activity derived from MEG resting-state recordings. The measures of brain activity we study are a) the Hilbert-transform derived amplitude envelope of the beamformer time series and b) functional networks; both measures derived by MEG resting-state recordings. Ten-minute MEG resting-state recordings were performed on healthy participants, with head position continuously recorded. The sources of the measured magnetic signals were localized via beamformer spatial filtering. Temporal independent component analysis was subsequently used to derive resting-state networks.
Significant correlations were observed between the beamformer envelope time series and head movement. The correlations were substantially reduced, and in some cases eliminated, after a participant-specific temporal high-pass filter was applied to those time series. Regressing the head movement metrics out of the beamformer envelope time series had an even stronger effect in reducing these correlations. Correlation trends were also observed between head movement and the activation time series of the default-mode and frontal networks. Regressing the head movement metrics out of the beamformer envelope time series completely eliminated these correlations. Additionally, applying the head movement correction resulted in changes in the network spatial maps for the visual and sensorimotor networks. Our results a) show that the results of MEG resting-state studies that use the above-mentioned analysis methods are confounded by head movement effects, b) suggest that regressing the head movement metrics out of the beamformer envelope time series is a necessary step to be added to these analyses, in order to eliminate the effect that head movement has on the amplitude envelope of beamformer time series and the network time series and c) highlight changes in the connectivity spatial maps when head movement correction is applied.