Maxfilter 2

Continuous data were preprocessed offline using the temporal extension of the signal space separation method (tSSS)^{72 (link)} implemented in Maxfilter 2.2 (Elekta-Neuromag). Briefly, tSSS subtracts external magnetic noise from the MEG recordings, corrects for head movements, and interpolates bad channels. Subsequent analyses were performed using the MatlabR2014B and FieldTrip toolbox version 20,170,911^{73 (link)}. Recordings were down-sampled to 500 Hz and segmented into epochs time-locked to picture presentation from 1000 ms before image onset to 1000 ms after image onset.
A semi-automatic procedure was employed to remove epochs containing electromyographic artifacts, SQUID jumps, and flat signals. Finally, a fast independent component analysis (fast ICA) was used to identify components reflecting blinks and electrocardiographic artifacts^{74 (link)}. Two participants were discarded from the final analysis due to a high number of blinking/muscular artifacts in the MEG recording (e.g. leading to > 40 kept trials in some conditions). Thus, the final MEG analysis was performed on a reduced sample of nineteen participants.

We analyzed resting-state MEG data of 604 participants (304 male and 300 females) aged 18–88 years old provided in the Cam-CAN database. For each subject, temporal Signal Space Separation (tSSS, MaxFilter 2.2, Elekta Neuromag, Oy, Helsinki, Finland) was applied to remove noise from external source and from HPI coil. The sampling frequency of recorded data was 1 kHz with a high-pass filter of 0.03 Hz. Independent component analysis had been performed by Cam-CAN to exclude signal components associated with eye movements. The resting-state recordings were each at least 8 min and 40 s in duration. The MEG sensor array consisted of 306-channel Elekta Neuromag Vectorview (102 magnetometers and 204 planar gradiometers). We analyzed the magnetometer sensors only- although we have analyzed all directions, the magnetometer time-series provided better and more pronounced avalanche distributions. This gave us as a result 102 channels of time-series activity for each participant. More details about the data acquisition pipeline can be found in Taylor et al. (2017 (link)). Computations were carried out using MATLAB (R2020a, The Mathworks Natick, MA), and the Python programming language (Python Software Foundation. Python Language Reference, version 3.8. available at https://www.python.org/), GNU Parallel (Tange, 2011 ).

The MEG data we analyzed was shared by the Cambridge Centre for Ageing and Neuroscience (CamCAN). CamCAN funding was provided by the UK Biotechnology and Biological Sciences Research Council (grant number BB/H008217/1), together with support from the UK Medical Research Council and University of Cambridge, UK. This data was obtained from the CamCAN repository (available at https://www.mrc-cbu.cam.ac.uk/datasets/camcan/; Shafto et al., 2014 (link); Taylor et al., 2017 (link)) and was conducted in accordance with the Helsinki declaration and approved by the Cambridgeshire 2 Research Ethics Committee (reference: 10/H0308/50).
MEG data was collected using a 306 sensor VectorView MEG system (Electa Neuromag, Helsinki). The 306 sensors consisted of 102 magnetometers and 204 planar gradiometers. The data were sampled at 1000 Hz and highpass filtered at 0.3 Hz. This data was run through temporal signal space separation (tSSS; Taulu et al., 2005 (link); MaxFilter 2.2, Elekta Neuromag Oy, Helsinki, Finland) to remove noise from external sources and to help correct for head movements (location of the head was continuously estimated using Head Position Indicator coils). MaxFilter was also used to remove the 50 Hz line noise and also to automatically detect and reconstruct noisy channels.

MEG recordings were acquired at a sampling rate of 1 kHz using a 306-channel whole-head neuromagnetometer (Triux, Elekta Oy, Helsinki, Finland) for about 60 minutes at rest. The subject’s head position inside the MEG helmet was continuously monitored by five head position identification coils located on the scalp. The locations of these coils, together with three anatomical landmarks (nasion, right and left preauriculars), and additional scalp points were digitized before the recording by means of a 3D digitizer (FASTRAK, Polhemus, Colchester, VT). The scalp surface points were used for the co-registration with the patient’s anatomical MRI. The raw MEG data were pre-processed off-line with the temporally extended Signal Space Separation method (tSSS) implemented in the Maxfilter 2.2 (Elekta Neuromag Oy, Helsinki, Finland) to suppress external interferences and correct for head movements (Taulu and Hari 2009 (link)), and next filtered at 0.1–100 Hz.
MRI images were acquired by means of a volumetric T1-weighted sequence on a 3T MR scanner (Philips Healthcare BV, Best, NL).

To remove external magnetic noise from the MEG recordings, data were preprocessed offline using the signal‐space‐separation method implemented in Maxfilter 2.1 (Elekta Neuromag).32 MEG data were also corrected for head movements, and substitutions were made for bad channels using interpolation algorithms implemented in the software. Subsequent analyses were performed using Matlab R2010 (Mathworks®, Natick, MA). Heartbeat and EOG artifacts were detected using independent component analysis (ICA) and linearly subtracted from recordings. The ICA decomposition was performed using the Infomax algorithm implemented in the Fieldtrip toolbox.33