External noise removed from magnetoencephalographic signal using independent component analyses of reference channels

Abstract

Background:Many magnetoencephalographs (MEG) contain, in addition to data channels, a set of reference channels positioned relatively far from the head that provide information on magnetic fields not originating from the brain. This information is used to subtract sources of non-neural origin, with either geometrical or least mean squares (LMS) methods. LMS methods in particular tend to be biased toward more constant noise sources and are often unable to remove intermittent noise. New Method:To better identify and eliminate external magnetic noise, we propose performing ICA directly on the MEG reference channels. This in most cases produces several components which are clear summaries of external noise sources with distinct spatio-temporal patterns. We present two algorithms for identifying and removing such noise components from the data which can in many cases significantly improve data quality. Results:We performed simulations using forward models that contained both brain sources and external noise sources. First, traditional LMS-based methods were applied. While this removed a large amount of noise, a significant portion still remained. In many cases, this portion could be removed using the proposed technique, with little to no false positives. Comparison with existing method(s):The proposed method removes significant amounts of noise to which existing LMS-based methods tend to be insensitive. Conclusions:The proposed method complements and extends traditional reference based noise correction with little extra computational cost and low chances of false positives. Any MEG system with reference channels could profit from its use, particularly in labs with intermittent noise sources.

Interference suppression techniques for OPM-based MEG: Opportunities and challenges

“…However, there are automated classification approaches for independent components, for example Corrmap ( Viola et al., 2009 ), IClabel ( Pion-Tonachini et al., 2019 ) and MEGnet ( Treacher et al., 2021 ), which may be adopted as OPM-based MEG systems become more standardised. Reference channel data, ECG and electrooculogram recordings can also be used to guide the automatic removal of components ( Hanna et al., 2020 ). Another disadvantage is that most ICA approaches are probabilistic, meaning that the order of the independent components is arbitrary, and the results may change if re-run.…”

Localization of beta power decrease as measure for lateralization in pre-surgical language mapping with magnetoencephalography, compared with functional magnetic resonance imaging and validated by Wada test

“…Subsequently, data were demeaned using a baseline interval from 0 to 500 ms before stimulus onset, and bandpass filtered from 1 to 45 Hz. Residual external noise was removed using independent component analysis (ICA) of reference channels (Hanna et al, 2020 ). Successively, ICA was administered once again, and the topography and time course of components were visually inspected for eye blinks and other remaining artifacts.…”

Extended Signal-Space Separation Method for Improved Interference Suppression in MEG

“…These methods may utilize signals from reference sensors measuring the external interference outside of the main sensor array [ 1 ] together with e.g. modelling based on independent component analysis (ICA) [ 2 ], [ 3 ]. Interference can also be suppressed without dedicated reference sensors; such methods can operate: i) in the spatial domain, such as signal-space projection (SSP) [ 4 ], signal-space separation (SSS) [ 5 ], [ 6 ] and generalized sidelobe canceller [ 7 ], ii) in the temporal domain, such as temporal filtering, iii) in the spatio-temporal domain, such as spatiotemporal SSS (tSSS) [ 8 ], [ 9 ] and dual signal-subspace projection (DSSP) [ 10 ], or iv) in the frequency domain, such as spectral signal-space projection (S3P) [ 11 ].…”