Modelling optically pumped magnetometer interference as a mean (magnetic) field

Abstract

Here we propose that much of the magnetic interference observed when using optically pumped magnetometers can be modeled spatially as a mean (magnetic) field. We show that this approximation reduces sensor level variability and substantially improves statistical power. This model does not require knowledge of the underlying neuroanatomy nor the sensor positions. It only needs information about the sensor orientation. Due to the model’s low rank there is little risk of removing substantial neural signal. However, we provide a framework to assess this risk for any sensor number, design or subject neuroanatomy. We find that the risk of unintentionally removing neural signal is reduced when multi-axis recordings are performed. We validated the method using a binaural auditory evoked response paradigm and demonstrated that the mean field correction increases reconstructed SNR in relevant brain regions in both the spatial and temporal domain. Considering the model’s simplicity and efficacy, we suggest that this mean field correction can be a powerful preprocessing step for arrays of optically pumped magnetometers.

Using OPMs to measure neural activity in standing, mobile participants

“…Data were recorded using a 16-bit precision analog-to-digital converter (National Instruments) with a sample rate of 6000 Hz. For our setup, the two sensitive OPM axes were oriented both radial and tangential to the head, and had the effect of increasing the dimensionality of the data and facilitating spatial interference suppression methods (Brookes et al, 2021; Tierney et al, 2020). This resulted in 90 channels of OP-MEG data.…”

“…Next, the data were cleaned using mean-field correction which approximates magnetic interference as a spatially constant field (see Tierney et al, 2020 for further details). This field is then removed from the data via linear regression.…”

“…S1. The maximum field change in each trial was reduced from an average of 654.7 pT to 242.1 pT following head position tracking regression; to 87.8 pT following mean-field correction (Tierney et al, 2020); and to 9.8 pT following the high-pass filter.…”

“…Fortunately, these artefacts typically have spatiotemporal properties distinct from neuromagnetic fields. Signal-processing techniques based on spatial and temporal filtering, such as beamforming (Brookes et al, 2008; Hillebrand & Barnes, 2005; Van Veen et al, 1997), can therefore be used for interference suppression during participant movement (Tierney et al, 2020).…”

“…Due to their reliability, AEFs are often used in the context of MEG for benchmarking purposes (e.g. Sekihara et al, 2002; Taulu & Hari, 2009; Tierney et al, 2020). AEFs are also a low-frequency phenomenon (1-40 Hz), that overlap in the frequency domain with movement artefacts (<10Hz).…”