Multi-channel whole-head OPM-MEG: Helmet design and a comparison with a conventional system

Abstract

Magnetoencephalography (MEG) is a powerful technique for functional neuroimaging, offering a non-invasive window on brain electrophysiology. MEG systems have traditionally been based on cryogenic sensors which detect the small extracranial magnetic fields generated by synchronised current in neuronal assemblies, however, such systems have fundamental limitations. In recent years, non-cryogenic quantum-enabled sensors, called optically-pumped magnetometers (OPMs), in combination with novel techniques for accurate background magnetic field control, have promised to lift those restrictions offering an adaptable, motion-robust MEG system, with improved data quality, at reduced cost. However, OPM-MEG remains a nascent technology, and whilst viable systems exist, most employ small numbers of sensors sited above targeted brain regions. Here, building on previous work, we construct a wearable OPM-MEG system with 'whole-head' coverage based upon commercially available OPMs, and test its capabilities to measure alpha, beta and gamma oscillations. We design two methods for OPM mounting; a flexible (EEG-like) cap and rigid (additively-manufactured) helmet. Whilst both designs allow for high quality data to be collected, we argue that the rigid helmet offers a more robust option with significant advantages for reconstruction of field data into 3D images of changes in neuronal current. Using repeat measurements in two participants, we show signal detection for our device to be highly robust. Moreover, via application of source-space modelling, we show that, despite having 5 times fewer sensors, our system exhibits comparable performance to an established cryogenic MEG device. While significant challenges still remain, these developments provide further evidence that OPM-MEG is likely to facilitate a step change for functional neuroimaging.

Keywords: Beta; Gamma; MEG; Magnetoencephalography; OPM; Optically pumped magnetometer.

Fig. A1.. Simulation showing how beamforming is affected by co-registration error in a rigid helmet or flexible cap.

A) Example of co-registration errors in the two cases. For the rigid helmet (upper panel), we assume we know the relative sensor locations and orientations accurately, so co-registration error is systematic across the helmet. For the flexible cap (lower panel), all sensor locations and orientations are acquired from the co-registration process independently, meaning co-registration error is random. The black circles show true sensor positions, and the coloured circles show measured locations with co-registration error. The left hand panel shows zero co-registration error; the right hand panel shows an error of 4 mm translation and a 4 degree rotation (about the origin). B) Summary measures of time course correlation (left) pseudo-Z-statistic (centre) and localisation accuracy (right) plotted as a function of co-registration error. Red shows the rigid cap, blue shows the flexible cap. The different lines show different regularisation values (zero (solid line); 10% (dashed line); 40% (dotted line)). Note x-axes represent both translation and rotation – e.g. an error of 5 means 5 mm and 5°.

Fig. A2.. Summary of results in Alpha band.

A) Beamformer pseudo-T-statistical images averaged over all 6 experimental runs for both participants. B) Glass brain, with the centre of the ellipsoids showing average peak location across runs. The size of the ellipsoids represents the standard deviation of the peak locations – and hence variability of localisation across runs. C) (Left) Ellipsoid volumes averaged across participants, (middle) Image consistency (correlation between pseudo-T-statistical images) collapsed across both participants, (right) Signal-to-Noise ratios for the three different systems in the alpha band. D) (Left) Input SNR at the best sensor, (middle) Output SNR measured in beamformer projected data, (right) Ratio of output to input SNR.

Fig. 1.. The OPM-MEG system.

A) Schematic diagram of the whole system. B) Magnetically shielded room. C) Flexible (EEG style) cap. D) Rigid additively manufactured helmet. Both helmets contain push-fit clips to house the 2nd generation QuSpin OPMs (shown inset in C) and D)).

Fig. 2.. Schematic diagram showing co-registration algorithm

for A) flexible cap and B) rigid helmet.

Fig. 3.. Test-re-test co-registration errors:

A) Location error for rigid helmet. B) Orientation error for rigid helmet. C) Location error for flexible helmet. D) Orientation error for flexible cap. Colder colours indicate a more reliable co-registration at that sensor.

Fig. 4.. Cortical coverage:

The plots on the top row show the sensor locations over the scalp. Lower plots show the norm of the forward fields, for each dipole location in the brain. Left, centre and right shows the rigid helmet, the flexible cap, and the cryogenic system, respectively. Note, the magnitude of the colour axis is different for each system.

Fig. 5.. Sensor space results for Subject 1.

Upper, middle and lower panels show results from the rigid, flexible, and cryogenic systems respectively. In all cases, the sensor space topography plots show estimated SNRs of the beta and gamma signals for each sensor; the six plots show six repeated measures in Subject 1. The line plots on the right-hand side show the oscillatory envelopes of the beta and gamma effects extracted from the sensor with the largest signal-to-noise ratio (all six runs are overlaid). An equivalent Figure for Subject 2 is shown in Supplementary Material, Figure S1.

Fig. 6.. Spatial signature of beta and gamma responses.

A) Beamformer pseudo-T-statistical images averaged over all 6 experimental runs for Subject 1. B) Glass brain, with the centre of the ellipsoids showing average peak location across runs. The size of the ellipsoids represents the standard deviation of the peak locations – and hence random variability of localisation across runs. C) Ellipsoid volumes averaged across participants. D) Image consistency (correlation between pseudo-T-statistical images) collapsed across both participants.

Fig. 7.. Beamformer-estimated (source space) neural oscillatory activity.

A) Oscillatory envelopes and time frequency spectra extracted from the locations of peak beta (left) and gamma (right) modulation. Top, centre and bottom rows show rigid, flexible and cryogenic systems respectively. B) Signal-to-Noise Ratios for the three different systems in the beta and gamma bands.

Fig. 8.. Helmet design comparison:

A) Input SNR at the best sensor. B) Output SNR measured in beamformer projected data. C) Ratio of output to input SNR.