Inferring laminar origins of MEG signals with optically pumped magnetometers (OPMs)

1643 

Abstract Submission 

Saskia Helbling¹

¹Ernst Strungmann Institute for Neuroscience, Frankfurt am Main, Germany

Saskia Helbling  
Ernst Strungmann Institute for Neuroscience
Frankfurt am Main, Germany

Magnetoencephalography (MEG) is a non-invasive technique that measures the tiny magnetic fields generated by neural currents in the brain. Conventional MEG operates with superconducting SQUID magnetometers that must be immersed in liquid helium for cooling, which introduces a substantial gap between the sensors and the scalp. Optically-pumped magnetometers (OPMs) are new, highly sensitive magnetometers that operate without the need for cryogenic cooling and can be placed close to the scalp, substantially improving sensitivity to cortical sources (Boto et al., 2016, Iivanainen et al., 2017). The typical spatial resolution achieved by conventional MEG is not sufficient for laminar inference. One strategy to distinguish between deep and superficial sources is to use high-precision forward models that exploit the small variations in the so-called lead fields between deep and superficial sources (Bonaiuto et al., 2018a and 2018b). On-scalp OPM-MEG has been suggested to further improve the discriminability of laminar sources. To investigate this, we simulate cortical sources at deep and superficial layers and infer their laminar origin using OPM sensor arrays with varying numbers of sensors and measurement axes.

We simulated OPM array designs with sampling distances between 25 and 55 mm in 10 mm increments and single, dual and triaxial sensor configurations, following Tierney et al. (2020). Current dipole sources were added to vertices of the white and the pial surface meshes reconstructed by FreeSurfer. For each surface, we randomly selected 60 vertices as cortical source locations. At each source location, a 20 Hz sinusoidal dipolar source patch was added to each of the 200 trials modelled per cortical source location. The dipolar sources were active for 400 ms. We determined the laminar origin of the simulated sources using a whole-brain and an ROI-based analysis, equivalent to those described in Bonaiuto et al. (2018), and used empirical Bayes beamformer (EBB) and multiple sparse priors (MSP) source reconstruction approaches. We compared the classification accuracy and bias of each analysis and source inversion algorithm using two-sided binomial tests and used logistic regression to evaluate changes in classification performance across sampling densities, number of axes and co-registration errors.

We find that for an OPM-MEG sensor array with an inter-sensor distance of 35 mm we are able to achieve highly accurate laminar inferences at SNRs of -30 dB or higher. However, at high SNRs, we observed a bias towards the deep surface for the whole-brain analysis when combined with the EBB source reconstruction approach. For the EBB approach, classification accuracy significantly increased with decreasing inter-sensor distances and increasing number of measurement axes at moderate to high SNRs. The MSP approach exhibited classification performance close to ceiling levels for SNRs of -30 dB or higher. We note that laminar inference was possible at a low sensor counts of 32 at sufficiently high SNRs. We observed a steep decrease in classification accuracy with increasing co-registration errors for the EBB approach, where laminar inference was not feasible anymore at a co-registration error of 4 mm. For the MSP approach, classification accuracy remained high across co-registration errors but still decreased significantly with increasing co-registration errors.

Overall, we find that laminar inference with OPM arrays is possible at relatively low sensor counts at moderate to high SNRs. Laminar inference improved with increasing spatial sampling densities and number of measurement axes. Challenges persist in dealing with biases at very low SNRs and a distinct bias toward the deep surface when combining EBB source reconstruction with the free energy whole-brain analysis. Adequate SNR through appropriate trial numbers and shielding, as well as precise co-registration, is crucial for reliable laminar inference with OPMs.

EEG/MEG Modeling and Analysis ¹

MEG ²

Cortical Layers

Data analysis

MEG

Modeling

Open-Source Code

Preprint

Source Localization