Authors :
Presenting Author: Mainak Jas, PhD – Massachusetts General Hospital
Seppo Ahlfors, PhD – Massachusetts General Hospital
Teppei Matsubara, MD, PhD – Massachusetts General Hospital
Padmavathi Sundaram, PhD – Massachusetts General Hospital
Catherine Chu, MD, MSC – Massachusetts General Hospital, Harvard Medical School, Kennedy Krieger Institute, Johns Hopkins University
Rationale:
Optically pumped magnetometers (OPMs) are a novel wearable sensor technology for the magnetoencephalogram (MEG) that can be operated at ambient temperature. This allows non-invasive measurement of epileptic activity during natural head movement, an important benefit in children where head movements may be exaggerated. Yet, when the OPM sensor moves in a residual background field (< 1 nT), it can cause artifacts that lead to errors in accurately locating the source of epileptic activity. Thus, while movement in conventional SQUID-based MEG is restricted by physical constraints of the fixed gantry, movement in OPM-based wearable MEG may be restricted by invisible constraints governed by the dipole localization accuracy.
Methods:
To determine the extent of these invisible constraints, we used 15 seconds of continuous head position indicator (cHPI) data from the MNE sample testing data (Fig. 1) and simulated its impact on the estimated dipole localization accuracy in both SQUID-MEG and OPM-MEG. While errors in SQUID-MEG localization are due to changes in the head position relative to the sensor, errors in OPM-MEG localization are due to detection of uncompensated background fields. The source time course was simulated with five randomly occurring interictal spikes (difference of Gaussians with amplitudes 300 nAm and 120 nAm) at the center of the ‘superior temporal’ label of the ‘aparc’ Freesurfer parcellation. Single-axis OPM sensors were determined by projecting the SQUID magnetometers 5 mm above the scalp surface with their sensitive axis oriented normal to the scalp. A 50 pT uniform background field (x, y and z direction) was simulated and its effect was computed based on how it interacted with the sensor’s orientation and position over time. Since the head position is monitored less frequently than the sampling rate, the MEG data was interpolated in the intervening time points. To fairly evaluate the impact of head movement on localization error, we applied pre-processing techniques to remove background field artifacts. In OPM-MEG data, we applied homogenous field correction (HFC) and in SQUID-MEG data, we applied signal space projection (SSP).
Results:
When the head remained still, SQUID-MEG gave the most accurate results (4.76 ± 2.99 mm localization error; Table 1). OPM-MEG dipole localization accuracy was worse compared to SQUID-MEG due to the higher noise floor (15 nT/√Hz compared to 2.5 nT/√Hz). OPM-MEG dipole localization error did not deteriorate with moderate amounts of movement (< 1 cm), especially when using HFC to remove the background field component. However, the error nearly doubled with larger movements. Although OPM-MEG dipole localization degraded with motion, SQUID-MEG errors were limited due to the small amount of movement possible inside the gantry. Conclusions:
While head position monitoring is standard practice in SQUID-MEG, it is typically not considered necessary in OPM-MEG since the sensors are head-mounted. Our findings highlight the importance of continuous head position monitoring with respect to the background field for quality assurance in OPM-MEG also.
Funding:
This work was funded by NIH grant R21NS140619.