Abstracts

Both electroencephalography (EEG) and magnetoencephalography (MEG) are used for interictal spike localization. However, the added value of MEG to the routine EEG recordings remails unclear. Here, we explore the signal-to-noise ratios (SNR) of MEG and EEG with cortical sources and spherically symmetric as well as realistically-shaped conductor models. We employed FreeSurfer to generate a mesh of about 260 000 cortical source locations and surfaces for three-layer boundary-element models (BEM) from 3T T1-weighted anatomical MRI. Spontaneous awake MEG (102 magnetometers and 204 gradiometers) and EEG (70 channels) were acquired (0.5-100 Hz) to estimate the variance of background activity which was considered noise in our simulations. We computed the forward fields for each cortical current dipole as well as for extended sources modelled as patches (radius 10 mm) using two conductor models (sphere model and BEM). Our simulations included: (1) Comparison of the SNRs predicted by the two conductor models for MEG data. (2) Comparison of the MEG and EEG SNRs for dipolar sources with BEM. (3) Comparison of the MEG and EEG SNRs for extended sources with BEM. The median noise standard deviations were: 55 fT/cm (gradiometers), 174 fT (magnetometers), and 9.6 uV (EEG). The BEM predicted a higher SNR at 92% and 96% of all cortical locations for MEG gradiometers and MEG magnetometers, respectively. For both sensor types, the predicted SNR was 2% better at the best BEM locations and 12% better at the best sphere locations. Locations with a higher SNR in BEM than in the sphere model showed a median improvement of [lt]1%. The median improvement of the sphere model for the remaining locations was likewise [lt]1%. The mean for both sets was equal SNR, with a standard deviation less than 1%. For dipole sources, the median difference between MEG and EEG SNRs was 0.25. The range of this difference was -2.6...2.0. MEG had a higher SNR at 67% of cortical locations. The median difference for locations where MEG shows a higher SNR was 0.44 and 0.37 for the locations where the opposite was true. The extended sources showed a similar result: median SNR difference 0.21, higher MEG SNR in 67% of the patches, median differences 0.37 (MEG better) and 0.35 (EEG better). The MEG SNR was not significantly affected by the choice of the conductor model. Under realistic background activity conditions, the MEG SNR is higher than that of EEG at two thirds of cortical source locations. This suggests that MEG may be a better tool for finding subtle interictal spikes in particular locations. Furthermore, the use of a spherical head model may be sufficient for spike identification. Finally, the data indicates that detection of cortically generated spikes is more effective when both MEG and EEG data are available. (Supported by The MIND Institute.)