Global sensitivity of MEG source analysis to tissue conductivity uncertainties

The influence of inter-individual variations of tissue conductivities on MEG source analysis is generally assumed to be small in comparison to EEG source analysis and the resulting effects on MEG source analysis have therefore been investigated much less. We perform an in-depth analysis of this influence, so that the results of this study are of importance to better interpret results of MEG source analysis and to improve applications that make use of both EEG and MEG, e.g., combined source analysis. MEG forward solutions for dipole sources regularly distributed in the gray matter compartment were simulated in a detailed five-compartment head model for three realistic sensor configurations using the FEM multipole approach. Subsequently, a generalized polynomial chaos approach (gPC) was employed to calculate MEG leadfields for varying tissue conductivities. Based on these gPC expansions, the sensitivity of MEG forward solutions towards tissue conductivity uncertainties and the influence on MEG source analysis was investigated. In general, our study shows that the influence of tissue conductivity uncertainties on MEG forward solutions and source analysis is clearly weaker than for the EEG, and confirms that MEG is fairly robust against tissue conductivity uncertainties. For all three investigated sensor configurations, we find very similar sensitivity distributions. A strong influence of tissue conductivity uncertainties on the topography of MEG forward solutions is found especially for quasi-radial sources as they are for example found on top of gyri. Furthermore, a strong influence of gray and white matter conductivity variations on the signal magnitude is found especially for sources on sulcal walls. For MEG source analysis, mean localization errors are below 2 mm in most regions, but significant errors are found in deep and temporal areas with mean localization errors of up to 5 mm. Significant effects on reconstructed source orientation and magnitude are particularly strong when no rank reduction is performed, i.e., the quasi-radial source orientation, which has a comparatively small contribution to the MEG signal, is not excluded. On the other hand, rank reduction makes it impossible to reconstruct the actual source orientation as long as this source is not exactly quasi-tangential. Even though the sensitivity of MEG source analysis towards tissue conductivity uncertainties is clearly weaker than for the EEG, it should not be completely neglected. The effects found for quasi-radial sources have little practical implications, since these sources have a very weak MEG signal and can therefore usually not be properly detected, but the effects found for quasi-tangential sources, such as variations of reconstructed source magnitude and orientation, could have significant effects in practice, e.g., in a combined analysis of EEG and MEG.