Cleaning fetal MEG using a beamformer search for the optimal forward model.

Abstract

We present a new method for improving the signal-to-noise ratio (SNR) of event-related fetal MEG signals based upon the SAM minimum-variance beamformer. SAM could separate the evoked response source activity from the remaining fMEG signal and interference if the evoked response source coordinates and forward model were know. However, this requires knowledge of both the coordinate of the evoked response source and its forward model. In late gestation, the vernix caseosa effectively insulates the fetus from the amniotic fluid. Hence, the forward model could be approximated by an equivalent current dipole in a homogeneously conducting sphere with its origin at the center of the fetal head. In the absence of accurate anatomical data, a beamformer could be used to evaluate all feasible source-origin combinations--selecting the combination giving the best SNR for the evoked response. Application of this approach to measured fMEG data reveals that the optimal model sphere location is described not by a single local sphere origin, but rather by all origins lying within an extended region. This result is explained by model predictions showing the same region of ambiguity [Vrba, 2004]. Although the model search does not localize the sources of the fetal evoked response, it does significantly improve SNR. This was demonstrated by analysis of fetal auditory evoked response data.