Maximum likelihood spatiotemporal EEG/MEG source analysis

EEG/MEG noise has an unequal variance and is correlated, both in space and in time. Noise variance may differ greatly between samples or sensors, and correlations between samples or sensors can be very high [1-4]. If these noise characteristics are neglected, then an EEG/MEG source analysis will yield unreliable results [e.g. 5, 6]. First, source parameter estimates will be inefficient. That is, their standard errors will be too high. Second, the estimated covariance matrix of the parameter estimates will be inaccurate. In general it will give a too optimistic impression of precision. Third, goodness of fit measures will be unreliable, which may result in overor undermodeling of the data. For these reasons, it is very beneficial to incorporate the spatiotemporal noise covariance in the analysis. Although the spatial covariance is incorporated quite often [6-11], the temporal covariance is disregarded up to now. Therefore, we are developing a method to incorporate the spatiotemporal noise covariance matrix. The essential feature of this method is that the estimation problem is split into two parts. First a model is fitted to the spatiotemporal noise covariance matrix. Then the source parameters are estimated given this noise model.