Spatiotemporal noise covariance model for MEG/EEG data source analysis
Sergey Plis, John St George, Sung Chan Jun, J. Paré-Blagoev, Douglas M. Ranken, David Maria Schmidt, C. C. Wood · arXiv (Cornell University) · 2005
A new method for approximating spatiotemporal noise covariance for use in MEG/EEG source analysis is proposed. Our proposed approach extends a parameterized one pair approximation consisting of a Kronecker product of a temporal covariance and a spatial covariance into 1) an unparameterized one pair approximation and then 2) into a multi-pair approximation. These models are motivated by the need to better describe correlated background and make estimation of these models more efficient. The effects of these different noise covariance models are compared using a multi-dipole inverse algorithm and simulated data consisting of empirical MEG background data as noise and simulated dipole sources.