Noise Learning in Empirical Bayesian Source Reconstruction Algorithms for Electromagnetic Brain Imaging
share
Electromagnetic brain imaging is the reconstruction of brain activity from electroencephalography (EEG) and magnetoencephalography (MEG) data. Robust estimation of the activity of multiple correlated electromagnetic brain sources has long been a challenging task, one that is significantly compounded by the effect of interference from spontaneous brain activity, sensor noise, and other artifacts. Empirical Bayesian-based algorithms, like Champagne, have been successful in addressing these issues in a principled fashion. Inherent to the success of these algorithms is the assumption that an estimate of the noise and interference can be obtained by a separate “baseline data”. However, in many scenarios, such baseline data is not always available or may be unreliable for noise estimation. Here, we propose robust methods to estimate heteroscedastic sensor noise covariance without the need for additional baseline or pre-stimulus data (Champagne with EM_NL and CB_NL). We demonstrate, both in simulations and with real data, that noise learning can result in robust reconstruction of complex brain source activity without the need for baseline data.
READ FULL TEXT