An ensemble de-noising method for spatio-temporal EEG and MEG data

EEG/MEG are important tools for non-invasive medical diagnosis and basic studies of the brain and its functioning, but often applications are limited due to a very low SNR in the data. Here, we present a discrete wavelet transform (DWT) based de-noising method for spatio-temporal EEG/MEG measurements collected by a sensor array. A robust threshold selection can be achieved by incorporating spatial information and pre-stimulus data to estimate signal and noise energies. Further improvement can be gained by applying a translation-invariant approach to the derived de-noising scheme. In simulations, the performance of the proposed method is evaluated in comparison to standard de-noising and low-rank approximation, which o ers some complementarity to our approach.