An Adaptive System for Improved Identification and Removal of Noise from Single Trial EEG/MEG via Model Order Estimation in ICA

An adaptive model order estimation method for Independent Component Analysis (ICA) in EEG/MEG data is presented. This technique seeks to extract the minimum number of components necessary for effective Blind Source Separation (BSS). Experimental results using synthesized noisy MEG data demonstrate the utility of this technique. Model order estimation is used in the extraction of baseline noise components which will serve as templates for subsequent identification and removal of noise. These templates are used to remove noise from a data set containing a somatosensory evoked response (SSR) potential; model order estimation was also used to decompose the SSR data set.