Multi-trial evoked EEG and independent component analysis.

Area of Science:

• Neuroscience
• Signal Processing

Background:

• Independent Component Analysis (ICA) is widely used for decomposing electroencephalography (EEG) data.
• ICA relies on the assumption of statistical independence among hidden components.
• This assumption is often violated in evoked, multi-trial, non-stationary EEG data, questioning ICA's applicability.

Purpose of the Study:

• To address the limitations of ICA in analyzing evoked EEG data.
• To introduce a preprocessing method that ensures ICA separation works effectively.
• To improve the identification of hidden components in evoked EEG and MEG.

Main Methods:

• A novel preprocessing technique is introduced for multi-trial EEG data.
• This method imparts a 'null conditional mean' (NCM) property to the hidden components.
• The NCM property is demonstrated to be sufficient for successful ICA separation.

Main Results:

• Theoretical validation of the mean subtraction method (NCM) was established.
• Numerical simulations mimicking transcranial magnetic stimulation (TMS)-evoked EEG data confirmed the method's efficiency.
• The NCM approach effectively suppresses stimulus-evoked artifacts, including muscular artifacts in TMS-EEG.

Conclusions:

• The NCM preprocessing significantly improves ICA results compared to conventional methods for multi-trial data.
• This methodology enhances the discovery of previously undetectable components in evoked EEG and MEG.
• The proposed approach offers a more robust application of ICA to complex neurophysiological data.