EEG Preprocessing and Denoising

Abstract

In this chapter, we first introduce physiological and non-physiological artifacts embedded in the raw EEG signals, e.g., ocular related artifacts (physiological) and power line interference (non-physiological). Then, we introduce the montage to describe and apply the location of scalp electrodes in the context of EEG studies. Further, we describe several preprocessing steps that are commonly used in the EEG preprocessing, including filtering, re-referencing, segmenting, removal of bad channels and trials, as well as decomposition of EEG using independent component analysis. More specifically, appropriate band-pass filtering can effectively reduce superimposed artifacts from various sources which are embedded in the EEG recordings. Re-referencing is a linear transformation of the EEG data, through which noise in the reference electrodes could turn into noise in the scalp electrodes. Data epochs that are time-locked to the specific events of interest should be extracted to facilitate the investigation of task/stimulus-related changes in EEG. Trials contaminated by artifacts, as well as bad channels that are not functioning properly for various reasons, should be excluded from further analysis. Given that the EEG data recorded from scalp electrodes can be considered as summations of neural activities, and that artifacts are independent with each other, independent component analysis could be a powerful and efficient strategy to separate artifact from EEG signals.