Effects of signal artefacts on electroencephalography spectral power during sleep: quantifying the effectiveness of automated artefact‐rejection algorithms
Muscle and ocular artefacts corrupt electroencephalography (EEG) signals recorded during sleep and alter their spectral power, potentially affecting the interpretation of such signals and their linkage to psychological disorders (Buysse et al., 2001; Cohen et al., 2013; Woodward et al., 2000). Hence, to process increasingly large streams of EEG recordings, some research groups use well‐established, automated artefact‐rejection algorithms (Brunner et al., 1996; Cohen et al., 2013; Doman et al., 1995), which provide for objective and consistent screening. However, the use of such algorithms has not been adopted universally, because the extent to which muscle and ocular artefacts affect spectral power values has not been quantified explicitly for whole‐night rapid eye movement (REM) and non‐REM (NREM) sleep. In addition, the extent to which these algorithms help match spectral power to their artefact‐free levels has not been assessed fully. Providing a quantitative characterization of the effects of artefacts on spectral power and the effectiveness of artefact‐rejection algorithms will help to increase awareness for the need to properly post‐process EEG signals before they are analysed.
The objective of this report is twofold: (1) to quantify the effects of muscle and ocular artefacts on the average EEG spectral power across different sleep states and EEG channels and (2) to assess the effectiveness of two previously developed automated algorithms—one for muscle artefacts (Brunner et al., 1996) and the other adapted for rejecting potential ocular artefacts (Cohen et al., 2013; Doman et al., 1995)—in minimizing the differences in the average spectral power of whole‐night EEG recordings with respect to their artefact‐free levels (annotated by visual detection and rejection).