Automated detection of EEG artifacts during sleep: preprocessing for all-night spectral analysis

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Automatic processing of large electroencephalogram (EEG) datasets for spectral analysis is challenging due to the impossibility of visual inspection for each epoch.
• Artifacts in EEG data can significantly distort spectral analysis results, necessitating robust detection methods.

Purpose of the Study:

• To develop and validate a simple artifact detection algorithm for automated, large-scale EEG data processing.
• To enable reliable off-line spectral analysis of extensive EEG recordings.

Main Methods:

• A chi-square (chi(2)) goodness-of-fit test to a Gaussian distribution (CSQ) was employed to assess EEG epochs (30 seconds long).
• The CSQ test quantifies non-stationarities in EEG amplitude distributions, generating a chi(2) coefficient.
• Epochs with chi(2) coefficients exceeding a heuristic threshold were identified as containing artifacts and discarded.

Main Results:

• The CSQ test demonstrated high sensitivity to non-stationarities indicative of artifacts in EEG epochs.
• A large chi(2) coefficient reliably signaled the presence of artifacts.
• This technique facilitated efficient data reduction and reliable automatic off-line spectral analysis of 50 nights of sleep EEG.

Conclusions:

• The chi-square goodness-of-fit test provides an effective and efficient method for automatic artifact detection in large EEG datasets.
• This algorithm supports reliable spectral analysis of EEG data by ensuring the exclusion of artifact-contaminated segments.
• The developed technique is suitable for processing extensive sleep EEG recordings automatically.