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

This paper describes a simple artifact detection algorithm which can be used when large amounts of EEG data are to be automatically processed via spectral analysis techniques in a general purpose digital computer, and visual inspection of each EEG epoch becomes an impossible task. The technique is based on a chi-square (chi(2)) goodness-of-fit test to a Gaussian distribution (CSQ), and it was applied to EEG epochs each 30 sec long. This test proved to be very sensitive to non-stationarities in the EEG amplitude distribution for a particular epoch, and it produced a large value for the chi(2) coefficient when an artifact was present. EEG epochs that gave rise to chi(2) coefficients of value larger than a heuristically determined minimum were discarded from further analysis. The above technique enabled efficient data reduction and reliable automatic off-line processing of 50 nights of sleep EEG via spectral techniques.