Automatic artifacts and arousals detection in whole-night sleep EEG recordings

In sleep electroencephalographic (EEG) signals, artifacts and arousals marking are usually part of the processing. This visual inspection by a human expert has two main drawbacks: it is very time consuming and subjective.
To detect artifacts and arousals in a reliable, systematic and reproducible automatic way, we developed an automatic detection based on time and frequency analysis with adapted thresholds derived from data themselves.
The automatic detection performance is assessed using 5 statistic parameters, on 60 whole night sleep recordings coming from 35 healthy volunteers (male and female) aged between 19 and 26. The proposed approach proves its robustness against inter- and intra-, subjects and raters’ scorings, variability. The agreement with human raters is rated overall from substantial to excellent and provides a significantly more reliable method than between human raters.
Existing methods detect only specific artifacts or only arousals, and/or these methods are validated on short episodes of sleep recordings, making it difficult to compare with our whole night results.
The method works on a whole night recording and is fully automatic, reproducible, and reliable. Furthermore the implementation of the method will be made available online as open source code.