An automated algorithm to identify and reject artefacts for quantitative EEG analysis during sleep in patients with sleep-disordered breathing.

Purpose: Large quantities of neurophysiological electroencephalogram (EEG) data are routinely collected in the sleep laboratory. These are underutilised due to the burden of managing artefact contamination. The aim of this study was to develop a new tool for automated artefact rejection that facilitates subsequent quantitative analysis of sleep EEG data collected during routine overnight polysomnography (PSG) in subjects with and without sleep-disordered breathing (SDB).

Methods: We evaluated the accuracy of an automated algorithm to detect sleep EEG artefacts against artefacts manually scored by three experienced technologists (reference standard) in 40 PSGs. Spectral power was computed using artefact-free EEG data derived from (1) the reference standard, (2) the algorithm and (3) raw EEG without any prior artefact rejection.

Results: The algorithm showed a high level of accuracy of 94.3, 94.7 and 95.8% for detecting artefacts during the entire PSG, NREM sleep and REM sleep, respectively. There was good to moderate sensitivity and excellent specificity of the algorithm detection capabilities during sleep. The EEG spectral power for the reference standard and algorithm was significantly lower than that of the raw, unprocessed EEG signal.

Conclusions: These preliminary findings support an automated way to process EEG artefacts during sleep, providing the opportunity to investigate EEG-based markers of neurobehavioural impairment in sleep disorders in future studies.