Muscle artifacts in the sleep EEG: Automated detection and effect on all‐night EEG power spectra

Abstract

Owing to the use of scalp electrodes in human sleep recordings, cortical EEG signals are inevitably intermingled with the electrical activity of the muscle tissue on the skull. Muscle artifacts are characterized by surges in high frequency activity and are readily identified because of their outlying high values relative to the local background activity. To detect bursts of myogenic activity a simple algorithm is introduced that compares high frequency activity (26.25-32.0 Hz) in each 4-s epoch with the activity level in a local 3-min window. A 4-s value was considered artifactual if it exceeded the local background activity by a certain factor. Sensitivity and specificity of the artifact detection algorithm were empirically adjusted by applying different factors as artifact thresholds. In an analysis of sleep EEG signals recorded from 25 healthy young adults 2.3% (SEM: 0.16) of all 4-s epochs during sleep were identified as artifacts when a threshold factor of four was applied. Contamination of the EEG by muscle activity was more frequent towards the end of non-REM sleep episodes when EEG slow wave activity declined. Within and across REM sleep episodes muscle artifacts were evenly distributed. When the EEG signal was cleared of muscle artifacts, the all-night EEG power spectrum showed significant reductions in power density for all frequencies from 0.25-32.0 Hz. Between 15 and 32 Hz, muscle artifacts made up a substantial part (20-70%) of all-night EEG power density. It is concluded that elimination of short-lasting muscle artifacts reduces the confound between cortical and myogenic activity and is important in interpreting quantitative EEG data. Quantitative approaches in defining and detecting transient events in the EEG signal may help to determine which EEG phenomena constitute clinically significant arousals.

The Combined Technique for Detection of Artifacts in Clinical Electroencephalograms of Sleeping Newborns

“…However in our experiments this information has not been used for learning the new rule. So the first practical result of our research is that the discovered recognition rule is comprehensible for EEG-experts and that this rule matches well with the rule described in [12]. The second practical issue is that we compared the performance of commonly used machine learning methods on sleep EEGs without using the additional information coming from channels besides EEG.…”

“…Table II we can conclude that our technique certainly outperforms the C4.5 and neural-network techniques on the sleep EEGs. However analyzing the induced DTs described above, we can see that both exploit only one feature AbsPowBeta2 presenting the power of a high frequency band mentioned in [12]. That is, these DTs testing one feature AbsPowBeta2 band without the information about background neural activity of sleeping newborns cannot perform enough well.…”

“…An analogous feature is tested for recognizing muscle artifacts in sleep EEG by using the threshold technique [12]. Additionally this technique evaluates the background neural activity at 3-minute window adjusted for adult EEGs.…”

“…In this research we use fruitful machine learning and pattern recognition methods to induce new rules for an automated recognition of EEG [5], [12]), for inducing these rules we do not use channels besides EEG. However, the experts labeling the EEG segments used additional information coming from other channels which were useful for visual recognition of the EEG artifacts.…”

“…Within our research we also evaluate how well the discovered rules can recognize the artifacts in the EEGs recorded from newborns of which background neural activity varies during sleep hours. Such variations heavily affect the accuracy of artifact recognition [12]. In this research, however, we do not use the average estimations of background neural activity as suggested in [12] and attempt learning a recognition rule exploiting the spectral features calculated for the current EEG segment.…”