QRS artifact elimination on full night sleep EEG

https://doi.org/10.1016/j.medengphy.2005.04.017Get rights and content

Abstract

Spectral analysis is now a standard procedure for analyzing the electroencephalograms (EEG) obtained by polysomnographic recordings. These numerical methods assume an artifact-free EEG since artifacts create spurious spectral components.

Our aim was the development of a QRS artifact removal technique that might be applied to full night EEG with a minimal human intervention. This technique should handle one EEG channel, with or without use of one ECG channel.

Variance minimization, independent component analysis (ICA), morphological filters (MF) have been implemented. Careful attention has been given to define the MF structuring element. The tests on artifact-simulated and real data were checked on the residual ECG spectral components present in the cleaned EEG.

The best results are obtained by the MF when the structuring element is an artifact template defined either directly on the EEG or on the ICA ECG component. Further developments are required to identify and subtract the T-wave artifacts.

Introduction

Polysomnographic recordings are the most common neurophysiological tools to analyze sleep. Electroencephalograms (EEG), electrooculograms (EOG), submental electromyograms (EMG) give important informations about the sleep organization and are routinely used to define the sleep stages [1] EEG are the basic signals in sleep research. Besides the sleep stages scoring, many sophisticated quantitative tools were developed to extract additional enlightenments from the EEG. Among these tools, spectral analysis is now a standard procedure for analyzing EEG. This technique allows for the separation of the EEG into several components at different frequencies, e.g. [2], [3], and is an usual routine to reduce the amount of data to be analyzed. All these numerical methods assume that EEG are artifact free. Artifacts introduce spurious components into the calculations and may therefore lead to incorrect interpretations of the results [4].

Among various artifact types, the cardiac activity may interfere with the EEG. Indeed, the electrical field of the heart propagates throughout the body and modifies the surface potential on the scalp. Dirlich et al. [5] showed the three-dimensional nature of these scalp potentials that are more important during the QRS and the T-waves of the electrocardiogram (ECG). Moreover, this group also demonstrated the scalp potentials dependence on the head orientation relative to the heart. Another EEG interference are the pulse artifacts. They occur with the recurrent motion of electrodes placed over a pulsating scalp artery, which alters the contact with the skin. This creates rhythmic waves synchronous with the heart. Consequently, these contaminations introduce an ECG artifact into the EEG, with high frequency components (QRS waves) and low frequency components (wave repetition at the cardiac frequency). Finally, all the EEG spectrum is distorted.

In this paper, our aim was the development of a QRS artifact removal technique that might be applied to full night sleep EEG with a minimal human intervention. This technique should handle one EEG channel of a recording, with or even without use of one ECG channel.

Section snippets

Subjects

All Subjects were recorded at the Sleep Laboratory of the ULB-Erasme Hospital. They were all part of a previous study [6]. For this investigation, three artifact-free records were arbitrarily selected and four artifacted EEG records were used. Subjects presenting a ECG artifact were not included into the above-mentioned results [6].

Details about the recording procedure may be found elsewhere [6]. Polysomnography was recorded with a 19-channel digital polygraph (Brainnet™, MEDATEC, Brussels,

Results

The procedures were first tested on simulated data for three subjects, then on real data for four subjects. Table 1 resumes some informations about these recordings. All 5-s records were used to perform QRS artifact elimination, even if the EEG presented other artifacts (body movements, low frequency drift, etc.). All EDF records were included for the calculation of the two INPS values and, with the simulated data, for the descriptive statistic of the difference EEG.

Calculations were performed

Discussion

We developed a ECG artifact removal technique that is applied to full night sleep EEG with a minimal human intervention. This technique was designed to handle recordings with one EEG channel at a time, with or even without one ECG channel.

The best artifact is the one that is not recorded. However, this ideal situation may not be encountered in practice. In the presence of an artifact, the user is faced with three options: to drop the recording, to use the recording despite the artifact

Conclusions

We developed an user-friendly technique to subtract the QRS artifact on full night sleep EEG in man. Variance minimization, independent component analysis and morphological filters have been implemented. Careful attention has been given to the definition of the MF structuring element. Methods performances have been checked on the residual ECG spectral components present into the cleaned EEG.

Results demonstrate that the ECG channel is mandatory to define the QRS positions. The best results are

Acknowledgments

This work was supported by the Erasme Hospital. We thank the National Fund for Scientific Research (Belgium) for its constant support. We also thank Bernard Jacques for technical assistance and the members of the sleep laboratory for their respective contribution. MD is a Research Associate of the National Fund for Scientific Research (Belgium).

Cited by (20)

View all citing articles on Scopus
View full text
Copyright © 2005 IPEM. Published by Elsevier Ltd. All rights reserved.