TU Ilmenau | Universitätsbibliothek | ilmedia, 2019
http://www.tu-ilmenau.de/ilmedia
Zelle, Dennis; Fiedler, Patrique; Haueisen, Jens:
Art
ifact reduction in multichannel ECG recordings acquired with textile
electrodes
Zuerst erschienen
in: Biomedical Engineering = Biomedizinische Technik. -
Berlin [u.a.] :
de Gruyter. - 57 (2012), Suppl. 1, Track-F, p. 171-174.
Erstveröffentlichung
: 2012-08-30
ISSN
(online): 1862-278X
ISSN (print):
0013-5585
DOI:
10.1515/bmt-2012-4401
[
Zuletzt gesehen: 2019-08-12]
„Im Rahmen der hochschulweiten Open-Access-Strategie für die Zweitveröffentlichung identifiziert
durch die Universitätsbibliothek Ilmenau.“
“Within the academic Open Access Strategy identified for deposition by Ilmenau University Library.”
„Dieser Beitrag ist mit Zustimmung des Rechteinhabers aufgrund einer (DFG-
geförderten) Allianz- bzw. Nationallizenz frei zugänglich.“
„This publication is with permission of the rights owner freely accessible due to
an Alliance licence and a national licence (funded by the DFG, German
Research Foundation) respectively.”
Artifact Reduction in Multichannel ECG Recordings Acquired with
Textile Electrodes
D. Zelle, P. Fiedler, J. Haueisen
Institute of Biomedical Engineering and Informatics, Ilmenau University of Technology, Ilmenau, Germany
patrique.fiedler@tu-ilmenau.de
Abstract
Textile electrodes integrated into clothes are an innovative approach for mobile ECG monitoring. However, the lack of
electrode fixation on the skin causes high susceptibility to artifacts due to movements and changing electrochemical
characteristics of the textile electrodes. In this paper we compare different artifact removal approaches concerning their
efficiency in realistic multichannel ECG recordings acquired with textile electrodes. We employed Principal Component
Analysis (PCA) and Independent Component Analysis (ICA) in time and frequency domain using FastICA and
Temporal Decorrelation Source Separation (TDSEP), respectively. Using textile electrodes comprising silver-coated
fibers, five Einthoven-I-leads were acquired during walking, running and extensive breathing. Horizontally aligned
electrodes each located on the left and right side of the shoulders, the chest and the back obtain the ECG signals. A
reference signal was recorded using self-adhesive Ag/AgCl electrodes placed at the inner forearms enabling calculation
of the correlation coefficient and the R-peak detection error. The methods using ICA enhance ECG recordings acquired
with textile electrodes for all test conditions. TDSEP in the time domain obtains the best results and successfully
removes artifacts in recordings of extensive breathing and walking. The results during running show considerable
improvements but no complete artifact separation. In conclusion, ICA represents a promising approach for artifact
reduction in multichannel ECG recordings acquired with textile electrodes.
1 Introduction
The evaluation of physical conditions of competitive
athletes using physiological parameters is an established
standard in sports medicine. Regular sports activity can
increase the personal health, particularly strengthening the
cardiovascular system [1]. Despite these advantages
regular physical activity may be associated with an
increased risk of Sudden Cardiac Death (SCD) [2], and
therefore the heart and the cardiovascular system are of
special interest. Anatomical and physiological disorders
imply an abnormal Electrocardiogram (ECG). Hence, ECG
is an effective tool in identifying risk factors for SCD.
However, recording the ECG with conventional
measurement methods, i.e. using self-adhesive silver/silver
chloride (Ag/AgCl) electrodes is not applicable during
exercise because of the distracting movement and
application time limitations. A novel approach to measure
the electrical heart activity is the use of textile electrodes
which include electrically conductive fibers to record the
ECG from the body surface. Enabling acquisition of
physiological parameters by integration of sensors in
clothing has already been successfully achieved for several
applications. Borges et al. [3] applied textile electrodes for
ECG acquisition in pregnant women. Coosemans et al. [4]
integrated woven stainless steel electrodes in a baby suit
for continuous ECG monitoring for prevention of sudden
infant death syndrome. Paradiso et al. [5] used conductive
and piezoresistive yarns to measure simultaneously the
ECG, the respiratory signal and the movement activity.
The use of electroconductive fibers has a number of
advantages compared to self-adhesive electrodes. The
wiring necessary to connect the electrodes to the
measurement instrumentation is integrated into clothing
and thus the ECG can be transmitted using wireless
communication [3-7] or data logging on a small device [8].
Moreover, textile electrodes are reusable, do not need
additional electrolyte gel and do not cause skin irritations,
thus enabling long-term acquisitions.
However, textile electrodes cannot be attached to the body
skin and thus are susceptible to artifacts due to extensive
breathing or body movements. Furthermore, electrodes
based on electrically conductive fibers are polarizable and
show unstable potentials as well as high impedances. The
amplitude of resulting artifacts may considerably exceed
the ECG signal amplitude thus impeding reliable medical
analysis. A possible approach for extracting the ECG
signal from highly disturbed recordings is the use of
multichannel sensor arrangements in combination with
multivariate statistics. Methods like Principal Component
Analysis (PCA) [9] and Independent component analysis
(ICA) [10-12] are based on multivariate statistics and have
been successfully utilized in several signal processing
applications on cardiological recordings.
In this paper we compare different techniques based on
multivariate statistics concerning their efficiency in
removing artifacts from realistic multichannel ECG
recordings acquired with textile electrodes.
Biomed Tech 2012; 57 (Suppl. 1) © 2012 by Walter de Gruyter · Berlin · Boston. DOI 10.1515/bmt-2012-4401
171
Bereitgestellt von | Technische Universität Ilmenau
Angemeldet
Heruntergeladen am | 12.08.19 14:44
2 Methods
2.1 Electrodes and Sensor Arrangement
Multichannel ECG recordings were acquired with textile
electrodes comprising silver-coated synthetic fibers
integrated in nonconductive fabrics. The size of each
textile electrode is approximately 1.615 cm. The
material is washable and allows multiple uses. Figure 1
shows an example of the textile electrodes used in this
study.
Figure 1 Textile electrodes consisting of electroconductive
fibers integrated in knitted fabrics.
In order to obtain five Einthoven-I-leads ten horizontally
aligned electrodes were located on the left and right side
of the shoulders, the chest and the back. A reference
signal was recorded using self-adhesive Ag/AgCl
electrodes placed at the inner forearms. Figure 2
illustrates this sensor arrangement. The patient ground
was placed at the neck of the subject.
Figure 2 Recording sites for multichannel ECG
acquisition using textile electrodes (grey) and self-
adhesive Ag/AgCl electrodes (white): a) frontal view, and
b) rearward view.
2.2 Implementation
Applying multivariate statistics to the acquired
multichannel ECG enables converting the input data into
a more meaningful representation yielding a matrix ,
which contains components that may be associated with
the ECG or the artifact signal, respectively. Neglecting the
artifact components in before reconstructing the ECG
signal results in an artifact-free ECG signal. Figure 3
schematically illustrates the use of multivariate statistics in
artifact reduction for ECG recordings.
PCA and ICA were implemented in Matlab (Mathworks,
Natick, USA) in order to determine . ICA was solved in
both time and frequency domain using the FastICA
algorithm [13] and the Temporal Decorrelation Source
Separation (TDSEP) algorithm [14]. The number of
iterations for the FastICA algorithm was limited to 200.
TDSEP was executed with ten correlation matrices time-
delayed by 2 ms each. The transformation from the time to
the frequency domain was performed by applying a Short-
Time Fourier Transform (STFT) with an FFT-length of 60
samples and a Hanning window of 117 ms and 90%
overlap.
Figure 3 Procedure for artifact reduction in multichannel
ECG recordings by applying multivariate statistics.
2.3 Identification of Artifact Components
After the input dataset has been transformed into a new
representation, the artifact components need to be
identified. Exploiting the morphological structure of the
ECG the kurtosis as fourth-order moment of a
probability density function (PDF) yields an objective
classification parameter. Signals with a Gaussian
distribution have (e.g. white noise). Castells et al.
[15] showed that PDFs of ECG signals can be assumed to
be super-Gaussian with high kurtosis values in the range of
, considerably exceeding the values of
noise and artifacts. Hence, components with high kurtosis
values are assumed to correspond to the ECG signal. Thus,
neglecting all other components removes both noise and
artifacts and yields to an artifact-free ECG reconstruction.
2.4 Experimental Setup
Multichannel ECG recordings were acquired from a
healthy male subject during extensive breathing, walking
and running using the sensor arrangement shown in
Figure 2. The data was recorded using shielded wires
connected to a multichannel amplifier (RefaExt, Advanced
Neuro Technology B.V., Enschede, Netherlands) with a
sampling rate of 512 Hz. The acquired ECG recordings
were preprocessed in Matlab using a wavelet
decomposition algorithm to remove baseline drift and
wavelet denoising to reduce power-line interference and
high-frequency noise. The artifact reduction procedure was
applied to sections of the ECG recordings with 60 seconds
length for each test condition and for interval lengths of
two, five and ten seconds. In order to evaluate the
performance of the individual techniques the sample
correlation coefficient and the beat detection error were
computed with respect to the ECG reference recording.
Beat detection was performed using the Pan-Tompkins
algorithm [16].
(a)
(b)
𝑿
Dataset
Transformation
𝒀
Component
classification
𝒀
Discard artifact
components
𝑿
Signal
reconstruction
Biomed Tech 2012; 57 (Suppl. 1) © 2012 by Walter de Gruyter · Berlin · Boston. DOI 10.1515/bmt-2012-4401
172
Bereitgestellt von | Technische Universität Ilmenau
Angemeldet
Heruntergeladen am | 12.08.19 14:44
3 Results
Figure 4 shows the overall results for the correlation
coefficient as boxplots according to the different test
conditions. Additionally, the black cross indicates the
mean value of the evaluated parameter. The methods using
ICA enhance recordings acquired with textile electrodes
for all test conditions with respect to the raw ECG signal.
The PCA achieves an increase in the correlation to the
reference signal for extensive breathing and walking but
cannot enhance the raw ECG acquired during running. The
results for the beat detection error closely correspond to
the results for the correlation coefficient .
Figure 4 Correlation coefficient in the time domain (TD)
and frequency domain (FD) according to the test
conditions extensive breathing, walking, and running.
TDSEP in the time domain produces the best results and
successfully removes artifacts in recordings of extensive
breathing and walking. After applying TDSEP the mean
value for the correlation coefficient increases to
0.85 (raw ECG: 0.39) and 0.79 (0.45), respectively. The
corresponding beat detection errors decrease to 0.10 (0.38)
and 0.06 (0.30). The results during running show
considerable improvements but no complete artifact
separation. However, the average correlation increases to
0.48 (0.20) and the beat detection error decreases to 0.62
(1.17).
Figure 5 shows an example of successful artifact reduction
achieved with TDSEP in the time domain for an ECG
recording acquired with textile electrodes from the chest.
The raw recording after signal conditioning is highly
contaminated with motion artifacts considerably exceeding
the ECG signal amplitude (Figure 5a). Figure 5b depicts
the ICA result using the TDSEP algorithm in the time
domain. Comparison with the reference signal acquired
with self-adhesive Ag/AgCl electrodes in Figure 5c shows
the successful artifact reduction as a result of the TDSEP
algorithm.
Figure 5 Artifact reduction with TDSEP for an ECG
recording from the chest acquired during walking: a)
artifact contaminated raw signal, b) signal after TDSEP
application, and c) reference signal.
The different test conditions considerably influence the
quality of the results. Increasing movement leads to
decreasing correlation coefficients and increasing beat
detection errors. Furthermore, the results show high
dependence on the recording sites implying that the body
movements may have a different impact on the signal
acquisition at the individual electrode positions. Moreover,
the number of channels included in the artifact reduction
influences the results for all employed methods.
Decreasing the channel number leads to decreasing
performance of the artifact removal procedure. In contrast,
the interval length does not show any influence on the
results indicating that a signal length of two seconds is
sufficient for artifact reduction with the investigated
methods.
Raw
FastICA (FD)
TDSEP (FD)
FastICA (TD)
TDSEP (TD)
PCA
(a) Extensive breathing
(b) Walking
(c) Running
Raw
FastICA (FD)
TDSEP (FD)
FastICA (TD)
TDSEP (TD)
PCA
Raw
FastICA (FD)
TDSEP (FD)
FastICA (TD)
TDSEP (TD)
PCA
Biomed Tech 2012; 57 (Suppl. 1) © 2012 by Walter de Gruyter · Berlin · Boston. DOI 10.1515/bmt-2012-4401
173
Bereitgestellt von | Technische Universität Ilmenau
Angemeldet
Heruntergeladen am | 12.08.19 14:44
4 Conclusion
Multivariate statistics combined with the temporal
structure of the ECG signal, as utilized in TDSEP, allow
extracting a reliable ECG from multichannel recordings
acquired with textile electrodes even during extensive
breathing or walking. For extensive physical activity it will
be necessary to further optimize electrode positions and
algorithm parameters in order to enable continuous mobile
ECG monitoring with electroconductive textiles.
5 References
[1] Bernadet P: Benefits of Physical Activity in the
Prevention of Cardiovascular Diseases. Journal
of Cardiovascular Pharmacology, vol. 25, 1995
[2] Corrado D, Migliore F, Bevilacqua M, Basso C,
Thiene G: Sudden cardiac death in athletes.
Herz, vol. 34, 2009.
[3] Borges LM, Rente A, Velez FJ, Salvado LR,
Lebres AS, Oliveira JM, Araujo P, Ferro J:
Overview of progress in Smart-Clothing project
for health monitoring and sport applications.
First International Symposium on Applied
Sciences on Biomedical and Communication
Technologies, 2008
[4] Coosemans J, Hermans B, Puers R: Integrating
wireless ECG monitoring in textiles. The 13th
International Conference on Solid-State Sensors,
Actuators and Microsystems (TRANSDUCERS
'05), 2005
[5] Paradiso R, Loriga G, Taccini N: A Wearable
Health Care System Based on Knitted Integrated
Sensors. IEEE Transactions on Information
Technology in Biomedicine, vol. 9, 2005
[6] Galeottei L, Paoletti M, Marchesi C:
Development of a low cost wearable prototype
for long-term vital signs monitoring based on
embedded integrated wireless module.
Computers in Cardiology, 2008
[7] Ottenbacher J, Romer S, Kunze C,
Grossmann U, Stork W: Integration of a
Bluetooth based ECG system into clothing.
Eighth International Symposium on Wearable
Computers (ISWC '04), 2004
[8] Pola T, Vanhala J: Textile Electrodes in ECG
Measurement. 3rd International Conference on
Intelligent Sensors, Sensor Networks and
Information (ISSNIP '07), 2007
[9] Castells F, Laguna P, Sörnmo L, Bollmann A,
Millet J: Principal Component Analysis in ECG
Signal Processing. EURASIP Journal on
Advances in Signal Processing, 2007
[10] Castells F, Cebrián A, Millet J: The role of
independent component analysis in the signal
processing of ECG recordings. Biomedizinische
Technik, vol. 52, 2007
[11] Milanesi M, Martini N, Vanello N, Positano V,
Santarelli MF, Paradiso R, De Rossi D, Landini
L: Multichannel Techniques for Motion Artifacts
Removal from Electrocardiographic Signals.
28th Annual International Conference of the
IEEE Engineering in Medicine and Biology
Society (EMBS '06), 2006
[12] DiPietroPaolo D, Müller HP, Tedeschi W, Park
JW, Jung F, Erné SN: Noise reduction in CHD
patients by means of BSS. International
Congress Series, vol. 1300, 2007
[13] Hyvarinen A: Fast and robust fixed-point
algorithms for independent component analysis.
IEEE Transactions on Neural Networks, vol. 10,
1999
[14] Ziehe A, Müller KR: TDSEP - an efficient
algorithm for blind separation using time
structure. Proceedings of the 8th International
Conference on Artificial Neural Networks
(ICANN'98), 1998
[15] Castells F, Rieta JJ, Millet J, Zarzoso V:
Spatiotemporal blind source separation approach
to atrial activity estimation in atrial
tachyarrhythmias. IEEE Transactions on
Biomedical Engineering, vol. 52, 2005
[16] Pan J, Tompkins WJ: A Real-Time QRS
Detection Algorithm. IEEE Transactions on
Biomedical Engineering, 1985
Biomed Tech 2012; 57 (Suppl. 1) © 2012 by Walter de Gruyter · Berlin · Boston. DOI 10.1515/bmt-2012-4401
174
Bereitgestellt von | Technische Universität Ilmenau
Angemeldet
Heruntergeladen am | 12.08.19 14:44
