Motion Artifact Processing Techniques for Physiological Signals

265 Pages·2013·41.76 MB·English

by  Kevin Sweeney

NATIONAL UNIVERSITY OF IRELAND MAYNOOTH
Motion Artifact Processing Techniques
for Physiological Signals
by
Kevin Sweeney
A thesis submitted in partial fulfillment for the
degree of Doctor of Philosophy
in the
Faculty of Science and Engineering
Department of Electronic Engineering
April 2013
Declaration of Authorship
I, Kevin Sweeney, declare that this thesis titled, ‘Motion Artifact Processing Techniques
for Physiological Signals’ and the work presented in it are my own. I confirm that:
(cid:4) This work was done wholly or mainly while in candidature for a research degree
at this University.
(cid:4) Where any part of this thesis has previously been submitted for a degree or any
other qualification at this University or any other institution, this has been clearly
stated.
(cid:4) Where I have consulted the published work of others, this is always clearly at-
tributed.
(cid:4) Where I have quoted from the work of others, the source is always given. With
the exception of such quotations, this thesis is entirely my own work.
(cid:4) I have acknowledged all main sources of help.
(cid:4) Where the thesis is based on work done by myself jointly with others, I have made
clear exactly what was done by others and what I have contributed myself.
Signed:
Date:
i
“A journey of a thousand miles begins with a single step.”
Lao Tzu
Abstract
The combination of reducing birth rate and increasing life expectancy continues to drive
thedemographicshifttowardanageingpopulationandthisisplacinganever-increasing
burden on our healthcare systems. The urgent need to address this so called health-
care “time bomb” has led to a rapid growth in research into ubiquitous, pervasive and
distributed healthcare technologies where recent advances in signal acquisition, data
storage and communication are helping such systems become a reality. However, similar
to recordings performed in the hospital environment, artifacts continue to be a major
issue for these systems. The magnitude and frequency of artifacts can vary significantly
depending on the recording environment with one of the major contributions due to
the motion of the subject or the recording transducer. As such, this thesis addresses
the challenges of the removal of this motion artifact removal from various physiological
signals.
The preliminary investigations focus on artifact identification and the tagging of physi-
ological signals streams with measures of signal quality. A new method for quantifying
signal quality is developed based on the use of inexpensive accelerometers which fa-
cilitates the appropriate use of artifact processing methods as needed. These artifact
processing methods are thoroughly examined as part of a comprehensive review of the
most commonly applicable methods. This review forms the basis for the comparative
studies subsequently presented. Then, a simple but novel experimental methodology
for the comparison of artifact processing techniques is proposed, designed and tested
for algorithm evaluation. The method is demonstrated to be highly effective for the
type of artifact challenges common in a connected health setting, particularly those con-
cerned with brain activity monitoring. This research primarily focuses on applying the
techniquestofunctionalnearinfraredspectroscopy(fNIRS)andelectroencephalography
(EEG) data due to their high susceptibility to contamination by subject motion related
artifact.
Using the novel experimental methodology, complemented with simulated data, a com-
prehensive comparison of a range of artifact processing methods is conducted, allowing
the identification of the set of the best performing methods. A novel artifact removal
techniqueisalsodeveloped,namelyensembleempiricalmodedecompositionwithcanon-
ical correlation analysis (EEMD-CCA), which provides the best results when applied on
fNIRS data under particular conditions. Four of the best performing techniques were
then tested on real ambulatory EEG data contaminated with movement artifacts com-
parable to those observed during in-home monitoring.
It was determined that when analysing EEG data, the Wiener filter is consistently
the best performing artifact removal technique. However, when employing the fNIRS
data, the best technique depends on a number of factors including: 1) the availability
of a reference signal and 2) whether or not the form of the artifact is known. It is
envisagedthattheuseofphysiologicalsignalmonitoringforpatienthealthcarewillgrow
significantly over the next number of decades and it is hoped that this thesis will aid in
the progression and development of artifact removal techniques capable of supporting
this growth.
Acknowledgements
Firstandforemost, IwouldliketoexpressmysincerestthankstomytwosupervisorsDr
Tom´as Ward and Dr Se´an McLoone for their constant direction and support throughout
my PhD. Their approachability, enthusiasm and in-depth knowledge in the area have
made the overall experience a very enjoyable one.
I would also like to express my gratitude to all the members of staff in the Electronic
Engineering Department in Maynooth for generating a atmosphere in which work, and
fun, came easily. In particular I would like to thank John Maloco and Denis Buckley for
their technical help throughout the research and Joanne Bredin and Ann Dempsey for
always being at the end of the phone/email when needed. Thanks to all the postgrads
in the department for the company throughout the four years, particularly Anthea,
Aodhan, Darren, Francesco, Giogio, Niall, Shane B and Violeta. Particular thanks are
due to Dr Hasan Ayaz, Dr Meltem Izzetoglu and Prof. Banu Onaral, and all in Drexel
university who were kind enough to share both their knowledge and their offices with
me. This research would not have started without the support from the Irish Research
Council for Science, Engineering and Technology, so again I give my thanks.
I would like to thank my parents, Kevin and Cecelia, who have been unwavering in their
love, not only during this PhD, but throughout my life. I could not have completed this
work without your help. To my sisters, Ciara and Siobh´an, your support has always
meant the world to me. I thank you for looking after your “student brother” more than
you might have had to. Thanks to all of the great friends I have had throughout my
time in Maynooth both during undergrad and postgrad. To Graham, Leanne, Shelley
and Sinead, thank you for being great housemates (even after I left). To Annie, Kate
and all the girls in Biology, thanks for the constant cups of tea and chats! A special
thanks to Ian Hatch and Mark Creevey who have been some of my closest friends for
the last 24 years! Long may it continue.
Finally I would like to thank Ciar´an Pollard, Damian Kelly, Helen O’Dowd, Lorcan
Walsh and Shane Lynn for being the best housemates and friends a person could ask
for. Each of you have made my time here some of the best in my life; thank you.
v
Contents
Declaration of Authorship i
Abstract iii
Acknowledgements v
List of Figures x
List of Tables xvii
Abbreviations xviii
1 Introduction 1
1.1 Objective . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Contributions of the Thesis . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.3 List of Publications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.4 Outline of the Thesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2 Physiological Background 9
2.1 Circulatory System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.1.1 Photoplethysmography (PPG) . . . . . . . . . . . . . . . . . . . . 11
2.1.2 Electrocardiography (ECG) . . . . . . . . . . . . . . . . . . . . . . 14
2.2 The Nervous System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.2.1 Electroencephalography (EEG) . . . . . . . . . . . . . . . . . . . . 16
2.2.2 Functional near-infrared spectroscopy (fNIRS) . . . . . . . . . . . 19
2.2.3 Electromyography (EMG) . . . . . . . . . . . . . . . . . . . . . . . 20
2.3 Artifact Modalities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.3.1 Environmental Artifacts . . . . . . . . . . . . . . . . . . . . . . . . 22
2.3.2 Experimental Error . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.3.3 Physiological Artifacts . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3 Review of Existing Artifact Removal Methods 27
3.1 Adaptive Filtering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.2 Wiener Filtering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
vi
Contents
3.3 Bayes Filters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.3.1 Kalman Filtering . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.3.2 Particle Filtering . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.4 Blind Source Separation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.4.1 Independent Component Analysis (ICA) . . . . . . . . . . . . . . . 36
3.4.2 Canonical Correlation Analysis (CCA) . . . . . . . . . . . . . . . . 39
3.5 Single-Channel Decomposition Methods . . . . . . . . . . . . . . . . . . . 40
3.5.1 Wavelet transform . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
3.5.2 Empirical Mode Decomposition . . . . . . . . . . . . . . . . . . . . 40
3.5.3 Morphological Component Analysis (MCA) . . . . . . . . . . . . . 41
3.6 Extension of BSS Methods for Single Channel Measurements . . . . . . . 42
3.6.1 Single-Channel ICA (SC-ICA) . . . . . . . . . . . . . . . . . . . . 42
3.6.2 Dynamical Embedding (DE) . . . . . . . . . . . . . . . . . . . . . 42
3.6.3 Wavelet ICA (WICA) . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.6.4 Empirical Mode Decomposition ICA (EMD-ICA) . . . . . . . . . . 44
3.7 Discussion and Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
4 Methodology for Comparison of Artifact Removal Techniques 49
4.1 Recording Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
4.1.1 fNIRS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
4.1.2 EEG . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
4.1.3 Accelerometer Data . . . . . . . . . . . . . . . . . . . . . . . . . . 52
4.2 Recording Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
4.2.1 fNIRS Experimental Setup . . . . . . . . . . . . . . . . . . . . . . 54
4.2.2 EEG Experimental Setup . . . . . . . . . . . . . . . . . . . . . . . 56
4.2.3 fNIRS Experimental Protocol . . . . . . . . . . . . . . . . . . . . . 57
4.2.4 EEG Experimental Protocol . . . . . . . . . . . . . . . . . . . . . . 60
4.3 Postprocessing of Recorded data . . . . . . . . . . . . . . . . . . . . . . . 62
4.3.1 Triggers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
4.3.2 EEG Post Processing . . . . . . . . . . . . . . . . . . . . . . . . . 63
4.3.3 fNIRS Post Processing . . . . . . . . . . . . . . . . . . . . . . . . . 64
4.3.4 Accelerometer Post Processing . . . . . . . . . . . . . . . . . . . . 66
4.4 Split of Training and Testing Data . . . . . . . . . . . . . . . . . . . . . . 68
4.5 Efficacy Metrics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
4.6 Importance of electrode/optode distance . . . . . . . . . . . . . . . . . . . 69
4.7 Automatic Artifact Removal . . . . . . . . . . . . . . . . . . . . . . . . . . 70
4.7.1 Ideal Automatic Artifact Removal for both fNIRS and EEG . . . . 72
4.7.2 Automatic Artifact Selection for fNIRS . . . . . . . . . . . . . . . 75
4.7.3 Automatic Artifact Selection for EEG . . . . . . . . . . . . . . . . 76
4.8 Synthetic Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
4.8.1 EEG . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
4.8.2 fNIRS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
4.8.3 Motion Artifact . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
4.9 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
5 Single Stage Artifact Removal Techniques 88
5.1 Data Tagging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
vii
Contents
5.2 Linear Time Invariant (LTI) Filtering Methods . . . . . . . . . . . . . . . 94
5.3 Wiener Filter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
5.3.1 The mathematics behind the Wiener filter . . . . . . . . . . . . . . 95
5.3.2 Requirements and limitations of the filter . . . . . . . . . . . . . . 98
5.3.3 Employing the Wiener filter . . . . . . . . . . . . . . . . . . . . . . 99
5.4 Kalman Filter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
5.4.1 The mathematics behind the filter . . . . . . . . . . . . . . . . . . 102
5.4.2 What input parameters are required by the user a priori? . . . . . 104
5.4.3 Employing the Kalman filter . . . . . . . . . . . . . . . . . . . . . 106
5.5 Morphological Component Analysis . . . . . . . . . . . . . . . . . . . . . . 107
5.5.1 Operation of the Morphological Component Analysis technique . . 107
5.5.2 Choice of the optimum dictionaries . . . . . . . . . . . . . . . . . . 110
5.5.3 Optimising the dictionary parameters . . . . . . . . . . . . . . . . 113
5.5.4 Employing the MCA algorithm . . . . . . . . . . . . . . . . . . . . 115
5.6 Adaptive Filter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
5.7 Results. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
5.7.1 Wiener Filter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
5.7.2 Kalman Filter. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
5.7.3 Morphological Component Analysis . . . . . . . . . . . . . . . . . 126
5.7.4 Adaptive Filter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
5.8 Discussion and Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
6 Two Stage Artifact Removal Techniques 136
6.1 Wavelet Transform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
6.1.1 Continuous Wavelet Transform . . . . . . . . . . . . . . . . . . . . 138
6.1.2 Discrete Wavelet Transform . . . . . . . . . . . . . . . . . . . . . . 140
6.1.3 Employing the wavelet transform . . . . . . . . . . . . . . . . . . . 145
6.2 Empirical Mode Decomposition . . . . . . . . . . . . . . . . . . . . . . . . 146
6.2.1 Operation of the technique . . . . . . . . . . . . . . . . . . . . . . 147
6.2.2 Ensemble Empirical Mode Decomposition (EEMD) . . . . . . . . . 152
6.3 Independent Component Analysis . . . . . . . . . . . . . . . . . . . . . . . 156
6.3.1 FastICA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
6.4 Canonical Correlation Analysis . . . . . . . . . . . . . . . . . . . . . . . . 161
6.4.1 Operation of CCA . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
6.5 Novel Combination of EEMD/Wavelets with CCA . . . . . . . . . . . . . 164
6.6 Combination of EEMD/Wavelets with ICA . . . . . . . . . . . . . . . . . 166
6.7 Results. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
6.7.1 Wavelet Transform . . . . . . . . . . . . . . . . . . . . . . . . . . . 168
6.7.2 (Ensemble) Empirical Mode Decomposition . . . . . . . . . . . . . 170
6.7.3 Wavelet-ICA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170
6.7.4 Wavelet-CCA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172
6.7.5 EEMD-ICA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
6.7.6 EEMD-CCA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175
6.8 Discussion and Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . 178
7 Case Study 187
7.1 EEG Measurement System . . . . . . . . . . . . . . . . . . . . . . . . . . 188
viii
Contents
7.2 Artifact Generation Protocol . . . . . . . . . . . . . . . . . . . . . . . . . 189
7.3 Efficacy Metrics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
7.4 Artifact Removal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193
7.4.1 Wiener Filter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
7.4.2 MCA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
7.4.3 Wavelet Transform . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
7.4.4 EEMD . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
7.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200
8 Conclusion 203
8.1 Summary of Contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . 204
8.2 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209
8.3 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211
A Publications Arising from this Work 212
B Data Sheet and Schematic for Accelerometer System 219
Bibliography 224
ix