Decomposition of the Cardiac and Respiratory Components from
Impedance Pneumography Signals
Marcel Mły
´
nczak
1
and Gerard Cybulski
1,2
1
Institute of Metrology and Biomedical Engineering, Faculty of Mechatronics, Warsaw University of Technology,
Boboli 8, 02-525, Warsaw, Poland
2
Department of Applied Physiology, Mossakowski Medical Research Centre, Polish Academy of Sciences, Pawinskiego 5,
02-106 Warsaw, Poland
Keywords:
Ambulatory Monitoring, Impedance Pneumography, Cardiorespiratory Activity, Decomposition.
Abstract:
Impedance pneumography (IP) measures changes of thoracic electrical impedance connected with change of
the air volume in the lungs. The electrode configuration used in IP applications causes that electrical heart
activity is visible in the IP signals. The aim of this paper is to assess the opportunity to decompose both respi-
ratory and cardiac components and its quality using various methods. Ten students performed static breathing
sequences, intended both for calibration and testing. Our prototype, Pneumonitor 2, and the reference pneu-
motachometer, were used. The accuracy of calculating tidal volume and heart rate, the calibration procedure
and the time of analysis, were considered. Mean 86.5% accuracy of tidal volume calculating and only 2.7%
error of heart rate estimation were obtained using moving average smoothing filters, for simple short record-
ing of free breathing calibration procedure, in three body positions. More sophisticated adaptive filtering also
provided good accuracy, however the processing time was 100-times higher, compared to simple methods.
It seems impedance pneumography, without ECG, could be enough for measuring basic cardiorespiratory
activity, particularly during ambulatory recordings, in which the least disturbing equipment is desirable.
1 INTRODUCTION
1.1 Problem
In nowadays there is increasing number of conditions,
when measuring respiration activity (e.g., tidal vol-
ume, TV ) could be necessary to improve the inference
about training, diagnostics or even treatment. There is
a strong need to perform such studies during natural
functioning of the subjects, due to reliability of data
gathered outside medical environment (Poupard et al.,
2008; Koivumaki et al., 2012).
However, direct method, pneumotachometry
(PNT ), could not be reliably carried out outside the
laboratory, due to its limitations (unportability and
use of face mask). From that perspective a devel-
opment of alternative methods, which could provide
the possibility to measure ventilation can be visi-
ble (Houtveen et al., 2006; Mły
´
nczak and Cybulski,
2012). One of those methods is impedance pneumog-
raphy (IP), which measures changes of transthoracic
electrical impedance, based on changes of amount of
air in the lungs. It could be used both in laboratory
and ambulatory settings (Seppa et al., 2010; Seppa
et al., 2013b).
The most common way of performing impedance
measurements is tetrapolar method, in which two
electrodes apply current signal and two remaining
measure voltage, which is related to the electrical
impedance. Various electrode configuration were
studied and utilized, however usually voltage elec-
trodes are positioned in the same place, on the midax-
illary line at about 5th-rib level (Seppa et al., 2013a).
This setting is similar to the one used in ambu-
latory ECG recordings, where single first ECG lead
configuration is applied. Shifting the placement from
the chest to the midaxillary line might change the
shape of the ECG signals, however it does not change
the timing relationships between particular waves and
cycles. Based on that, cardiac components were ob-
served in the IP signals, because of the electrical, not
mechanical, activity of the heart (Seppa et al., 2011).
The sample raw IP signal, obtained during the pre-
liminary measurements, is presented in the Figure 1.
In order to provide accurate respiratory param-
eters (particularly peak-related values, like TV , or
26
´
CyÅ
ˇ
Dczak M. and Cybulski G.
Decomposition of the Cardiac and Respiratory Components from Impedance Pneumography Signals.
DOI: 10.5220/0006107200260033
In Proceedings of the 10th International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC 2017), pages 26-33
ISBN: 978-989-758-212-7
Copyright
c
2017 by SCITEPRESS – Science and Technology Publications, Lda. All rights reserved
Figure 1: Sample of the raw IP signal consisting of both
respiratory and cardiac components.
maximum flow during inspiration and expiration)
these cardiac components should be removed and
somehow ignored from the I P signal.
However, preprocessing methods could impact on
the recorded signal. Firstly, by degrading the cor-
respondence between respiratory IP component and
reference signal in the areas of minimums and max-
imums of the signals. Secondly, due to the non-zero
mean value of the cardiac component of the IP signal
between the beginning of P wave and the ending of T
wave, particularly caused by QRS cycle and T wave.
1.2 Related Work
The problem has already been observed in the past
and some algorithms were used or introduced in the
literature, as in the author’s previous paper (Mły
´
nczak
et al., 2015).
Apart from simple filtration or moving average,
Savitzky-Golay-related smoothing method was car-
ried out, due to its well performance at minimums
and maximums (Savitzky and Golay, 1964). For the
graphical comparison the basic methods were applied
on the sample raw IP signal and the outputs were
showed in the Figure 2.
There are also more sophisticated approaches pre-
sented in the literature. Seppa et al. proposed
the method to remove cardiac components adaptively
based on simultaneous recording of ECG, which was
inspired from Schuessler’s work on removing cardio-
genic oscillations from esophageal pressure signals
(Seppa et al., 2011; Schuessler et al., 1998). Reinsh
et al. and Schoenberg et al. suggested smoothing
splines (Reinsch, 1967; Schoenberg, 1964; Poupard
et al., 2008). Yasuda and Barros offered the method
of filtering non-correlated noise in impedance cardio-
graphy, which could be applied for impedance pneu-
mography too (Barros et al., 1995).
Figure 2: The comparison of basic filtration methods out-
puts applied on the sample raw IP signal.
Empirical Modes Decomposition (EMD) and En-
semble Empirical Modes Decomposition (EEMD)
were also evaluated, however the choice of modes,
which are strictly related to respiratory activity, was
not clear and trivial (Wang et al., 2016). From the
other hand, wavelet denoising, based on different
wavelets families, seems to be reliable tool to remove
cardiac component in the IP signal as well (Math-
works, 2016).
Adaptive filtering and Scaled Fourier Linear Com-
biner (SFLC) was used by Yasuda et al. to assess
obstructive sleep apnea and central hypopnea based
on changes of thoracic impedance and HRV (Yasuda
et al., 2005). Cardiac-related artefacts could be also
attenuated during motion artefact removing process,
presented in Ansari et al. work (Ansari et al., 2016).
However, it is worth noted, that cardiac-related
component in the IP signal was usually treated as
noise (Seppa et al., 2010), and still all mentioned al-
gorithms intended to remove cardiac component from
the IP signal, in order respiratory one to be as shape-
connected to reference as it possible.
1.3 Objectives
Consequently, it seems important to evaluate whether
there is the possibility to separate both components
in IP signal and use both of them as valuable sig-
nals; respiratory component to calculate quantitative
volume-, flow- and time-related parameters, and car-
diac one to measure heart rate (HR), tachogram and
heart rate variability (HRV ) parameters.
Therefore, the aim of this paper is to assess
the quality of various preprocessing methods, which
could be applied on raw IP signal in order to separate
respiratory and cardiac components, concerning ve
Decomposition of the Cardiac and Respiratory Components from Impedance Pneumography Signals
27
aspects:
What calibration procedure could provide the best
data for further measurements?
What are determination coefficients (R
2
) of the
calibration model between respiratory component
of IP to the reference for calibration data?
What are inspiratory and expiratory tidal volumes
(TV
in
& TV
ex
) between respiratory component of
IP to the reference for testing data?
Is cardiac component subtracted from raw signal
is comparable to the single-lead E CG signal in
terms of HR and HRV calculation possibilities?
What is the analysis duration and complexity?
The goal of the work is to indicate the most ro-
bust algorithm from both respiratory, and cardiac per-
spective. Different compromise approaches were dis-
cussed.
2 METHODS
2.1 Subjects & Devices
The participants of the study were 10 generally
healthy students (all males, without any respiratory
disease reported), who were informed about the aim
of the study and wrote the consents.
The Table 1 presents the basic information about
the study group.
Table 1: Information about the study participants.
Minimum Mean Maximum
Weight [kg] 65.0 77.4 100.0
Height [cm] 171.0 179.3 187.0
BMI 20.75 24.14 33.41
Age 19 23 27
The Flow Measurement System with a Spirometer
Unit M909 and a Fleisch-type Heatable Flow Trans-
ducer 5530, with a Conical Mouthpiece M9114 con-
nected to the PNT sensor, was used without any re-
sistance as a reference device, made by Medikro Oy
(Kuopio, Finland). The system was calibrated with
the 3L syringe to provide accurate flow values every-
day.
IP signals were gained from our impedance pneu-
mography prototype, Pneumonitor 2, which also en-
ables to measure ECG signal and motion (from 3-axis
accelerometer module). Single-lead ECG signal from
our prototype was treated as reference to analyze car-
diac components extracted from IP.
Originally, sampling frequency of the reference
device was 200Hz, however due to the Pneumonitor
2 setting ( f
s
= 250Hz), all signals were transformed
(interpolated) to that sampling frequency.
2.2 Protocol & Analysis
The scheme of analysis is presented in the Figure 3.
IP
PNT
Filtering
Detrending
Integration
Breathing phases
establishing
Applying calibration coefficients
Preprocessed IP
Integrated PNT
Calibration
IP
Filtering
Detrending
Preprocessed IP
Decomposition
PNT
Integration
Integrated PNT
ECG
HRV curve
estimating & analysis
Cardiac IP Respiratory IP
Inspiration & Expiration
TV calculating & analysis
Calibration
Main measurements
Figure 3: The scheme of analysis, distinguishing signals,
operations and final analyses; the whole is performed for
each body position separately.
In order to deal with first question marked in ob-
jective subsection, we proposed three types of cali-
bration protocols, regarding time of registration and
breathing regularity:
1. free breathing registered for 30 seconds, consid-
ered as the simplest and the quickest procedure,
hereafter called ’Calibration Procedure 1’,
2. free breathing recorded for 2 minutes, to evalu-
ate the impact of longer measurement, hereafter
called ’Calibration Procedure 2’,
3. fixed breathing, shallow and deep alternately, 4
times each, for three frequencies, 6, 10 and 15
breaths per minute (BPM), to check, whether
adding various rates and depths of breathing may
improve the calibration quality meaningly, here-
after called ’Calibration Procedure 3’.
As we observed, body position has significant im-
pact on the values of calibration coefficients provided
for each subjects, therefore all 3 calibration procedure
were performed in three body positions (supine, sit-
ting and standing).
Finally, we asked to perform the test procedure
consisting of 6 normal breaths and then 6 deep breaths
(the difference was subjective), for three breathing
rates (6, 10 and 15 BPM) and for three body posi-
tions (listed above). That data were also used to per-
form calibration (hereafter called Calibration Pro-
cedure 4’), in order to evaluate the largest possible
BIOSIGNALS 2017 - 10th International Conference on Bio-inspired Systems and Signal Processing
28
accuracy, due to use of the same measurement both
for calibration and for testing.
The reference breathing phases for each recording
were calculated using simple amplitude thresholding
of the flow-related raw PNT signal. Three states, in-
spiration, expiration and breathing pause, were auto-
matically marked. Some heuristics were also added
in order to remove the artefacts and very short phases
and errors. Then, the volume-related reference sig-
nals were obtained by linearly detrended integration,
using Simpson’s rule, of pneumotachometry signal.
All IP signals (intended both for calibration and
for testing purposes) were next processed by various
techniques, listed below:
1. Moving average smoothing with 0.5 second win-
dow, considered as mild one.
2. Moving average smoothing with 1 second win-
dow, as proposed by Koivumaki et al. (Koivumaki
et al., 2012).
3. Moving average smoothing with 1.5 second win-
dow, intended to be strong one.
4. Savitzky-Golay 2nd-order filter with 25 probes
window (Savitzky and Golay, 1964).
5. Savitzky-Golay 7th-order filter with 25 probes
window.
6. Subtraction of raw IP signal and the noise com-
ponent from least mean square adaptive filtration,
then smoothed by 200 ms window.
7. The same subtraction as in algorithm 6, however
smoothed stronger, by 400 ms window.
8. The process of 25-fold decimation (performed
twice using 5-fold coefficient), then applying 10th
order least-square FIR filter with 1 Hz pass and
2.5 Hz stop frequencies, at the end the spline inter-
polation to return to original sampling frequency.
9. The same process as in algorithm 8, but use of
10th order stable Chebyshev IIR 1 Hz pass fre-
quency filter.
10. Wavelet denoising using soft heuristic SURE
thresholding and scaled noise option, on detail
coefficients obtained from the decomposition at
level 5 by ’sym8’ wavelet.
11. Wavelet denoising using minimax thresholding at
level 5 by ’db5’ wavelet.
12. Smoothing Splines, presented by Reinsh (Rein-
sch, 1967).
The output of each algorithm was treated as a res-
piratory component in IP signal.
The calibration was performed by calculating the
single coefficient from linear modeling of the volume-
related reference and processed respiratory IP com-
ponent, after detrending and mean removal (the post-
processing removes the need to take into account the
intercept coefficient of linear modeling). The cali-
bration was performed for signals gained during each
breathing protocols and for all three body positions.
The calibration coefficients and determination coeffi-
cients of the linear model were stored.
In order to compare TV values, separate inspira-
tory and expiratory TV were estimated as the differ-
ence between the extremums before and after each
breathing phase, e.g., for inspiration, TV
in
was calcu-
lated as the difference between the maximum found
during short breathing pause after inspiration and
before expiration, and the minimum found during
breathing pause before that inspiration, as suggested
by Poupard et al. (Poupard et al., 2008).
ECG signals were also compared for signals mea-
sured for all three body positions. However, due to
the fact, that for 4 participants reference ECG signals
were weak during sitting and standing, only those,
which were recorded during supine position, were
taken into account. The R points were automatically
marked using simple thresholding technique.
The cardiac IP component was estimated as a dif-
ference between raw signal and respiratory compo-
nent. We evaluated whether the possibility to extract
the R points from cardiac IP component was linked
with the equivalent of signal-to-noise ratio (SN, as-
sumed as the ratio between mean absolute value of the
cardiac IP component after mean removal, and mean
absolute value of the respiratory IP component after
mean removal as well).
All analyses were performed using MATLAB
software. The processing time of the algorithms
were measured with the computer processor Intel i5
(1200MHz), without any accelerations. It is worth
noted, that the specific numbers are of secondary im-
portance, the most essential is their mutual relation-
ship.
3 RESULTS
Breathing phases were firstly estimated from refer-
ence signal. The sample output of the algorithm for
the first participant, from 1st calibration procedure,
for supine body position, is presented in the Fig. 4.
Next, we performed the calibration for all signals
gained from all participants. In the Fig. 5 sample rela-
tionship between respiratory IP component and refer-
ence with linear model between them, are presented.
Decomposition of the Cardiac and Respiratory Components from Impedance Pneumography Signals
29
Figure 4: Sample output of the algorithm to detect breathing
phases from reference PNT signal.
Figure 5: Sample relationship between respiratory compo-
nent of impedance pneumography signal and reference, and
linear calibration model.
In the Tables 2 and 3 we gathered mean determina-
tion coefficients of linear models, and mean process-
ing times of the algorithms, respectively, depending
on the calibration procedures.
Further, we used calculated calibration coeffi-
cients to evaluate the accuracy of tidal volumes, mea-
sured both for inspiration and expiration. We gathered
all accuracies for both breathing phases in the Table 4,
distinguishing algorithm type and calibration proce-
dure. The accuracy is shown as absolute and relative
errors. We also presented the compatibility and the
Bland-Altman plots for the calculated tidal volumes,
for 1st calibration procedure and for 7th algorithm.
They are presented in the Fig. 6 and 7, respectively.
Table 2: The mean determination coefficient of the linear
model, depending on the algorithm and calibration proce-
dure.
Algorithm Procedure [%]
1 2 3 4
1 94.2 91.1 92.2 92.1
2 95.3 91.2 92.3 92.2
3 94.4 89.6 92.0 92.0
4 82.8 82.7 90.8 90.2
5 79.5 79.9 90.1 89.5
6 90.9 88.9 91.8 91.5
7 93.3 90.6 92.1 91.9
8 93.1 90.6 92.0 91.8
9 3.7 11.4 46.8 50.2
10 91.2 89.1 91.6 91.2
11 88.7 87.2 91.6 91.1
12 92.9 87.2 91.6 91.6
Table 3: The mean processing times of the algorithm, de-
pending on the calibration procedure.
Algorithm Procedure
1 2 3 4
1 [ms] 3.4 7.1 11.4 18.3
2 [ms] 4.4 9.9 11.6 15.9
3 [ms] 5.9 11.7 13.1 19.3
4 [ms] 5.6 12.3 13.5 18.3
5 [ms] 5.5 11.6 12.5 17.3
6 [s] 0.17 1.06 1.77 3.52
7 [s] 0.16 1.06 1.70 3.73
8 [ms] 24.4 26.0 36.3 42.3
9 [s] 0.15 0.14 0.17 0.19
10 [ms] 29.9 37.6 43.5 56.1
11 [ms] 24.8 29.8 27.9 38.8
12 [s] 1.63 25.53 45.16 106.09
Then, we compared the tachograms estimated
from reference ECG and cardiac component of IP
signal for all subjects, for supine body position. The
sample signals obtained for first subject are presented
in the Fig. 8.
The comparison of the accuracy of the heart rate
estimation for cardiac components derived from IP
signal using different algorithms is presented in the
Table 5.
The minimal overall error of cardiac calculations
from IP was obtained for third algorithm.
We observed no statistically significant correspon-
dence between the accuracy of cardiac calculations
from IP signals, and the SN ratio.
BIOSIGNALS 2017 - 10th International Conference on Bio-inspired Systems and Signal Processing
30
Figure 6: Compatibility plot for tidal volumes, calculated
both for inspirations’ and expirations’ values, for all partic-
ipants, for each body positions, for 1st calibration procedure
and for 7th algorithm.
Figure 7: Bland-Altman plot for tidal volumes, calculated
both for inspirations’ and expirations’ values, for all partici-
pants, for each body positions, for 1st calibration procedure
and for 7th algorithm.
4 DISCUSSION
In this work various algorithms were evaluated for
decomposition of the respiratory and cardiac com-
ponents in IP signal accuracy. Among them no one
seemed to be the best in every assessed aspect. Still
important thing remains, what will be the best criteria
for decomposition optimization.
The results showed, that from respiratory perspec-
tive moving average smoothing with short window
or 7th algorithm provided the best results, however
the time of operation was quite long for the second
Figure 8: Sample comparison of the cardiac IP component
with reference ECG signal, and the HRV curves derived
from those signals.
one. In cardiac analysis, moving average smoothing
with longer window appears to have the best compat-
ibility to ECG reference. It seems that mixing the
approaches, depending on the analyzed signal, could
give the most reliable results.
The results also demonstrated, that because of the
linear relationship between respiratory IP component
and reference volume signal, there is no need to dis-
tinguish the depth and rate of breathing during cali-
bration procedure. On the other hand, the shorter time
of calibration registration allows to get higher deter-
mination coefficient of linear model, and increase the
mean accuracy of tidal volume calculation.
The purpose of the testing was not to evaluate
the accuracy of using calibration coefficient indepen-
dently on body position; all results assumed different
coefficients calculated for different positions.
There are some limitations of this study. There
were only 10 participants, only males - this proba-
bly had no direct impact on the results, but it seems
worth to apply the analysis in women’s signals, which
are quite different, e.g., due to the geometry of the
chest. Further, the measurements were carried out
only in static conditions, without considering motion
artifacts; and there was no multi-lead ECG reference
device. At the end, in ambulatory situations, the reg-
istrations are longer and would be more diversified,
which may affect the overall accuracy.
Therefore, future plans include performing mea-
surements during dynamic conditions, which imitate
natural functioning of subjects, and then changing the
way of pre-processing (particularly detrending part)
from ”global” approach, to the ”local” one, in order
to deal with non-linear and non-stationary drift of the
baseline impedance.
From the analytics perspective, there is a need for
Decomposition of the Cardiac and Respiratory Components from Impedance Pneumography Signals
31
Table 4: The comparison of tidal volume estimating accu-
racy, regarding algorithm type and calibration procedure;
all absolute errors are in milliliters, except 9th algorithm
(marked in italics), in which are in liters, and all relative
errors are in percent; statistical significance is presented by
abbreviations: A - absolute error, R - relative error, p -
p-value of paired T test.
Alg. Procedure
1 2 3 4
1 A 214.7 240.5 251.0 165.3
R 13.8 16.0 17.1 11.8
p 0.91 0.90 0.83 0.76
2 A 206.0 245.7 284.2 205.7
R 13.5 16.7 19.3 14.4
p 0.74 0.77 0.77 0.71
3 A 238.1 278.5 338.7 273.9
R 15.6 19.1 23.2 19.2
p 0.57 0.61 0.71 0.68
4 A 317.9 298.9 251.9 181.2
R 19.8 18.6 18.1 14.0
p 0.91 0.94 0.88 0.84
5 A 343.5 311.8 264.1 200.6
R 21.6 19.5 19.8 16.1
p 0.91 0.93 0.87 0.84
6 A 246.8 255.5 246.4 160.7
R 16.1 17.0 17.2 12.1
p 0.96 0.94 0.83 0.75
7 A 223.8 243.0 247.9 161.8
R 14.7 16.4 17.2 11.9
p 0.90 0.88 0.81 0.71
8 A 234.3 251.2 240.1 153.0
R 14.9 16.5 16.5 11.2
p 0.99 0.99 0.88 0.80
9 A 1.60 1.49 1.19 1.13
R 97.1 92.0 85.5 82.2
p 0.94 0.72 0.01
0.00
10 A 246.6 258.4 248.0 162.3
R 15.7 16.9 17.0 11.8
p 0.94 0.96 0.92 0.90
11 A 267.9 273.1 249.3 165.5
R 17.1 17.7 17.3 12.3
p 0.92 0.94 0.92 0.89
12 A 309.9 320.9 379.2 317.4
R 19.3 21.6 26.2 22.5
p 0.56 0.59 0.72 0.74
further improvement and assessment of the compo-
nents’ decomposition, e.g., using time series algo-
rithms utilized in econometrics field.
The possibility to remove the classical ECG reg-
istration from ambulatory cardiorespiratory measure-
ments in the situations, in which direct sophisticated
analysis of multi-lead ECG signal is not necessary,
has to be evaluated in the broader manner.
Table 5: The comparison of heart rate derived from car-
diac IP components and compatibility of the shape of HRV
curve; only the best algorithms are presented; fifth subject
did not have any recognizable cardiac component, there-
fore was removed from presenting; abbreviations: A - abso-
lute error [BPM], R - relative error in percent, CC - cross-
correlation coefficient between HRV curves derived from
ECG reference and cardiac IP component, SN - the equiva-
lent of signal-to-noise ratio.
Subject Algorithm
2 3 7 12
1 A 0.05 0.18 0.47 0.16
R 0.1 0.3 0.7 0.2
CC 0.44 0.45 0.83 0.40
SN 0.09 0.10 0.06 0.09
2 A 3.91 1.26 1.68 3.70
R 5.8 1.9 2.5 5.5
CC 0.46 0.40 0.38 0.50
SN 0.40 0.46 0.32 0.51
3 A 2.17 1.04 3.99 2.53
R 3.2 1.5 6.0 3.8
CC 0.40 0.54 0.51 0.48
SN 0.15 0.17 0.13 0.19
4 A 0.66 1.38 0.12 0.66
R 1.1 2.3 0.2 1.1
CC 0.43 0.60 0.28 0.43
SN 0.56 0.57 0.53 0.56
6 A 17.25 8.38 28.86 12.26
R 32.7 15.9 54.6 23.2
CC 0.03 0.09 0.09 0.01
SN 0.34 0.42 0.26 0.43
7 A 0.01 0.01 0.02 0.01
R 0.01 0.02 0.02 0.02
CC 1.0 1.0 1.0 1.0
SN 0.25 0.26 0.22 0.25
8 A 0.05 0.05 0 0.05
R 0.1 0.1 0 0.1
CC 0.98 0.98 1.0 0.98
SN 0.20 0.22 0.20 0.21
9 A 0.02 0.03 0.02 0.02
R 0.02 0.04 0.02 0.02
CC 1.0 0.91 1.0 1.0
SN 0.32 0.37 0.31 0.39
10 A 5.62 1.44 0.87 1.61
R 8.6 2.2 1.3 2.5
CC 0.02 0.28 0.34 0.33
SN 0.13 0.15 0.11 0.15
5 CONCLUSIONS
Impedance pneumography could be used to measure
respiratory activity, intended primarily for ambulatory
conditions. In order to get volume values, the calibra-
BIOSIGNALS 2017 - 10th International Conference on Bio-inspired Systems and Signal Processing
32
tion for each participant is needed. However, short
recording of free breathing, for various body posi-
tions, seemed to be enough to get the highest accuracy
of tidal volume estimation.
Raw impedance signal obtained from the chest
consists of both respiratory and cardiac components,
the second most commonly regarded as an element
to be removed. The study showed, that there is the
possibility to measure and extract each component
from impedance pneumography separately, with re-
liable accuracy, 86.5% and 97.3%, respectively for
tidal volume and heart rate estimation.
Simple moving average smoothing (with 1s win-
dow for respiratory analysis, and with 1.5s window
for cardiac one) were the best algorithm regarding
compromise between tidal volume and heart rate ac-
curacy, and time of processing. More sophisticated
adaptive filtering also provided good accuracy, how-
ever the processing time was 100-times higher, com-
paring to simple methods.
Cardiac component is not equally visible in ev-
ery participant, however obtained compatibility be-
tween ECG reference seems promising, particularly
concerning ambulatory long-term measurements.
ACKNOWLEDGEMENTS
This study was supported by the research programs of
institutions the authors are affiliated with.
REFERENCES
Ansari, S., Ward, K., and Najarian, K. (2016). Motion Ar-
tifact Suppression in Impedance Pneumography Sig-
nal for Portable Monitoring of Respiration: an Adap-
tive Approach. IEEE J. Biomed. Heal. Informatics,
11(4):1–1.
Barros, A. K., Yoshizawa, M., and Yasuda, Y. (1995). Fil-
tering noncorrelated noise in impedance cardiogra-
phy. IEEE Transactions on Biomedical Engineering,
42(3):324–327.
Houtveen, J. H., Groot, P. F. C., and De Geus, E. J. C.
(2006). Validation of the thoracic impedance derived
respiratory signal using multilevel analysis. Int. J.
Psychophysiol., 59(2):97–106.
Koivumaki, T., Vauhkonen, M., Kuikka, J. T., and Haku-
linen, M. a. (2012). Bioimpedance-based measure-
ment method for simultaneous acquisition of respi-
ratory and cardiac gating signals. Physiol. Meas.,
33(8):1323–1334.
Mathworks (2016). Wavelet Denoising.
http://www.mathworks.com/help/wavelet/ug/wavelet-
denoising.html. [Online; accessed 07-August-2016].
Mły
´
nczak, M. and Cybulski, G. (2012). Impedance pneu-
mography: Is it possible? In XXX Symp. Photonics
Appl. Astron. Commun. Ind. High-Energy Phys. Exp.
(Wilga 2012), volume 8454, pages 84541T–19.
Mły
´
nczak, M., Niewiadomski, W.,
˙
Zyli
´
nski, M., and Cybul-
ski, G. (2015). Verification of the Respiratory Param-
eters Derived from Impedance Pnumography during
Normal and Deep Breathing in Three Body Postures.
In IFMBE Proc., volume 45, pages 162–165.
Poupard, L., Mathieu, M., Sart
`
ene, R., and Goldman, M.
(2008). Use of thoracic impedance sensors to screen
for sleep-disordered breathing in patients with cardio-
vascular disease. Physiol. Meas., 29(2):255–267.
Reinsch, C. H. (1967). Smoothing by spline functions. Nu-
merische Mathematik, 10(3):177–183.
Savitzky, A. and Golay, M. J. E. (1964). Smoothing and
Differentiation of Data by Simplified Least Squares
Procedures. Anal. Chem., 36(8):1627–1639.
Schoenberg, I. J. (1964). Spline functions and the problem
of graduation. Proceedings of the National Academy
of Sciences, 52(4):947–950.
Schuessler, T. F., Gottfried, S. B., Goldberg, P., Kearney,
R. E., and Bates, J. H. T. (1998). An adaptive filter
to reduce cardiogenic oscillations on esophageal pres-
sure signals. Ann. Biomed. Eng., 26(2):260–267.
Seppa, V. P., Hyttinen, J., Uitto, M., Chrapek, W., and
Viik, J. (2013a). Novel electrode configuration for
highly linear impedance pneumography. Biomed. Eng.
/ Biomed. Tech., 58(1):35–38.
Seppa, V.-P., Hyttinen, J., and Viik, J. (2011). A method
for suppressing cardiogenic oscillations in impedance
pneumography. Physiol. Meas., 32(3):337–345.
Seppa, V. P., Uitto, M., and Viik, J. (2013b). Tidal breath-
ing flow-volume curves with impedance pneumogra-
phy during expiratory loading. In Proc. Annu. Int.
Conf. IEEE Eng. Med. Biol. Soc. EMBS, volume 2013,
pages 2437–2440.
Seppa, V.-P., Viik, J., and Hyttinen, J. (2010). Assessment
of pulmonary flow using impedance pneumography.
IEEE Trans. Biomed. Eng., 57(9):2277–2285.
Wang, Z., Wu, D., Chen, J., Ghoneim, A., and Hossain,
M. A. (2016). A Triaxial Accelerometer-Based Hu-
man Activity Recognition via EEMD-Based Features
and Game-Theory-Based Feature Selection. IEEE
Sens. J., 16(9):3198–3207.
Yasuda, Y., Umezu, A., Horihata, S., Yamamoto, K.,
Miki, R., and Koike, S. (2005). Modified thoracic
impedance plethysmography to monitor sleep apnea
syndromes. Sleep Med., 6(3):215–224.
Decomposition of the Cardiac and Respiratory Components from Impedance Pneumography Signals
33