Medical Engineering and Physics 38 (2016) 671–678

Contents lists available at ScienceDirect

Medical Engineering and Physics

journal homepage: www.elsevier.com/locate/medengphy

Communication

A robust approach for ECG- based analysis of cardiopulmonary

coupling

Jiewen Zheng

a , Weidong Wang

b , Zhengbo Zhang

b , c , ∗, Dalei Wu

d , Hao Wu

b , Chung-kang Peng

e , f

a The Quartermaster Research Institute of the General Logistic Department, Beijing, P R China b Department of Biomedical Engineering, Chinese PLA (People’s Liberation Army) General Hospital, Beijing, P R China c Harvard-MIT Division of Health Sciences and Technology, Massachusetts Institute of Technology, Cambridge, MA 02139, USA d Department of Computer Science and Engineering, The University of Tennessee at Chattanooga, Chattanooga, TN 37403, USA e Center for Dynamical Biomarkers and Translational Medicine, National Central University, Chungli, Taiwan f Division of Interdisciplinary Medicine & Biotechnology and Margret & H.A. Rey Institute for Nonlinear Dynamics in Medicine, Beth Israel Deaconess

Medical Center, Harvard Medical School, Boston, MA 02215, USA

a r t i c l e i n f o

Article history:

Received 8 January 2015

Revised 8 December 2015

Accepted 22 February 2016

Keywords:

ECG-derived respiration

Cardiopulmonary coupling

Adaptive filtering

Phase synchronization

Ensemble empirical mode decomposition

a b s t r a c t

Deriving respiratory signal from a surface electrocardiogram (ECG) measurement has advantage of simul-

taneously monitoring of cardiac and respiratory activities. ECG-based cardiopulmonary coupling (CPC)

analysis estimated by heart period variability and ECG-derived respiration (EDR) shows promising ap-

plications in medical field. The aim of this paper is to provide a quantitative analysis of the ECG-based

CPC, and further improve its performance. Two conventional strategies were tested to obtain EDR signal:

R-S wave amplitude and area of the QRS complex. An adaptive filter was utilized to extract the com-

mon component of inter-beat interval (RRI) and EDR, generating enhanced versions of EDR signal. CPC

is assessed through probing the nonlinear phase interactions between RRI series and respiratory signal.

Respiratory oscillations presented in both RRI series and respiratory signals were extracted by ensemble

empirical mode decomposition for coupling analysis via phase synchronization index. The results demon-

strated that CPC estimated from conventional EDR series exhibits constant and proportional biases, while

that estimated from enhanced EDR series is more reliable. Adaptive filtering can improve the accuracy

of the ECG-based CPC estimation significantly and achieve robust CPC analysis. The improved ECG-based

CPC estimation may provide additional prognostic information for both sleep medicine and autonomic

function analysis.

© 2016 IPEM. Published by Elsevier Ltd. All rights reserved.

1

e

[

t

s

i

p

k

e

c

w

G

a

u

t

s

n

R

i

p

d

c

n

a

h

1

. Introduction

Human cardiovascular and respiratory systems interact with

ach other and present effects of modulation and synchronization

1,2] . The cardiovascular and cardiorespiratory systems are charac-

erized by a complex interplay of several linear and nonlinear sub-

ystems [3] . Analysis of cardiopulmonary coupling (CPC) provides

nformation about the coupling dynamics between the cardiores-

iratory systems at different time scales [4] , leading to improved

nowledge of the interacting regulatory mechanisms under differ-

nt physiological and pathophysiological conditions [1] . Currently,

ardiac inter-beat (R–R) interval (RRI) dynamics analysis has been

idely investigated in the evaluation of autonomic nervous system

∗ Corresponding author. Department of Biomedical Engineering, Chinese PLA

eneral Hospital, 28 Fuxing Rd., Beijing, PR China. Tel.: + 86 10 66937921.

E-mail address: zhengbozhang@126.com (Z. Zhang).

p

s

p

d

ttp://dx.doi.org/10.1016/j.medengphy.2016.02.015

350-4533/© 2016 IPEM. Published by Elsevier Ltd. All rights reserved.

ctivities [5] . Compared with RRI dynamics analysis, CPC analysis

tilizes information presented in both cardiovascular and respira-

ory variables. It has been demonstrated that CPC analysis during

leep can characterize different sleep stages and identify sleep ap-

ea events, overcoming some potential limitations encountered in

RI dynamics analysis techniques [6,7] . A recent study shows the

mportance of monitoring CPC, proving that the strength of CPC

rogressively decreased as a function of sympathetic activation in-

uced by head-up tilt test [8] . Therefore, CPC analysis might over-

ome or at least complement traditional univariate analysis tech-

iques such as RRI dynamics analysis.

For the investigation of cardiorespiratory interaction, bivariate

pproaches are commonly applied [1] . For CPC analysis, both heart

eriod variability and respiratory signal are required. Respiratory

ignal is usually measured by motion or volume change during res-

iration, or by airflow [9] . Compared with other techniques, ECG-

erived respiration (EDR) is a more convenient and cost-effective

672 J. Zheng et al. / Medical Engineering and Physics 38 (2016) 671–678

Fig. 1. Structure of adaptive estimation of respiratory signal by RLS algorithm, with

EDR series as the primary input and RRI series as the reference input.

m

i

h

w

a

0

2

n

o

7

[

a

r

r

p

y

c

a

2

2

p

f

i

t

e

i

i

2

t

Q

t

E

w

e

d

m

2

i

a

p

o

t

t

s

a

alternative for respiration monitoring, which derives respiratory

signal from ECG recording without using special-purpose hardware

[10] . Hence, with solely ECG recording, information of heart period

variability and respiratory signal can be derived and CPC analysis

can be achieved. Recent studies have demonstrated the effective-

ness of EDR and ECG-based CPC analysis as a screening or diag-

nostic tool for sleep disordered breathing [6,7,11,12] .

EDR is a signal-processing technique deriving respiratory wave-

forms from ordinary ECG [10,13] . Recently, many algorithms have

been developed such as amplitude modulation of ECG wave [14] ,

autoregressive model-based method on heart rate series [15] ,

band-pass filter on ECG waveform [16] , and other methods com-

bining both beat morphology and HR information [17,18] . For sin-

gle ECG recording, amplitude-based EDR algorithms have been re-

ported with satisfactory performance, especially in the context of

sleep apnea monitoring [19,20] . Two forms of amplitude-based al-

gorithms are commonly studied and implemented: R-S wave am-

plitude and area of the QRS complex [10,16,21] . As EDR is an in-

direct recovery of respiratory waveform from recorded ECG, it is

relatively noisy. Many studies have been performed to evaluate

and improve its performance [9,11,16–19,22–24] . However, most of

these studies focused on the accuracy of breathing rate detection

[9,16,18,22,24] . Little attention has been paid on the performance

of EDR for CPC analysis [6,12] .

The aim of our work is to provide a quantitatively evaluation

on the ECG-based CPC analysis, compared to CPC values assessed

from real respiratory signal such as respiratory inductive plethys-

mography (RIP). As EDR is a surrogate respiratory signal and rel-

atively noisy, a recursive least squares (RLS) adaptive filter is em-

ployed to improve its performance for CPC analysis. In this study,

CPC was assessed through probing the nonlinear phase interac-

tions between the corresponding components extracted from RRI

series and respiratory signals. The performance of the ECG-based

CPC analysis was tested on paced breathing (PB) dataset and Fan-

tasia Dataset [25] . In the PB dataset, a stepwise PB procedure was

performed to generate different levels of cardiorespiratory interac-

tion, testing the degree of agreement between the ECG-based CPC

and the gold standard through the range of measurements. As el-

derly subjects usually exhibit attenuated amplitude of respiratory

sinus arrhythmia (RSA) [26] , we are curious to explore the effec-

tiveness of the RLS adaptive filter in improving CPC estimation for

elderly subjects based on the Fantasia dataset. Two datasets both

have respiratory signals recorded by RIP. Compared with other res-

piration monitoring techniques, RIP has the advantages of greater

accuracy and better sensitivity [27] .

2. Materials and methods

2.1. Participants and protocol

2.1.1. PB dataset

We collected the PB dataset by a stepwise PB protocol, which

was reviewed and approved by the Ethics Committee of Chinese

PLA General Hospital. Twenty healthy volunteers (8 females and

12 males, with an average age of 30) participated in this experi-

ment. Volunteers were excluded when they had a history of respi-

ratory, cardiovascular or neurological diseases. During the exper-

iment, each volunteer underwent a stepwise PB procedure in a

sitting position at rest. The stepwise PB procedure consisted of

one 3-min spontaneous breathing session and six PB sessions. Each

PB session had a pre-defined breathing rate with a constant in-

spiration to expiration ratio (1:2), lasting 3 min . The six PB ses-

sions were arranged in a stepwise order with guiding breathing

rate changing from high to low in a protocol of [14, 12.5, 11,

9.5, 8 and 7] breath/min (BPM). During the stepwise PB procedure,

participants were guided by six melodies to perform the six PB

aneuvers. Each melody comprises two different tones, with a ris-

ng tone guiding inhalation while a lower tone accompanying ex-

alation. Physiological variables of ECG (lead II) and respiration

ere recorded during the experiment. All signals were sampled

t 10 0 0 Hz with amplifier bandwidths of 0.05–100 Hz for ECG and

.05–20 Hz for respiration [28] .

.1.2. Fantasia dataset

The Fantasia dataset includes continuous recording of sponta-

eous respiration and ECG signals from twenty young (21 –34 years

ld, mean age: 26) and twenty elderly (68 –85 years old, mean age:

5) rigorously screened healthy subjects in supine resting position

25] . Each subgroup of subjects includes equal numbers of men

nd women. In this dataset, each record includes ECG and respi-

ation sampled at 250 Hz, lasting 120 min . In our study, from each

ecord, we randomly extracted 5-min stable signals to compare the

erformance of the ECG-based CPC analysis to the reference anal-

sis from real respiratory signal. For the breath-by-breath phase

oupling algorithm used in our study (see Section 2.3 ), 5 min data

re enough to produce accurate CPC estimation.

.2. Signal pre-processing

.2.1. QRS complex detection

To achieve accurate QRS complex detection and EDR extraction,

ower line interference and baseline wander were first removed

rom raw ECG waveforms. ECG beat detection was performed us-

ng Hamilton & Tompin’s QRS detector [29] , and each beat annota-

ion was verified by visual inspection. With this algorithm, we got

ach R wave location and RRI time series. Normal-to-normal sinus

ntervals, extracted from the R –R interval time series, were linearly

nterpolated at an even interval of 0.2 s (i.e., 5 Hz).

.2.2. Deriving two basic types of EDR

The two commonly used EDR algorithms: R-S wave ampli-

ude (with EDR series noted as EDR_RS) [16,21] and area of the

RS complex [10] (with EDR series noted as EDR_area) were

ested to obtain EDR on a beat-to-beat basis. Both EDR_RS and

DR_area series were interpolated to 5 Hz. These two EDR series

ill further be processed by an adaptive filter to generate two

nhanced types of EDR. RIP signals in both the PB and Fantasia

atasets were also down-sampled to 5 Hz through a decimation

ethod.

.2.3. Enhanced types of EDR by adaptive filtering

The structure of the adaptive filter used in our study is shown

n Fig. 1 . In our study, we chose EDR series as the primary input

nd RRI series as the reference input to the adaptive filter. The out-

ut of the adaptive filter (EDR

′ in Fig. 1 ) is the common component

f respiration-induced oscillations presented in both EDR and RRI

ime series. RLS algorithm was employed by taking into account

he advantages of fast convergence speed and without eigenvalue

pread problem [18] . Based on the two basic types of EDR (EDR_RS

nd EDR_area), two enhanced types of EDR (noted as EDR_RS_filt

J. Zheng et al. / Medical Engineering and Physics 38 (2016) 671–678 673

Fig. 2. Time series of RRI, RIP, EDR_RS, EDR_area, EDR_RS_filt and EDR_area_filt under various paced breathing conditions from one subject.

f

R

t

2

r

c

e

or EDR_RS and EDR_area_filt for EDR_area) were derived by the

LS adaptive filter, respectively. In our case, the forgetting factor of

he RLS algorithm was set to 0.996 and the length of the filter to

0 [18] .

With the aforementioned signal pre-processing steps, time se-

ies of RRI, RIP, EDR_RS, EDR_area, EDR_RS_filt and EDR_area_filt

an be acquired. Fig. 2 shows the time series of RRI and differ-

nt types of respiratory signals (RIP and EDR based) under various

674 J. Zheng et al. / Medical Engineering and Physics 38 (2016) 671–678

Fig. 3. An example of the original time series and their corresponding dominant respiratory component (DRC) extracted from the original time series in the PB dataset. (a)

The original time series of RIP, RRI and different types of EDR, (b) DRCs extracted from the original time series by EEMD approach.

2

t

d

p

a

w

v

s

p

p

s

P

w

a

t

n

W

t

O

w

i

c

t

(

d

s

2

p

p

T

a

v

i

t

w

paced breathing conditions from one subject. For comparison and

clarity, all time series were normalized.

2.2.4. Extraction of respiratory component by EEMD

Cardiovascular and cardiorespiratory systems are characterized

by a complex interplay of several linear and nonlinear subsystems

[3] , and interactions between heart rate variability and respira-

tion are characterized by alternation of periods of quasiperiodicity

(weak coupling) with periods of relatively high coordinated activ-

ity [30] . Therefore, CPC is an intermittent phenomenon character-

ized by non linear dynamics and non-stationarities. To address this

challenge, ensemble empirical mode decomposition (EEMD) was

applied to all the cardiorespiratory series and respiration induced

oscillations presented in the respiratory band were extracted to as-

sess the CPC [31] .

EEMD is a noise-assisted EMD algorithm [31] , which was de-

signed to extract dynamical information from non-stationary and

nonlinear signals at different time scales [32,33] . The essence of

the EMD is to decompose a signal into different components called

intrinsic mode function (IMF) that satisfy the necessary conditions

for a meaningful Hilbert transform calculation. Compared with

EMD, EEMD can be used without subjective intervention and pro-

duces a set of IMFs that bear the full physical meaning by elim-

inating the issue of mode mixing [31] . In our case, since ventila-

tion is the driving factor for respiration-induced oscillations pre-

sented in all the series, we selected the IMF in the respiratory

band as the dominant respiratory component (DRC) [34] . For the

Fantasia dataset, it was observed that respiration-related oscilla-

tions in all cardiorespiratory series lie in IMFs 3-5, we identified

DRC in each cardiorespiratory series from IMFs 3-5 according to

the average frequency of each IMF. For the PB dataset, it was found

that respiration-related oscillations in all series lie in IMFs 3-6

during the PB procedure, however, the dominant respiratory IMF

might differ between different PB sessions. Therefore, for the PB

dataset, after decomposing each time series by EEMD, we further

segmented each series into seven subsections (one spontaneous

breathing subsection and six PB subsections) according the step-

wise PB protocol. Thus, for each series in each subsection, IMF in

the respiratory band was determined as the DRC. Fig. 3 shows the

original cardiorespiratory series and their corresponding DRCs ex-

tracted from the PB dataset.

.3. Phase coupling analysis

The concept of phase synchronization has been developed for

he identification of interactions between respiratory and car-

iovascular systems [1,35] , and recent studies have shown that

hase dynamics are more sensitive to coupling strength than other

mplitude-derived markers [34,36] . Therefore, in our study, CPC

as estimated from Hilbert transformation of the extracted DRC

ia phase synchronization index. With the extracted DRC from each

eries, Hilbert transformation was applied to provide instantaneous

hase of signal, and point-by-point phase difference between res-

iratory signal and RRI was calculated. We quantified the phase

ynchronization as

= ( cos ϕ i ) 2 + ( sin ϕ i )

2 (1)

here ϕ i is the point-by-point phase difference of paired series,

nd ( cos ϕ i ) and ( sin ϕ i ) represent the average value during a cer-

ain time interval [37] .

If phase difference ϕ i = ϕ Resp − ϕ RRI ( ϕ Resp for respiratory sig-

al and ϕ RRI for RRI) is constant, both series are synchronized.

hen respiratory signal and RRI are highly synchronized, the dis-

ribution of phase difference ϕ i should be within a narrow range.

therwise, for non-synchronized signals, the distribution should be

ide. The phase synchronization was quantified by this normalized

ndex ranging from 0 to 1, with larger values indicating higher syn-

hronization level between cardiorespiratory systems. Fig. 4 shows

he extracted DRCs from each time series at the PB rate of 14 BPM

from the same subject as in Fig. 2 ), along with the corresponding

istribution of the phase difference between RRI and respiratory

ignal (RIP and different types of EDR).

.4. Statistical analysis

To determine the performance of the ECG-based CPC analysis,

hase couplings assessed from different types of EDR were com-

ared with the gold standard assessed from RIP using both paired

-test and Wilcoxon rank sum test. The effectiveness of the RLS

daptive filter in improving the CPC estimation was tested. A p

alue of less than 0.05 (two-sided) was used as the level of signif-

cance. Results are reported as mean ± standard deviation. To test

he degree of agreement between the ECG-based CPC estimation

ith the gold standard, Bland –Altman plot was also performed

J. Zheng et al. / Medical Engineering and Physics 38 (2016) 671–678 675

Fig. 4. An example of the extracted DRCs from each cardio-respiratory time series at the PB rate of 14 BPM, and the corresponding distribution of phase difference (a)

between RRI and respiratory signal ; (b) between RRI and RIP; (c) between RRI and EDR_area; (d) between RRI and EDR_RS; (e) between RRI and EDR_area_filt; (f) between

RRI and EDR_RS_filt .

[

R

3

3

d

E

e

c

d

d

s

C

s

t

b

T

f

t

e

S

s

s

t

n

3

d

e

t

c

s

e

b

t

i

m

38] . All statistical analyses were performed using MATLAB version

2010b (MathWorks, Natick, Massachusetts, USA).

. Results

.1. CPC estimation

The results of CPC estimation are shown in Table 1 for the PB

ataset. From Table 1 , it can be seen that the two basic types of

DR underestimate CPC in all breathing conditions, while the two

nhanced types of EDR produce more accurate CPC values very

lose to the gold standard estimated from RIP. As one subject un-

ertook seven breathing maneuvers during the stepwise PB proce-

ure, for each subject we took the mean value of seven CPC mea-

ures for statistical analysis. It was found that the differences in

PC estimation between the two basic types of EDR and the gold

tandard are relatively large and statistically significant. However,

here are no statistically significant differences in CPC estimation

etween the two enhanced types of EDR and the gold standard.

he results clearly indicate that adaptive filter can improve the per-

ormance of the ECG-based CPC estimation significantly for both

wo basic EDR algorithms. Moreover, there are no significant differ-

nces in CPC estimation between the two enhanced types of EDR.

imilar conclusions can be observed from the Fantasia dataset, as

hown in Table 2 . The results from the Fantasia dataset demon-

trated that even for aged group with potential attenuated ampli-

ude of RSA, ECG-based CPC estimation can also be improved sig-

ificantly by adaptive filtering.

.2. Bland –Altman plot of the CPC estimation

Fig. 5 shows the Bland –Altman plot of CPC estimation in the PB

ataset. It is very clear that both two enhanced types of EDR gen-

rate lower biases in CPC estimation (as shown in Fig. 5 (c) and (d))

han the basic types of EDR (as shown in Fig. 5 (a) and (b)). Similar

onclusion goes to the Fantasia dataset (Bland –Altman plot was not

hown). Moreover, from Fig. 5 (a) and (b), it was found that there

xists a proportional bias in CPC estimation from either of the two

asic types of EDR. The basic types of EDRs tend to produce bet-

er CPC estimations when the breathing rate gets lower. The ex-

stence of the proportional bias indicates that ECG-based CPC esti-

ation from either of the two basic types of EDR does not perform

676 J. Zheng et al. / Medical Engineering and Physics 38 (2016) 671–678

Table 1

Comparison of CPC estimation between RRI and different types of respiratory signal in the PB dataset at different breathing conditions, with

CPC values assessed from RIP signal as gold reference. The results are reported as mean ± standard deviation.

Respiratory

signals

Different breathing conditions ( n = 20) Combined

results

( n = 20) SB 14 BPM 12.5 BPM 11 BPM 9.5 BPM 8 BPM 7 BPM

RIP 0.86 ± 0.10 0.93 ± 0.04 0.93 ± 0.05 0.93 ± 0.05 0.90 ± 0.08 0.88 ± 0.09 0.82 ± 0.15 0.89 ± 0.09

EDR_area 0.69 ± 0.13 0.80 ± 0.18 0.83 ± 0.11 0.83 ± 0.17 0.81 ± 0.16 0.79 ± 0.17 0.78 ± 0.15 0.79 ± 0.16 ∗

EDR_RS 0.75 ± 0.12 0.78 ± 0.14 0.82 ± 0.12 0.86 ± 0.09 0.84 ± 0.16 0.81 ± 0.20 0.80 ± 0.15 0.81 ± 0.14 §

EDR_area_filt 0.80 ± 0.17 0.86 ± 0.13 0.92 ± 0.08 0.93 ± 0.08 0.90 ± 0.12 0.89 ± 0.10 0.82 ± 0.13 0.87 ± 0.13 ♦

EDR_RS_filt 0.85 ± 0.08 0.90 ± 0.08 0.93 ± 0.06 0.93 ± 0.06 0.92 ± 0.07 0.89 ± 0.10 0.82 ± 0.16 0.89 ± 0.10 #

Note : SB: spontaneous breathing; BPM: breathing per minute. ∗ p < 0.001 for comparison of CPC estimation between RIP and EDR_area (both by paired T -test and Wilcoxon rank sum test) . § p < 0.001 for comparison of CPC estimation between RIP and EDR_RS (both by paired T -test and Wilcoxon rank sum test) . ♦ p = 0.302 by paired T -test and p = 0.352 by Wilcoxon rank sum test for comparison of CPC estimation between RIP and EDR_area_filt . # p = 0.757 by paired T -test and p = 0.979 by Wilcoxon rank sum test for comparison of CPC estimation between RIP and EDR_RS_filt .

Table 2

Comparison of CPC estimation between RRI and different types of respiratory signal in the Fanta-

sia dataset, with CPC values assessed from RIP signal as gold reference. The results are reported as

mean ± standard deviation.

Different groups RIP EDR_area EDR_RS EDR_area_filt EDR_RS_filt

Young group ( n = 20) 0.80 ± 0.10 0.58 ± 0.20 0.56 ± 0.14 0.80 ± 0.11 0.77 ± 0.12

Elderly group ( n = 20) 0.69 ± 0.19 0.48 ± 0.20 0.50 ± 0.18 0.71 ± 0.13 0.70 ± 0.12

Total ( n = 40) 0.75 ± 0.15 0.53 ± 0.20 0.53 ± 0.16 0.76 ± 0.13 0.74 ± 0.12

o

a

e

d

s

t

t

p

e

t

a

t

d

p

b

g

a

e

t

s

s

p

a

c

l

t

p

s

s

a

t

d

r

g

e

r

i

v

consistently through the range of measurements. In contrast, for

the enhanced types of EDR, there are no obvious proportional bi-

ases in CPC estimation. This result indicates that the enhanced

types of EDRs can produce robust ECG-based CPC estimations in

different breathing conditions.

4. Discussion

Prior work has documented the effectiveness of EDR for breath-

ing rate detection [16,22–24] . Several techniques for improving the

signal quality of EDR have been developed [24] . However, little at-

tention has been paid on the performance of EDR for CPC analy-

sis. In this study, we not only provided a quantitatively analysis on

the ECG-based CPC, but also used the RLS adaptive algorithm to

improve its performance. Moreover, to achieve more reliable phase

coupling analysis, EEMD was applied to cardiorespiratory signals to

extract respiration-related oscillations adaptively, instead of using a

low-pass filter with a constant cut-off frequency [37] . Our results

demonstrate that the basic EDR algorithms underestimate CPC in

all breathing conditions, while the enhanced types of EDR by adap-

tive filtering can produce more accurate and robust CPC estimation.

The Bland –Altman plots in Fig. 5 show that there are propor-

tional biases in CPC estimation for both two basic types of EDR.

The biases at high breathing rates are larger than those at low

breathing rates. This phenomenon is easy to explain. As breath-

ing rate gets lower, the synchronization between heart rate and

respiration becomes stronger. Therefore, basic form of EDR tends

to perform better in CPC analysis at low breathing frequency. The

enhanced types of EDR not only reduced the biases in CPC analy-

sis, but also eliminated the proportional biases. Thus, the enhanced

types of EDR can achieve robust ECG-based CPC analysis in differ-

ent breathing conditions.

As elderly subjects exhibit reduced amplitude of RSA [26] , the

effectiveness of adaptive filter in improving CPC estimation was

tested on elderly subjects by using the Fantasia dataset. The re-

sults clearly indicated that even for elderly subject group, CPC es-

timation could also be improved significantly by adaptive filter. In

this study, the forgetting factor of RLS filter was set to 0.996 and

the length of the filter to 20. Better results might be achieved by

ptimizing the parameters of the forgetting factor and filter length

ccording to the clinical applications under consideration.

One interesting phenomena was found that the strength of CPC,

ither estimated from the direct respiratory measurement or from

ifferent types of EDR-based measurements, was improved over

pontaneous breathing during the first three PB procedures, and

hen progressively deteriorated when the breathing rate was lower

han 9.5 BPM. Amplitude and phase dynamics are two major as-

ects in CPC investigation, and amplitude dynamics was consid-

red in most of studies [4] . In our previous study, we investigated

he effect of PB on the amplitude of RRI and demonstrated that the

mplitude-modulation effect of PB becomes more evident when

he breathing rate decreases [39] . The variation of phase coupling

uring the PB procedure indicates that in contrast to the RRI am-

litude, the PB rate might have an optimal range to achieve the

est coherent state between respiration and heart rate.

Our results provide compelling evidence that the basic EDR al-

orithms underestimate CPC, and the enhanced types of EDR by

daptive filtering are effective in improving CPC estimation. How-

ver, some limitations are worth noting. It should be noted that

he CPC investigated in our study is different from the ECG-based

leep spectrogram proposed by Thomas et al. [6] The ECG-based

leep spectrogram combines both amplitude and frequency cou-

ling, taking into account of coherence and cross spectrum of RRI

nd EDR [6] . The CPC investigated in our study reflects the syn-

hronization between respiration and RSA. Hence, it should be re-

ated to the coherence in the frequency domain. Therefore, to test

he value of our algorithm in sleep medicine, information of am-

litude coupling may also need to be included for coupling analy-

is. The future work of our study is to calculate the ECG-derived

leep spectrogram described in [6] by integrating adaptive filter

nd EEMD algorithms, testing whether these approaches can lead

o improvement in sleep quality measurement and sleep apnea

etection. Second, since the RLS filter uses the RRI series as one

eference input to enhance EDR quality, the performance of our al-

orithm might be compromised when the RRI series contains inad-

quate amount of oscillations caused by RSA. Therefore, an index

eflecting the amount of RSA oscillations in RRI series or an index

ndicating the EDR signal quality should be developed to super-

ise the performance of the algorithm during applications. Finally,

J. Zheng et al. / Medical Engineering and Physics 38 (2016) 671–678 677

Fig. 5. Comparison of CPC estimation from different type of EDRs and from RIP signal, with CPC values assessed from RIP signal as gold reference. (a) Difference in CPC

estimation between RIP and EDR_area; (b) Difference in CPC estimation between RIP and EDR_RS; (c) Difference in CPC estimation between RIP and EDR_area _filt; (d)

Difference in CPC estimation between RIP and EDR_RS_filt.

f

r

v

5

h

e

p

a

p

a

a

i

C

E

p

t

J

A

F

F

e

o

2

R

uture work need to conduct to investigate the applications of the

obust ECG-based CPC analysis in the evaluation of autonomic ner-

ous system activity.

. Conclusion

In conclusion, CPC estimated from conventional EDR series ex-

ibits constant and proportional biases, while that estimated from

nhanced EDR series is more reliable. Adaptive filtering can im-

rove the accuracy of the ECG-based CPC estimation significantly,

chieving reliable and robust result. The methods developed in this

aper may lead to improved performance of the ECG-based CPC

nalysis for cardiorespiratory interaction and autonomic function

nalysis, providing additional prognostic information in sleep qual-

ty measurement and sleep apnea detection.

onflict of interest

The authors declare that they have no conflict of interest.

thical approval

We collected a paced breathing dataset according to a stepwise

aced breathing protocol, which was reviewed and approved by

he local ethics committee of Chinese PLA General Hospital. The

udgement’s reference number is IRB-BME-028.

cknowledgments

This project was supported by the National Natural Science

oundation of China (grant no. 61471398 ), Beijing Natural Science

oundation (grant nos. 3102028 , 31220343 ), General Logistics Sci-

nce Foundation (grant no. CWS11C108), and National Key Technol-

gy Research and Development Program (grant nos. 2013BAI03B04,

013BAI03B05).

eferences

[1] Schulz S , Adochiei FC , Edu IR , Schroeder R , Costin H , Bar KJ , et al. Cardiovas-cular and cardiorespiratory coupling analyses: a review. Philos Trans A Math

Phys Eng Sci 2013;371:20120191 . [2] Garcia AJ , Koschnitzky JE 3rd , Dashevskiy T , Ramirez JM . Cardiorespiratory cou-

pling in health and disease. Auton Neurosci 2013;175:26–37 .

678 J. Zheng et al. / Medical Engineering and Physics 38 (2016) 671–678

[

[

[

[

[

[

[

[

[3] Voss A , Schulz S , Schroeder R , Baumert M , Caminal P . Methods derived fromnonlinear dynamics for analysing heart rate variability. Philos Trans A Math

Phys Eng Sci 2009;367:277–96 . [4] Bahraminasab A , Kenwright D , Stefanovska A , Ghasemi F , McClintock PV . Phase

coupling in the cardiorespiratory interaction. IET Syst Biol 2008;2:48–54 . [5] Heart rate variability: standards of measurement, physiological interpreta-

tion and clinical use. Task Force of the European Society of Cardiologyand the North American Society of Pacing and Electrophysiology. Circulation

1996;93(5):1043–65 .

[6] Thomas RJ , Mietus JE , Peng CK , Goldberger AL . An electrocardiogram-basedtechnique to assess cardiopulmonary coupling during sleep. Sleep

2005;28:1151–61 . [7] Harrington J , Schramm PJ , Davies CR , Lee-Chiong TL Jr . An electrocardio-

gram-based analysis evaluating sleep quality in patients with obstructive sleepapnea. Sleep Breath 2013;17:1071–8 .

[8] Porta A , Bassani T , Bari V , Tobaldini E , Takahashi AC , Catai AM , et al. Mod-

el-based assessment of baroreflex and cardiopulmonary couplings duringgraded head-up tilt. Comput Biol Med 2012;42:298–305 .

[9] O’Brien C , Heneghan C . A comparison of algorithms for estimation of arespiratory signal from the surface electrocardiogram. Comput Biol Med

2007;37:305–14 . [10] Moody GB , Mark RG , Zocola A , Mantero S . Derivation of respiratory signals

from multi-lead ECGs. In: Proceedings of IEEE conference on computers in car-

diology, vol. 12; 1985. p. 113–16 . [11] Khandoker AH , Palaniswami M . Modeling respiratory movement signals dur-

ing central and obstructive sleep apnea events using electrocardiogram. AnnBiomed Eng 2011;39:801–11 .

[12] Schramm PJ , Thomas R , Feige B , Spiegelhalder K , Riemann D . Quantitative mea-surement of sleep quality using cardiopulmonary coupling analysis: a retro-

spective comparison of individuals with and without primary insomnia. Sleep

Breath 2013;17:713–21 . [13] Wang RC , Calvert TW . A model to estimate respiration from vectorcardiogram

measurements. Ann Biomed Eng 1974;2:47–57 . [14] Dobrev D , Daskalov I . Two-electrode telemetric instrument for infant heart rate

and apnea monitoring. Med Eng Phys 1998;20:729–34 . [15] Thayer JF , Peasley C , Muth ER . Estimation of respiratory frequency from autore-

gressive spectral analysis of heart period. Biomed Sci Instrum 1996;32:93–9 .

[16] Boyle J , Bidargaddi N , Sarela A , Karunanithi M . Automatic detection of respi-ration rate from ambulatory single-lead ECG. IEEE Trans Inf Technol Biomed

2009;13:890–6 . [17] Leanderson S , Laguna P , Sornmo L . Estimation of the respiratory frequency us-

ing spatial information in the VCG. Med Eng Phys 2003;25:501–7 . [18] Varanini M , Emdin M , Allegri F , Raciti M , Conforti F , Macerata A , et al. Adap-

tive filtering of ECG signal for deriving respiratory activity. Comput Cardiol

1990:23–6 . [19] Pogach MS , Punjabi NM , Thomas N , Thomas RJ . Electrocardiogram-based sleep

spectrogram measures of sleep stability and glucose disposal in sleep disor-dered breathing. Sleep 2012;35:139–48 .

[20] de Chazal P , Heneghan C , Sheridan E , Reilly R , Nolan P , O’Malley M . Automatedprocessing of the single-lead electrocardiogram for the detection of obstructive

sleep apnoea. IEEE Trans Biomed Eng 2003;50:686–96 .

[21] Nemati S , Malhotra A , Clifford GD . Data fusion for improved respiration rateestimation. EURASIP J Adv Signal Process 2010;2010:926305 .

22] Bailon R , Sornmo L , Laguna P . A robust method for ECG-based estimationof the respiratory frequency during stress testing. IEEE Trans Biomed Eng

2006;53:1273–85 . [23] Langley P , Bowers EJ , Murray A . Principal component analysis as a tool for an-

alyzing beat-to-beat changes in ECG features: application to ECG-derived res-piration. IEEE Trans Biomed Eng 2010;57:821–9 .

[24] Widjaja D , Varon C , Dorado AC , Suykens JA , Van Huffel S . Application of ker-

nel principal component analysis for single-lead-ECG-derived respiration. IEEETrans Biomed. Eng 2012;59:1169–76 .

25] Saeed M , Villarroel M , Reisner AT , Clifford G , Lehman LW , Moody G , et al. Mul-tiparameter intelligent monitoring in intensive care II: a public-access inten-

sive care unit database. Crit Care Med 2011;39:952–60 . 26] De Meersman RE . Aging as a modulator of respiratory sinus arrhythmia. J

Gerontol 1993;48:B74–8 .

[27] Carry PY , Baconnier P , Eberhard A , Cotte P , Benchetrit G . Evaluation of respi-ratory inductive plethysmography: accuracy for analysis of respiratory wave-

forms. Chest 1997;111:910–15 . 28] Zhang Z , Zheng J , Wu H , Wang W , Wang B , Liu H . Development of a respiratory

inductive plethysmography module supporting multiple sensors for wearablesystems. Sensors (Basel) 2012;12:13167–84 .

29] Hamilton PS , Tompkins WJ . Quantitative Investigation of Qrs detection

rules using the Mit/Bih arrhythmia database. IEEE Trans Biomed Eng1986;33:1157–65 .

[30] Schafer C , Rosenblum MG , Kurths J , Abel HH . Heartbeat synchronized withventilation. Nature 1998;392:239–40 .

[31] Wu Z , Huang NE . Ensemble empirical mode decomposition: a noise-assisteddata analysis method. Adv Adapt Data Anal 2009;1:1–41 .

32] Huang NE , Shen Z , Long SR , Shen SS , Qu WD , Gloersen P , et al. The empirical

mode decomposition and the Hilbert spectrum for nonlinear and non-station-ary time series analysis. Proc R Soc London Ser A 1998;454:903–93 .

[33] Janusauskas A , Marozas V , Lukosevicius A . Ensemble empirical mode de-composition based feature enhancement of cardio signals. Med Eng Phys

2013;35:1059–69 . [34] Manor BD , Hu K , Peng CK , Lipsitz LA , Novak V . Posturo-respiratory synchro-

nization: effects of aging and stroke. Gait Posture 2012;36:254–9 .

[35] Pereda E , De la Cruz DM , De Vera L , Gonzalez JJ . Comparing generalized andphase synchronization in cardiovascular and cardiorespiratory signals. IEEE

Trans Biomed Eng 2005;52:578–83 . 36] Hu K , Peng CK , Czosnyka M , Zhao P , Novak V . Nonlinear assessment of cere-

bral autoregulation from spontaneous blood pressure and cerebral blood flowfluctuations. Cardiovasc Eng 2008;8:60–71 .

[37] Cysarz D , Bussing A . Cardiorespiratory synchronization during Zen meditation.

Eur J Appl Physiol 2005;95:88–95 . [38] Bland JM , Altman DG . Statistical methods for assessing agreement between

two methods of clinical measurement. Lancet 1986;1:307–10 . 39] Wang B , Zhang Z , Wang W . Research on the respiratory sinus arrhythmia in

the process of guided breathing. J Biomed Eng 2012;29:45–50 69 .