Analysis of Heart Rate Variability Using Time-Varying
Filtering of Heart Transplanted Patients
Ghailen Laouini, Olivier Meste, Marianna Meo
To cite this version:
Ghailen Laouini, Olivier Meste, Marianna Meo. Analysis of Heart Rate Variability Using TimeVarying Filtering of Heart Transplanted Patients. International Conference of the IEEE Engineering
in Medicine and Biology Society, Aug 2012, United States. hal-00738441
HAL Id: hal-00738441
https://hal.science/hal-00738441
Submitted on 4 Oct 2012
HAL is a multi-disciplinary open access
archive for the deposit and dissemination of scientific research documents, whether they are published or not. The documents may come from
teaching and research institutions in France or
abroad, or from public or private research centers.
L’archive ouverte pluridisciplinaire HAL, est
destinée au dépôt et à la diffusion de documents
scientifiques de niveau recherche, publiés ou non,
émanant des établissements d’enseignement et de
recherche français ou étrangers, des laboratoires
publics ou privés.
34th Annual International Conference of the IEEE EMBS
San Diego, California USA, 28 August - 1 September, 2012
Analysis of Heart Rate Variability
Using Time-Varying Filtering of Heart Transplanted Patients
Ghailen Laouini, Olivier Meste and Marianna Meo, member IEEE
Abstract— In this paper, we analyze the heart rate variability
(HRV), obtained by using the time-varying integral pulse
frequency modulation (TVIPFM) which is well adapted to the
exercise stress testing.
We consider that the mean heart period is varying function of
time, during exercise. This technique allows the estimation of
the autonomic nervous system modulation (ANS) from the beat
occurrences. The estimated respiratory sinus arrhythmia is then
filtered in the time-frequency domain around the respiration
using a time-varying filter. It is proven that the Spectrogram
is a convenient time-frequency representation that allows the
implementation of such filter. The recorded data comes from
exercise test performed by ten heart transplant patients. The
magnitude of the filtered modulation of the heart rate due to
respiration is compared to the date of transplantation taking
into account the volume of respiration. It reveals that the
normalized magnitude of the filtered variability, is significantly
increased as the age of transplantation is higher with a high
correlation coefficient (R=0.74, p=0.01). This correlation raised
to 0.82 when considering dynamic behavior of the parameters.
Applied to our dataset, standard parameter fails to exhibit
such correlation.
I. INTRODUCTION
Several works [1] have used the Integral pulse frequency
modulation (IPFM) model to explain the regulation of the
heart rate by the ANS, in different physiological situations,
even during stress testing exercise [9]. However the IPFM
model assumes a constant threshold [2,3] considering a
constant mean heart period. During dynamic protocol the
use of the time-varying threshold IPFM (TVIPFM) model to
correct the heart rate variability is relevant for the estimation
of the modulating signal [4,10].
Many authors have emphasized the close relationship between respiration and HRV [6,7,8]. This component appears
to be the principal non stationary modulation of high frequency. Since it is embedded in a mixture of modulating
signals, its analysis requires a filtering that is time-varying in
that case. The performance of stress test exercises allows the
use of the respiratory sinus arrhythmia (RSA) information for
the ANS tone assessment. This modulation is effective with
respects to the innervation of the heart. After orthotopic heart
transplantation the loss of the innervation should suppress
this modulation. During long-term follow-up it has been
shown [11] that the HRV is higher as long is the age of
the transplantation, with a strong correlation. In the study
[11] the variability measurement is addressed by using the
standard deviation of the data during resting conditions.
G. Laouini is with Laboratory I3S, University of Nice and CNRS, Sophia
Antipolis 06903, France, (e-mail:laouini@i3s.unice.fr)
O. Meste is with Laboratory I3S, University of Nice and CNRS, Sophia
Antipolis 06903, France, (e-mail:meste@i3s.unice.fr)
978-1-4577-1787-1/12/$26.00 ©2012 IEEE
Thus a correlation between this parameter and the age of
the transplantation is exhibited under controlled protocol,
limiting the impact of such analysis. It would be of greater
interest to show this correlation with relaxed and dynamic
conditions corresponding to increasing physical exercise.
The paper is organized as follows. The methods and
materials are presented in Section II, where we use a mathematical development to demonstrate the feasibility of filtering
in the time-frequency domain, using the Spectrogram. The
TVIPFM model is described as well in this section to obtain
a corrected HRV signal. Section III presents the results and
Section IV the discussion.
II. METHODS
A. Mathematical development
In this study the signal to be filtered is non stationary.
It means that the characteristics of the bandpass filter must
vary with time. The time-frequency transformation that will
permit this particular filtering is the Spectrogram defined by:
Z
S(t, f ) = | x(s)h∗ (s − t)e−i2πf s ds|2
where h(t) stands for the smoothing window. In the following it will be demonstrated that by using this representation
in addition with a matched template G(t, f ), an efficient
time-varying filter can be applied. Note that continuous time
is considered in the entire development but results are still
valid with sampled time. The spectrogram being the squared
magnitude of the short Fourier Transform, it is linked to the
Wigner-Ville (WV) distribution by the convolution products
in time and frequency:
S(t, f ) = (Wh ∗ ∗Wx )(t, f )
Assuming that the WV transform of h(t) is separable such
that Wh (t, f ) = g(t)l(f ) (verified when h(t) is a gaussian
function), when x(t) is a linear frequency modulation, i.e.
x(t) = exp(iπαt2 ), the Spectrogram is then:
Z Z
S(t, f ) =
g(t − s)l(f − ν)δ(s − αν)dsdν
S(t, f ) =
Z
g(t − αν)l(f − ν)dν
Note that the signal x(t) is locally approximated by this
model whose modulation slope is α. This assumption is
valid with respect to the respiration component where even
during exercise the frequency varies smoothly. This quantity
3436
is nonzero when the two windows g and l are simultaneously
nonzero.
g(t) 6= 0 f or − T < t < T
(1)
l(f ) 6= 0 f or − F < f < F
where g(t) and l(f ) are assumed hermitian symmetric.Then:
g(t − αν) 6= 0 f or −Tα+t < ν < +Tα+t
(2)
l(f − ν) 6= 0 f or − F + ν < f < F + ν
It can be concluded that for a given time t0
t0 = 0, the beat occurrence time series can be generated as
solution of [5]:
Z tk
1 + m(t)
dt
k=
T (t)
0
with its extension to continuous case:
Z t
1 + m(s)
ds
k(t) =
T (s)
0
the instantaneous heart rate is defined as below
−T + t0
+T + t0
−F +
<ν<F+
α
α
dHR (t) =
So [−F + −Tα+t0 ; F + +Tα+t0 ] is the support of S(t0 , f )
along the frequency axis. The mean value of this interval is
t0
α that corresponds to the instantaneous frequency of x(t)
for t = t0 .As shown in the sequel, because this result means
that the support is centered on the instantaneous frequency
of x(t), the integration around this interval will provide the
magnitude without bias. This allows us to impose to the
template G(t, f ), that will be multiplied with S(t, f ), to be
nonzero around f = tα0 for t = t0 in order to get rid off the
other components. It is well known that for the Spectrogram,
the marginal is:
Z
S(t, f )df = (px ∗ ph )(t)
with px (t) being the squared envelop |x(t)|2 of x(t) . If
this envelop is locally a constant A, then
R the computation
of this integration will produce A2 if |h(t)|2 = 1. This
result imposes to the smoothing window h(t) to be energy
normalized. The slow variation of the respiration component
in the heart period perfectly match the local assumption.
This demonstration proves the role of the time-varying
filtering around the respiration to retrieve the envelope of
the respiration component observed within the HRV signal.
This modulation will be corrected or not by the TVIPFM
model.
The template G(t, f ) that conveys the knowledge of the
respiration frequency is defined as:
G(t, f ) = 1 if f ∈ BR ∀t
(3)
G(t, f ) = 0 otherwise
With BR is the band frequency of respiration. Details on
the computation of the time-varying filter using G(t, f ) is
fully provided in [7].
The TVIPFM model is based on the hypothesis that the
ANS influence on the sinus atrial SA node can be represented
by the modulating signal m(t).
In the TVIPFM model the integral of 1+m(t) is compared to
a time-varying threshold T (t), representing the time-varying
mean heart period. Assuming that the first beat occurs at time
1
T (t)
Supposing the variations of the term
are slower than
m(t)
T ,
and their spectral components don’t
those of the term
overlap, we define the time-varying heart rate as:
dHRM (t) =
1
T (t)
and the heart rate variability signal is defined as
dHRV (t) = dHR (t) − dHRM (t) =
m(t)
T (t)
So, to obtain the modulating signal, we have to correct the
HRV signal dHRV (t) by the time-varying mean heart rate
dHRM (t).
In the TVIPFM, the variations of the time-varying threshold
TAC (t), are small compared to its mean value TDC . Rewriting the instantaneous heart rate
dHR (t) =
1 + m(t)
1 + m(t)
1 + m(t)
=
=
(t)
T (t)
TDC + TAC (t)
TDC (1 + TTAC
)
DC
≈
TAC (t)
1 + m(t)
(1 +
)
TDC
TDC
with the assumption m(t) < 1, we can neglect the second
order term, then we obtain:
dHR (t) =
1 + m(t) −
TAC (t)
TDC
TDC
we assume
dHRM (t) =
B. The time-varying threshold Integral pulse frequency
modulation model
1 + m(t)
dk(t)
=
T (t)
dt
1−
TAC (t)
TDC
TDC
=
1
TDC
To obtain the modulating signal m(t), first we estimate
dHR (t), where k(t) is estimated from pairs (tk , k), then
we estimate the dHRM (t) term by low-pass filtering dHR (t)
and we calculate the dHRV (t) term. Finally, the modulating
signal is obtained as:
3437
m(t) =
dHRV (t)
dHRM (t)
0.02
0.015
0.01
magnitude
Recorded signals have been obtained from surface electrodes that produce interferences such as noise and baseline
wander. So, to exploit all signals we must cancel the disturbances caused by breathing. First we filter the signal using
500-th order high-pass finite impulse response filter in order
to remove the baseline generated by the respiration. After
using a threshold technique we demodulate the filtered signal
to detect all the time occurrences of R-waves, namely the tk .
The differences of consecutive tk provide the heart period.
Finally, the HRV is computed by removing the trend of the
heart period. In addition to the ECG, the respiration signal is
recorded to extract the frequency and the volume. Ten heart
transplant subjects with different ages of transplantation [4160] months are analyzed by using different parameters. The
parameter introduced in [11] is indeed a standard deviation
(stand) evaluated on the HRV at rest. It will be compared
to the quantity mag(t) corresponding to the magnitude of
the time-varying filtered respiration component extracted
from the HRV. When corrected by the TVIPFM model the
subscript will be tv. Because the respiration volume has
not been controlled during the protocol, the time-varying
quantities will be normalized by the respiration volume
computed from the respiration signal. This normalization
denoted by the subscript resp will be exclusively applied to
mag because it contains only the respiration influence. All
these quantities are computed at rest and at the maximum
of the exercise. An interval of 100 seconds will be used
to compute parameter stand and to average the quantities
mag(t). In summary, the analysis will provide the parameters
stand, mag, magtv and magtv,resp .
is an abrupt increase which gradually decreases as the heart
rate decreases reaching values close to those observed at the
beginning of the test.
0.005
0
−0.005
−0.01
−0.015
0
200
400
600
800
1000
time−seconds
1200
1400
1600
0
200
400
600
800
1000
time−seconds
1200
1400
1600
0
200
400
600
800
1000
time−seconds
1200
1400
1600
0.018
0.016
0.014
0.012
magnitude
III. M ATERIALS AND RESULTS
0.01
0.008
0.006
0.004
0.002
0
100
90
80
normalized volume
0.5
0.45
normalized frequency
0.4
0.35
70
60
50
40
30
0.3
20
0.25
10
0.2
0
0.15
0.1
Fig. 2.
The corrected variability with the TVIPFM model (top), the
magnitude of the envelop filtered around the respiration frequency (middle)
and the breathing volume (bottom), the red vertical lines correspond to
the begining and the maximum of the exercise, respectively. Note that the
magnitude unit is meaningless because of the correction procedure, the
volume has been normalized by the greater value within all the subjects
0.05
0
500
1000
1500
k
2000
2500
Fig. 1. The spectrogram of the heart rate variability with the template
G(t, f ) boundaries (in black) defined by the respiratory frequency band
Figure (1), reveals that the RSA modulation is high at the
beginning of exercise, then the magnitude of this modulation
decreases abruptly after the beginning of exercise due to
the exercise pressor reflex causing ventilatory responses to
exercise. Then period from second 1400 to 1800 corresponds
to the increasing exercise. The intensity of the RSA is at
lower values during this interval. During the recovery, there
Figure (2) shows first the HRV corrected with the TVIPFM
model (upper). This variability is then filtered in the timefrequency domain, producing an estimation of the RSA
magnitude (middle). If needed, the volume of the respiration
is computed from the respiration signal (bottom). From this
function of time (middle) an average over 100 seconds is
computed in the resting period and at the maximum of the
exercise.
3438
TABLE I
T HE CORRELATION OF PARAMETERS WITH THE AGE OF TRANSPLANTATION
stand
mag
magtv
magtv,resp
(rest)
R=0.25, p=0.48
R=0.43, p=0.22
R=0.61, p=0.06
R=0.74, p=0.01
(max)
R=0.27 , p=0.45
R=-0.21, p=0.56
R=-0.14, p=0.69
R=0.09 , p=0.80
−4
6
x 10
Arbitrary Unit
5
4
3
2
1
0
0
20
40
60
80
months
100
120
140
160
Fig. 3. At rest. The magnitude of the heart rate variability corrected with
TVIPFM model filtered around the frequency band of respiration and normalized with the respiratory volume, function of the age of transplantations
(rest)-(max)
R=0.11 , p=0.76
R=0.29 , p=0.40
R=0.67 , p=0.03
R=0.82, p=0.003
and the instantaneous power is then calculated by integrating
over frequencies to obtain the magtv .
Probably because no respiration profiles have been imposed in the protocol, the normalization with the respiration
volume has increased the correlation. In contrast to standard
parameter, namely the standard deviation stand, the proposed one corroborates the physiologic expectation on our
data set. The reinnervation should be higher as the age of
the transplantation is longer, inducing a higher variability.
Note that the global variability addressed by the parameter
stand didn’t convey any positive result. The high correlation
between the RSA and the age of transplantation is due to
the TVIPFM correction and the time-varying filtering. This
significant correlation is revealed at rest and not during
exercise. The dynamic behavior corresponding to (rest)(max) is even more correlated. This result is of great interest
because it highlights the gain of the HRV dynamics due to
the innervation process.
R EFERENCES
Table I, displays all the parameters extracted from the
ten patients. Its clear that magtv,resp outperforms the other
parameters whereas the stand parameter fails in exhibiting
a linear correlation. The difference (rest)-(max) is provided
in order to investigate a dynamic behavior. Once again,
magtv,resp shows a clear correlation. Figure (3), represents
the magnitude of the heart rate variability magtv,resp (rest),
filtered around the frequency band of respiration and normalized with the respiratory volume, function of the age
of transplantation. It shows the high correlation (R=0.74,
P=0.01), between the age of the transplantation and the RSA
with TVIPFM correction and respiration normalization.
One of the important results in this paper is that from the
parameters obtained from the HRV, only the variability corrected by the TVIPFM and normalized is highly correlated
with the age of the transplantation. The result was expected
from the literature but with a parameter that fails in that
case (stand). However, since a higher variability is related
to a reinnervation, our result is in line with the physiology
knowledge.
IV. D ISCUSSION
In this paper, a method of time-varying filtering is proposed, centering the filter around the respiration frequency
band. The relevance of using the Spectrogram for this aim is
proven. In combination with this filter we use the TVIPFM
model which has a time-varying mean heart period adapted
to non stationary conditions. The time-varying mean heart
rate is estimated by low-pass filtering dHR (t). Thus, the
modulating signal is filtered in the time-frequency domain
[1] E. Bayly, Spectral analysis of pulse frequency modulation in the
nervous system. IEEE Trans Biom., vol. BME-15, pp. 257-265, 1968.
[2] J. Mateo and P. Laguna, Improved hart rate variability time-domain
signal construction from the beat occurence time according to IPFM
model. IEEE Trans Biom., vol. 47, pp. 985-996, 2000.
[3] J. Mateo and P. Laguna, Analysis of heart rate variability in the
presence of ectopic beats using the heart timing signal. IEEE Trans
Biom., vol. 50, no. 3, pp. 334-343, 2003.
[4] R. Bailon, G. Laouini, M. Orini, P. Laguna, O. Meste, The integral
pulse frequency modulation model with time-varying threshold: application to heart rate variability analysis during exercise stress testing.
IEEE Trans Biom., vol. 58, no. 3, pp. 642-652, 2011.
[5] O. Rompleman, J. Snijders and C. van Spronsen, The measurement
of heart rate variability spectra with the help of a personal computer.
IEEE Trans Biom., vol. 29, no. 7, pp. 503-510, 1982.
[6] O. Meste, G. Blain and S. Bermon. Some considerations on the IPFM
Model for the heart rate variability analysis.Computers in Cardiology.,
2003, p. 709-712, 2003.
[7] O. Meste, B. Khaddoumi, G. Blain and S. Bermon. Time-varying
analysis methods and models for the respiratory and cardiac system
coupling in graded exercise. IEEE Transactions on Biomedical Engineering., vol. 52, no. 11, pp. 1921-1930, 2005.
[8] G. Blain G, O. Meste and S. Bermon. Assessment of ventilatory
thresholds during graded and maximal exercise test using time-varying
analysis of respiratory sinus arrythmia. Br. J. Sports Med., vol 39, no.
7, pp. 448-452, 2005.
[9] R. Bailon G, L. Mainardi, M. Orini, L. Sornmo, and P. Laguna.
Analysis of heart rate variability during exercise stress testing using
respiratory information, Biomed Signal Process Control., vol. 5, no.
4, pp. 299-310, 2010.
[10] G. Laouini, A. Cabasson, G. Blain, P. Bonizzi, O. Meste, and S.
Bermon, Evidence of the influence of respiration on the heart rate variability after human heart transplantation: Role of observation model, in
Proc. Comput. Cardiol., vol. 36, pp. 409-412, 2009. http://cinc.mit.edu,
2009
[11] A. F. Folino, G. Buja, M. Miorelli, U. Livi, A. Nava, G. Thiene,
S. Dalla Volta, Heart rate variability in patients with orthotopic heart
transplantation: Long-term follow-up., Clinical cardiology, vol. 16, no.
7, pp. 539-542, 1993
3439