EMBC`2009 - 1st paper - final

Document technical information

Format pdf
Size 1.8 MB
First found May 22, 2018

Document content analysis

Category Also themed
Language
English
Type
not defined
Concepts
no text concepts found

Persons

Charles Stewart Parnell
Charles Stewart Parnell

wikipedia, lookup

Organizations

Places

Transcript

Assessing Systolic Time-Intervals from Heart Sound: a Feasibility
Study
P. Carvalho, R. P. Paiva, R. Couceiro, J. Henriques, I. Quintal, J. Muehlsteff, X. L. Aubert, M. Antunes
Abstract— Systolic time intervals are highly correlated to
fundamental cardiac functions. In this paper we investigate the
feasibility of using heart sound (HS) to accurately measure the
opening and closing moments of the aortic valve, since these are
crucial moments to define the main systolic timings of the heart
cycle, i.e. the pre-ejection period (PEP) and the left ventricular
ejection time (LVET). We introduce a HS model, which is
applied to define several features that can be applied to identify
clear markers that define these moments in the HS. Using these
features and a comparative analysis with registered
echocardiographies from 17 subjects, the results achieved in
this study suggest that HS can be used to accurately estimate
LVET and PEP.
S
I. INTRODUCTION
Robert Hooke’s discovery of the diagnostic
potential of heart sound (HS), cardiac auscultation has
been a key instrument for non-invasive and low-cost
diagnosis. In the last few decades, due to the introduction of
new and powerful diagnostic tools such as ultrasound and
Doppler imaging, as well as due to the high proficiency
required for accurate heart auscultation, heart sound
auscultation has been relegated to a minor role in daily acute
medical practice. Conversely, developments in digital signal
processing and analysis are leading to renewed interest in
heart sound. It has emerged as a powerful (easy to use, low
intrusive, repeatable and accurate) and inexpensive biosignal to develop monitoring systems, mainly in the context
of chronic disease management, where low-cost and reliable
solutions for cardiovascular function assessment are required
for long-term patient follow-up.
There are several theories regarding the origin of heart
sounds. One of the most accepted theories is the valve
theory. In opposition to the cardiohemic theory where HS is
attributed to vibration of the entire cardiohemic system
(heart cavities, blood and valves), in the valvular theory the
first heart sound (S1) is produced by the closure of the mitral
INCE
This work was supported in part by the EU FP7 project HeartCycle (FP7
– 216695).
P. Carvalho, R. Paiva, R. Couceiro and J. Henriques are with the
Department of Informatics Engineering, Science and Technology Faculty of
the University of Coimbra, Pólo II, Coimbra, Portugal (phone: +351
239790000; e-mail: {carvalho, ruipedro, rcoureir, jh}@dei.uc.pt).
I. Quintal is with the Echocardiography Department of the Centro
Hospitalar de Coimbra, Coimbra, Portugal.
J. Muehlsteff and X. Aubert are with Philips Research Laboratories
Europe, Aachen, Germany (e-mail: {Jens.Muehlsteff, Xavier.
Aubert}@philips.com).
M. Antunes is with the Cardiothoracic Surgery Center of the Hospitais
da
Universidade
de
Coimbra,
Coimbra,
Portugal
(email:
[email protected])
and the tricuspid valves, which may not coincide, whereas
the second heart sound (S2) is linked to the closure of the
aortic and pulmonary valves. In the later, the first component
is always the aortic component, which may coincide with the
pulmonary valve closure. The first heart sound (S1) is
mainly characterized by two well-defined high frequency
components. As was pointed out by Lakier et al. [1], there is
good evidence that the first component of S1 is induced by
the (eventually non-simultaneous) closure of the mitral and
the tricuspid valves, while regarding the second component,
it is suspected to be mainly related to the vibration induced
by the aortic valve opening.
The timings between HS’s main components, its
morphology as well as its spectral content can be applied to
directly estimate relevant cardiac parameters [2][12].
Currently it is observed that most computer-aided diagnostic
systems based on heart sound focus on the morphology and
spectral content analysis of its components. For instance,
Teitz et al. [3] have introduced algorithms for heart valve
dysfunction diagnosis for prosthetic heart valves. Other
successful applications of computer-aided heart sound
analysis are for the detection of heart failure using the third
HS component [4] and the assessment of the PulmonaryAortic Pressure [6]. Xiao et al. [7] proposed two indicators
of cardiac reserve using the analysis of the amplitude of the
first heart sound.
Using heart sounds as a reference for systolic and diastolic
time intervals measurement is not a new idea. It was a
common practice prior to the introduction of ultrasound. The
procedure was based on the combination of the carotid pulse
(variants exist where other markers for the systolic ejection
are applied), the HS and the ECG. By neglecting the time
distortion caused by pulse propagation, the carotid pulse
wave was analyzed to identify the aortic valve timings and
enabled the estimation of the left ventricular ejection time
(LVET). The pre-ejection period (PEP) was measured
indirectly by subtracting LVET from RS2 [5][12], where
RS2 is the time interval from the ECG R-peak to S2.
One aspect that has not been fully explored using HS is
the possibility to accurately measure the main cardiac time
intervals using this signal without resorting to the carotid
pulse. Several studies [9][10] have shown that cardiac
systolic and diastolic time intervals are highly correlated to
major and fundamental cardiac functions. Of major
relevance in assessing cardiac reserve and the left ventricular
function are PEP and LVET. By definition, PEP is the time
interval between the start of ventricular depolarization and
the moment of aortic valve opening (for detection accuracy
Fig. 1. Data registration. (top) Typical registration error profile using the
R-R ECG interval. (bottom) Registration results showing
echocardiography (second from top), superimposed ECGs (third from top)
and HS.
in this paper PEP is defined as the interval between the Rpeak of the ECG and the opening of the aortic valve),
whereas the LVET is defined as the time interval of left
ventricular ejection, which occurs between the opening of
the aortic valve and its subsequent closure. PEP is an index
of the left ventricular function and reflects changes in
myocardial contractility, left ventricular end-diastolic
volume and aortic diastolic pressure. Another important
application of PEP is in non-invasive beat-by-beat
estimation of blood pressure. A review on the prominent role
of PEP in the assessment of BP surrogates can be found in
Muehlsteff et al. [11]. The left ventricular ejection period
(LVET) can also be related to contractility and to cardiac
output [10]. It is by itself a measure of cardiac function [12].
In this paper the goal is to assess the feasibility to
accurately extract the systolic time intervals (PEP and
LVET) using HS and ECG. The underlying hypothesis is
that the first and the second heart sounds encode the
movements of the aortic valve and that these components
exhibit noticeable and specific signatures that enable their
identification using this signal. The interested reader is
referred to [15] where we introduce an algorithm to identify
PEP and LVET from HS.
In section II the experimental design of the study and the
data analysis procedure are described. In section III the main
results are presented and discussed. Finally, in section IV the
main conclusions are presented.
II. FEASIBILITY STUDY
As already mentioned, the primary goal of the study is to
evaluate the feasibility in using heart sounds to determine
the opening and the closing of the aortic valve, in order to
define the main systolic time intervals, i.e. PEP and LVET.
Heart valve movements are not instantaneous, but rather
transitory processes that have their intrinsic dynamic. Due to
this reason, PEP as well as LVET definitions do not have
precise measurement points. This may induce considerable
variations / imprecisions between different measurement
techniques in assessing many PEP and LVET related
parameters. Therefore, a secondary goal of this study is the
assessment of the accuracy in using heart sounds to measure
the start time of both processes of aortic valve movement,
i.e. opening and closing. This is particularly important, since
the uncertainty underlying PEP and LVET measurements is
also common to other measurement principles such as
impedance cardiography (ICG), one of the reference
methods for portable devices in measuring these parameters.
As was stated by Ermishkin et al. [13], compared to the PEP
and LVET values extracted using the Gold Standard method
– the echocardiography, PEP values extracted from the ICG
using a visual inspection method for B-point determination
based upon the dZ/dt is delayed by 3-20ms relative to the
onset of blood flow in the left ventricular trace.
A. Experimental Setup
17 volunteer students at the Centro Hospitalar de Coimbra
have been asked to participate in the data collection study
aimed at the simultaneous collection of heart sounds (HS)
and echocardiography (echo). A synchronous ECG with
each of the above signals was also acquired and served as a
reference signal for co-registration. The population was not
balanced for gender (14 male and 3 female). All persons
involved in this study did not have any known congenital or
other heart disease. The biometric characteristics of the
population were:
• Age: 22.53 ±3.81 years
• BMI: 23.27 ± 2.15 Kg/m2
The measurement protocol was conducted by an
authorized medical specialist and consisted of several
acquisitions of echocardiography in different modes
(Doppler and M-mode) and heart sound collection sites
(LSB – left sternum border, and apex). More precisely the
following steps were observed:
• The patient was in supine position, turned left
(approximately 45º) – the usual echo observation
position for the aortic valve.
• The echo device was configured for M-mode and the
stethoscope was positioned in the Apex region.
• Runs of 8 sec. data acquisitions of HS, Echo and ECG
were performed repeatedly.
• The echo was configured for Doppler-mode and the
stethoscope was positioned in the LSB region.
• Runs of 8 sec. data acquisitions of HS, Echo and ECG
were performed repeatedly.
The following signals have been acquired:
• Echocardiography and ECG using a Siemens Acuson
CV70 device. This device produces a DICOM output
Fig. 2. Annotation of aortic valve timings using two different modes of
echocardiography. (top) Doppler mode echocardiography. (bottom)
M-mode echocardiography.
•
with images of time resolution equivalent to 272 Hz. The
echo DICOM SOP (Service Object Pair) does not allow
for simultaneous recording of time-series and images.
Therefore, the echo ECG was recorded as part of the
image at 272 Hz. (see fig. 2)
Heart Sounds and ECG: a Meditron Stethoscope and
Analyzer were applied to record HS and ECG at 44.1
kHz. The bandwidth of the HS sensor is 20 kHz.
B. Data Analysis
As already mentioned, the data streams originating from
the HS and the Echocardiography were synchronized using
the simultaneously acquired ECG signals. The algorithm
applied for data registration using the ECG signals is based
on the analysis of the ECG’s R-R interval matching least
square error minimization. Let R1(k), k=1,…,n, and R2(w),
w=1,…,m (n<m) be the R-R intervals of the ECGs to be
registered. The registration instant t is obtained from:
 n

2
1
t = arg min 
R1( k ) − R2 ( w + k )) 
(1)
(

w=1,...,m−n 
 n k=1
A typical fitting error profile and HS-echo registration result
is shown on top of fig. 1.
The data provided by the Echo device required some post€
processing in order to extract the ECG from the DICOM
image and to compose a linear image from the circular 3sec.
image buffer shown on-screen and saved by the device to
DICOM using M-JPEG type video coding.
The annotations of the opening and closing instants of the
aortic valve using the echocardiographies were performed
under the supervision of an experienced clinical expert in
echocardiography. In the M-mode echocardiographies it is
observed that the closing of the aortic valve is not always
identifiable. In these situations only the opening of the aortic
valve was annotated. For this mode, the opening of the aortic
valve was annotated at the start of the dynamic opening
process, i.e. when the aortic valve leaflets start the opening
movement (see fig. 2). Regarding the Doppler mode
echocardiographies, the opening instant of the aortic valve
was annotated as the onset of the ejection lobe of the left
∑
Fig. 3. Features extracted and applied to visually annotate the onset of
the opening of the aortic valve (all values are normalized; derivatives
are clipped to negative values).
ventricle, while the closing point was defined immediately
before the onset of the closing click produced by the residual
reflux after the aortic valve cusps have closed, as can be
observed in fig. 2. It should be noted that this click
corresponds to the end of the dynamic process related to the
closing of the valve cusps.
The annotations of the heart sounds were carried out by
one of the team members. This has been performed without
echo reference, i.e. only the HS and ECG signals as well as
features extracted from the HS were shown during
annotation. Regarding S1 annotation, it is assumed that it is
mainly characterized by two well-defined high frequency
components: the working hypothesis in this study is that the
first of these components corresponds to the closing of the
mitral and (eventually non-simultaneous) the tricuspid
valves; the second one is induced by the opening of the
aortic valve. Both components are assumed to follow a
signal model described by amplitude modulated chirp
signals. It is also assumed that both components only
marginally overlap in time. From the above, we assume that
the first heart sound is modeled by eq. (2), where Ai(t) is the
non-linear time dependant gain and ϕi(t) represents the time
dependent instantaneous frequency of the signal component.
2
s( t ) =
∑ A (t) sin(ϕ (t)),
i
i=1
€
i
Ai ( t ) ≥ 0
(2)
TABLE I
PEP IDENTIFICATION RESULTS
*
Signal
Average±SD
Range
Echo
HS
Error*
54.04±18.22
52.82±17.13
5.81±4.91
[22.06, 109.97]
[27.77, 106.58]
-
|PEPHS-PEPEcho|.
TABLE II
LVET IDENTIFICATION RESULTS
Signal
Average±SD
Range
Echo
HS
Error**
266.01±27.67
255.13±25.41
14.76±10.94
[180.54, 328.95]
[176.85, 326.05]
-
**
|LVETHS-LVETEcho|.
TABLE III
AVC IDENTIFICATION RESULTS
Signal
Echo
HS
Error§
§
Average±SD
Range
320.05±26.46
307.95±22.84
-15.48±11.06
[222.63, 386.76]
[250.49, 362.98]
-
AVCHS-AVCEcho, where AVC stands for Aortic Valve Closure
Furthermore, it is assumed that Ai(t) is a smooth function,
i.e. with much lower frequency content compared to ϕi(t).
Since on the one hand the instantaneous frequency of the
vibration induced during valve membrane movement is a
function of the pressure across the valves [14] and, on the
other hand, the pressure differences that govern the
movements of the mitral and the aortic valves tend to be
significantly different, given the marginal overlap of both
signal components, a noticeable variation in (i)
instantaneous frequency and (ii) energy is expected in the
signal. These are the main features applied for the detection
of the opening of the aortic valve in S1. The instantaneous
energy and frequency are computed using the Wigner-Ville
transform and the non-linear gain is obtained using a
combined homomorphic and low-pass filtering approach. In
fig. 3 these features are shown for one S1 sound. As can be
observed there are significant changes in the instantaneous
frequency and power of the signal that enable the detection
of the onset of the aortic valve opening. An algorithm based
on some of these features is introduced in [15]. To detect the
closing of the aortic valve, the high frequency signature
developed in [8] is applied as a reference.
III. RESULTS
The main results obtained in this study are summarized in
tables I, II and III. The achieved results suggest that it is
possible to accurately identify the systolic time intervals
using HS. As can be observed in fig. 4, both PEP as well as
LVET obtained from HS closely follow with high
correlation the beat-by-beat values identified for these
intervals using the echocardiography. Namely, it is observed
that the correlation between the PEPs measured using
Fig. 4. Observed PEP and LVET values using heart sound and
echocardiography for 17 subjects. (top) beat-by-beat values. (middle)
PEP estimation difference dispersion as a function of the beat-by-beat
(PEPHS+PEPEcho)/2. (bottom) LVET error dispersion as a function of the
beat-by-beat (LVETHS+LVETEcho)/2.
echocardiography and HS is 0.91 (p<0.0001). On average, it
is seen that using HS the opening of the aortic valve is
detected before (see fig. 4 middle) it is observed in the
echocardiography. The average bias is of the same order as
the time resolution of the echo (≈3.7ms). It should be noticed
that in both measurement scenarios the observed PEPs are
lower than those that are usually reported in literature for
similar populations. This is linked to the fact that PEP is
usually defined as the time interval between the Q-peak of
the ECG to the onset of the opening of the aortic valve,
while in this paper we have used the R-peak as the reference
ECG point.
Regarding the measured LVET beat-by-beat series
extracted from the echocardiographies and HS, it is observed
that the correlation coefficient is 0.84 (p<0.0001), which is
slightly lower compared to PEP. On the other hand, the
average delay observed is significantly higher than the
echo’s time resolution. Namely, it is observed that on
average the onset of the aortic valve cusps closing
movement is detected 15.48ms earlier compared to
echocardiography (see table III). This can be attributed to
the fact that HS enables the detection of the onset of the
aortic valve closing process, while echocardiography enables
its detection near the closing click induced by the cusps, i.e.
at the end of the dynamic process.
As already mentioned, there is ample evidence that ICG
does not enable a precise detection of the onset of the aortic
valve opening and closing process [13]. During the opening
dynamics, the preferred measurement point is the B-point,
which, according to Ermishkin [13], introduces a delay of 320ms relative to the onset of blood flow. The B-point might
not be the only factor contributing to the delay. The
sensitivity of the ICG to changing impedance induced by the
blood flow during the left ventricle ejection might itself be
the origin of the delay.
of Dr. Leitão Marques from CHC in facilitating the
arrangements for the data acquisition part of the study is also
acknowledged.
IV. CONCLUSION
[5] D. B. Newlin and R. Levenson, Pre-ejection period: Measuring betaadrenergic influences upon the heart, Psychophysiology, 1979, vol. 16,
pp. 546-553.
We have investigated the possibility of using heart sounds
to accurately measure the main systolic heart time intervals,
i.e. the pre-ejection period and the left ventricle ejection
time. The working hypothesis was that heart sounds encode
clear makers that enable the detection of the opening and the
closing of the aortic valve. To evaluate this hypothesis we
have conducted a comparative echocardiography-heart
sound study on 17 healthy subjects. A heart sound model for
the first heart sound was introduced and several features
were defined in order to enable the definition of clear
markers in the heart sound signal to identify the aortic valve
components of the S1 and S2 (the reader is referred to [15]
for an algorithm that uses some of these features to
automatically measure the systolic time intervals from heart
sounds). The achieved results strongly support the view that
heart sound can be applied to detect the onset of the aortic
valve movement processes. This seems to be a significant
result, since other competing approaches for LVET and PEP
measurement (e.g. the ICG approach) tend to exhibit biases
in determine these moment estimations, leading to possible
inaccuracies in cardiac function assessment.
It should be noted that the dataset was collected from a
population that is not the typical population with coronary
heart diseases (both due to the average age as well as BMI).
Hence, it is acknowledged that the study should be pursued
with new measurements of patients from the typical target
population for heart disease management. This will also
enable to extend the available database of HS and
Echocardiography with synchronized ICG signals in order to
compare more accurately the two measurement principles.
ACKNOWLEDGMENT
The authors want to express their gratitude to the 17 students
from the Centro Hospitalar de Coimbra (CHC) who
volunteered for this study. The authors would also like to
recognize and to express their appreciation to the Centro
Hospitalar de Coimbra for supporting the study. The effort
REFERENCES
[1] J. Lakier, V. Fritz, A. Pocock, J. Barlow, Mitral components of the first
heart sound, British Heart Journal, 1972, vol. 34, pp.160-166.
[2] C. Alhstrom, Nonlinear Phonocardiographic Signal Processing,
Dissertation No 1168, Linkoping University, 2008.
[3] T. Eitz, D. Fritzsche, O. Grimmig, I. Frerichs, A. Frerichs, G. Hellige,
K. Minami, R. Körfer, Acoustic Phenomena and Valve Dysfunction in
Cardiac Prostheses: Data Acquisition and Collection via the Internet,
The Journal of Heart Valve Disease, 2003, vol. 12, pp. 414-419.
[4] D. Kumar, P. Carvalho, M. Antunes, J. Henriques, R. Schmidt, J.
Habetha, Third Heart Sound Detection Using Wavelet Transform–
Simplicity Filter, Int. Conf. of the IEEE Engineering in Medicine and
Biology Society, 2007.
[6] V. Nigam, R. Priemer, A dynamic method to estimate the time split
between the A2 and the P2 components of the S2 heart sound,
Physiological Measurement, 2006, vol. 27, pp. 553-567.
[7] S. Xiao, Z. Wang, D. Hu, Studying Cardiac Contractility change trend
to evaluate cardiac reserve, IEEE Engineering in Medicine and
Biology, 2002, pp. 74-76.
[8] D. Kumar, P. Carvalho, M. Antunes, J. Henriques, L. Eugénio, R.
Schmidt, J. Habetha, Detection of S1 and S2 Heart Sounds by High
Frequency Signatures, Int. Conf. of the IEEE Engineering in Medicine
and Biology Society, 2006.
[9] D. Cokkinos, E. Heimonas, J. Demopoulos, A. Haralambakis, G.
Tsartsalis, C. Gardikas, Influence of heart rate increase on uncorrected
pre-ejection period/left ventricular ejection time (PEP/LVET) ratio in
normal individuals, British Heart Journal, 1976, vol. 38, pp. 683-688.
[10] S. Finckelstein, J. Cohn, Method and apparatus for measuring cardiac
output, US patent 5241966, 1993.
[11] J. Muehlsteff, X. Aubert, M. Schuett, Cuffless estimation of systolic
blood pressure for short effort bicycle tests: the prominent role of the
pre-ejection period, Int. Conf. of the IEEE Engineering in Medicine and
Biology Society, 2006.
[12] M. Tavel, Clinical Phonocardiography and external pulse recording,
Year Book Medical Publishers, Inc.
[13] V. Ermishkin, E. Lukoshkova, E. Bersenev, M. Saidova, V. Shitov, O.
Vinogradova, V. Khayutin, Beat-by-beat changes in pre-ejection period
during functional tests evaluated by impedance aortography: a step to a
left ventricular contractility monitoring, Int. Conf. on Electrical
Bioimpedance and the 8th Conference on Electrical Impedance
Tomography, 2007, pp. 655-658.
[14] K. Lim, Y.-C. Liew, C.-H. Oh, Analysis of mitral and aortic valve
vibrations and their role in the production of the first and second heart
sound, Phys. Med. Biol.., 1980, vol. 25, no. 4, pp. 727-733.
[15] R. P. Paiva, P. Carvalho, J. Henriques, M. Antunes, J. Muehlsteff, X.
Aubert, Assessing PEP and LVET from Heart Sounds: Algorithms and
Evaluation, Int. Conf. of the IEEE Engineering in Medicine and
Biology Society, 2009 (submitted).
[16] M. Wong, Y. Zhang, The effects of pre-ejection period on the blood
pressure estimation using pulse transit-time, 5th Int. Workshop on
Wearable and Implantable Body Sensor networks, 2008.
×

Report this document