WindkesselPPG

WindkesselPPG

Non-invasive Estimation of Cardiac
Output from Finger Photoplethysmogram
Based on Windkessel Model
Carmen C. Y. Poon

Windkessel
18
18PPGPhysioflow2
Windkessel
Physioflow0.12
1.07 L/min

PPG
ABSTRACTCardiac output (CO) is an important index for the diagnosis and treatment of cardiovascular disease,
which is the major cause of death in many countries. Pulse contains abundant cardiovascular information and plays an
important role in cardiovascular system valuation. This study attempts to estimate CO non-invasively both in bedside
monitoring and ambulatory situation from photoplethysmgram (PPG), which can be simply obtained by an inexpensive
photo sensor. In the experiment, finger PPG was captured continuously for 18 minutes while the subjects were asked
to sit up, lie down and perform cycling exercise. CO measured by impedance cardiography was used as the reference.
Blood pressure was measured every two minutes by a mercury sphygmomanometer. Based on the Windkessel model,
an equation was developed to estimate CO from PPG after calibrated the arterial compliance with the BP change. The
results of the study found that the proposed method can estimate CO within 0.121.07 L/min of the reference.

KEYWORDS Cardiac Output; Photoplethysmogram; Windkessel Model; Arterial Compliance

1 Introduction

C ardiovascular disease has been the leading cause of

mortality in most countries nowadays [1]. Cardiac
output (CO) defined as the volume of blood being
the direct Ficks technique, thermodilution, partial CO2
rebreathing, Doppler ultrasound, magnetic resonance
imaging, and impedance cardiography. Most of these
pumped into the aorta by a ventricle per minute [2], is methods are costly, invasive and they are mainly used in
an important index to evaluate heart function and blood fixable situation, such as bedside monitoring. In contrast,
circulation, and it plays an important role in disease a number of recent studies have focus on methods that
diagnosis, cardiovascular state or therapeutic effect estimate CO non-invasively, continuously, economically
evaluation. For instance,research indicates that CO and simply, for example, some group calculate CO by
monitoring is very significant for heart failure patient and analyzing the contour of blood pressure [5] or blood
cardiopulmonary exercise test [3]. CO can be calculated volume [6]. In this study, we intend to study a new
as the product of heart rate (HR) and stroke volume (SV). technique that estimate CO from photoplethysmogram
The average CO is approximately 5L/min when at rest, (PPG) [7], which captures the blood volume pulse non-
but it may differ from subject to subject. As reviewed invasively by an inexpensive photo sensor, and has been
in [4], conventional methods on monitoring CO include applied widely in the measurement of oxygen saturation,

 Vol. 4 No.6/ Jun. 2010

blood pressure, cardiac output and arterial stiffness. interval of the k-th beat, tk+1 is the onset of the next beat,
and C and R are the arterial compliance and resistance in
2 Background that beat.
It is assumed that blood is mainly ejected into the arterial
T he proposed method was based on the Windkessel
model, a basic and classic lumped model that has
been widely used in research of the cardiovascular
system during systole and the volume of blood ejected is
nearly zero during diastole.
system. For example, some groups use it to estimate
CO from the waveform of blood pressure [8, 9]. In this
(3)
study, a two-element Windkessel model which described
the cardiovascular system in terms of compliance and
resistance was adopted to compute CO from PPG
signal. Fig.1 presents the analogy circuit of two-element Based on this assumption, we integrate the right hand side
Windkessel model, where I(t) is the instantaneous blood of the lower equation of (3) and find that blood pressure
volume which is pumped into the arterial system, see waveform is a pure decaying exponential during diastole,
(1) [10], P(t) represents the blood pressure measured from and the decay time constant, , which is the product of R
the peripheral circulation, R and C represent the total and C, and M can be calculated with . Where
peripheral resistance (TPR) and the compliance of large Sd is the diastole area of the blood pressure waveform,
arteries (AC) respectively. Fig.2 shows the simulation Ps and Pd are the systolic and diastolic blood pressures
waveform of input flow and output pressure of the two- respectively. On the other hand, the blood volume of
element Windkessel model. single beat that is SV can be obtained from (3) during
systole as follows:

(4)
(1)
where Ts is the systole interval, Ss is the systole area of
the blood pressure waveform.

3 Methodology

A. Calibration
Based on the two-element Windkessel model and
the analysis of contour of PPG, we can find that
the waveform of blood pressure is similar to the
Fig.1 Analogy circuit for the two-element Windkessel model
photoplethysmogram which represents the blood volume
information. Therefore, in this study, we estimated the
area ratio Ss/Sd in (4) by As/Ad , where As and Ad are the
systole and diastole area of blood volume waveform,
i.e. PPG. We used blood pressure measured by the
auscultative method for P s and P d. We can therefore
estimate CO from PPG using the following equation:

(5)

Fig. 2 the waveforms of blood low and pressure of model In order to estimation CO by (5), we proposed a novel
simulation method to calibrate the arterial compliance C in this
One can obtain the following model equation of a single paper.
beat for the circuit shown in Fig.1: In some previous studies, people tended to consider C as
a constant within a range of blood pressure [9], or calibrate
. (2) it by least square estimation methods [8, 11]. In this paper,
we intend to tackle the problem from the physiological
where t k is the onset of the k-th beat, T k is the time aspect. It has been found that the relationship between


 WindkesselPPG

the blood vessel volume and blood pressure can be Excluding 36 trials for calibration, 288 trials were used
described by the following model [12]: for CO estimation. We calculated the root-mean-square
error by Equation (10) [11] to analyze the error between
(6) estimated CO (COest) and reference CO (COref).
where V represents the volume of blood vessel, m , b and
n are parameters that characterize the vessel properties (10)
and are approximately constant for the same subject.
Therefore, the arterial compliance, as defined in (7),
should vary with blood pressure changes.

(7)

Assume that we can obtain C1 and C2, which are the

arterial compliances of a subject in two physiological
states, we can then deduce from (7) to calculate the
parameter b with (8) and arterial compliance C with (9) Fig. 3 The recorded 18-min PPG data from subject ID3
respectively:

, (8)

(9)
Fig. 4 A segment of PPG data during 140~145s
In this study, we proposed using data measured while
a subject is sitting up and lying down as the two 4 RESULT
physiological states and pressure for calibration.
B. Experiment
Eighteen young healthy volunteers participated in this
T he results of this study found that using the proposed
method, CO can be estimated within 0.121.07
L/min of the reference for all subjects. The range of
experiment, including 10 males and 8 females, and their COest and COref and their RMSE are shown in TABLE
ages ranged from 22 to 35 years old. For each subject, a I. The smallest and largest RMSE are 5.13% and 18.87%
continuous 18-min data was recorded, which is shown respectively. Fig. 5 shows the Bland-Altman plot for all
in Fig. 3 and 4. During the experiment, the subjects trials of all subjects.
were first asked to sit for 2 minutes and lied down for 2
minutes, and they were then asked to perform a cycling Table I Estimated and Reference CO and Their Rmse
exercise in the supine position for 6 minutes and rest
for another 6 minutes before sitting up for 2 minutes.
PPG was captured from the left index finger of the
subjects and CO monitored by impedance cardiography
(PhysioFlowTM, France) continuously throughout the
experiment was used as the reference.
Data of each subject were divided into 18 trials and the
averaged of each trial was calculated. Blood pressure was
measured every 2 minutes during the experiment by an
experienced nurse using a mercury sphygmomanometer.
For trials without a cuff reading, the average of two
adjacent blood pressure readings was used instead. Data
of the first minute of sitting (i.e. the first trial) and the
first minute of lying down (i.e. the third trial) were used
for calibration. For two subjects, whose BP showed
no changes during change of posture, the first dataset
recorded during exercise was used instead for calibration.
The rest of the datasets were used for estimation.


 Vol. 4 No.6/ Jun. 2010
[5] Jin Li, Lin Yang, Song Zhang, et al, Computation of Cardiac
Output by Pulse Wave Contour, Bioinformatics and Biomedical
Engineering, 1088-1090, 2007.
[6] Hao Xinghai, Xu Min, Zheng Huiping, et al, Comparative Study
of Finger Volume Blood Flow Pulse Method and Themdilution
Method Regarding Cardiovascular Function, Chinese Journal of
Medical Device, 18(7), 6-8, 2005.
[7] John Allen, Photoplethysmography and its application in clinical
physiological measurement, Physiological Measurement, 28,
R1-R39, 2007.
[8] TA Parlikar, T Heldt, GV Ranade, et al, Model-Based Estimation
of Cardiac Output and Total Peripheral Resistance, Computers
in Cardiology, 34, 379-382, 2007.
Fig. 5 Bland-Altman analysis of COest and COref
[9] Zhenwei Lu, Ramakrishna Mukkamala, Continuous cardiac
output monitoring in humans by invasive and noninvasive
5 Conclusion peripheral blood pressure waveform analysis, J Appl Physiol,
101(2), 598-608, 2006.

I n this study, we developed a non-invasive method [10] , , , ,

, 2006.
based on the two-element Windkessel model to
[11] Parlikar T, Modeling and Monitoring of Cardiovascular
estimate CO from PPG. A new calibration method using
Dynamics for Patients in Critical Care, Ph.D. Thesis,
the change of body posture is proposed for the method. Massachusetts Institute of Technology, Cambridge, MA, June,
The results suggested that the estimated CO can track 2007.
CO and its variability in most subjects; however, there [12] Zhaorong Liu, Kenneth P. Brin, Frank C. P. Yin, Estimation of
are occasional trials where the difference between the total arterial compliance: an improved method and evaluation
estimated and reference CO can be over 4L/min. There of current methods, the American Physiological Society,
H588-H600, 1986.
maybe several reasons for this phenomena. First, some
[13] Kemps HMC, Thijssen EJT, Schep G, et al, Evaluation of
other factors for example the inertia of blood volume two methods for continuous cardiac output assessment during
and central venous pressure, which were not considered exercise in chronic heart failure patients, Journal of applied
in this study, also influence CO or arterial compliance, physiology, 105(6), 1822-1829, 2008.
and were not taken into account by the two-element
Windkessel model. Second, there is a recent report [13] on
the inaccuracy of the CO measurements by impedance

cardiography, which can also contribute to the difference
between the estimated and reference CO. Third, the Photoplethysmogram
relationship between blood volume and blood pressure is
still ambiguous. Last, the systolic pressure and diastolic
pressure which used in this study are measured by a /
mercury sphygmomanometer every two minutes, so the
pressure measurement was not continuous or beat-by-
2
beat. In the future study, the proposed method should
be improved and evaluated against invasive methods
such as thermodilution or Doppler ultrasound method
to confirm the accuracy of this non-invasive estimation
of CO, and other indictors would be introduced instead
of the systolic pressure and diastolic pressure to trace
cardiac output.

References
[1] World Health Organization, The Global Burden of Disease:
2004 Update, Switzerland: WHO Press, 8-10, 2008.
[2] http://en.wikipedia.org/wiki/Cardiac_output
[3] Tanino Y, Shite J, Paredes OL, et al. Whole body bioimpedance
monitoring for outpatient chronic heart failure follow up [J].
Circulation Journal, 73: 1074-1079, 2009.
[4] S Jhanji, Dawson, RM Pearse, Cardiac output monitoring: basic
science and clinical application, Anaesthesia, 172-181, 2008.

10