31st Annual International Conference of the IEEE EMBS

Minneapolis, Minnesota, USA, September 2-6, 2009

Noninvasive Cardiac Output Estimation Using a Novel

Photoplethysmogram Index
L. Wang, Emma Pickwell-MacPherson, Y. P. Liang and Y. T. Zhang

Abstract—Cardiac output (CO) monitoring is essential for noninvasive, accurate, reliable and continuous. At present,
indicating the perfusion status of the human cardiovascular no single method meets all these criteria. Intermittent
system under different physiological conditions. However, it is thermodilution is widely accepted as the clinical golden
currently limited to hospital use due to the need for either standard. This method requires the insertion of a pulmonary
skilled operators or big, expensive measurement devices.
artery catheter (PAC) to obtain one measurement per 3-4
Therefore, in this paper we devise a new CO indicator which
can easily be incorporated into existing wearable devices. To minutes [1]. It is too invasive and non-continuous. Two
this end, we propose an index, the inflection and harmonic area existing less invasive and continuous methods are
ratio (IHAR), from standard photoplethysmographic (PPG) oesophageal Doppler monitoring and CO2 re-breathing, but
signals, which can be used to continuously monitor CO. We both of these require skilled operators and expensive
evaluate the success of our index by testing on sixteen measurement devices [1]. Amongst the currently used
normotensive subjects before and after bicycle exercise. The
methods, impedance cardiography is probably the only
results showed a strong intra-subject correlation between
IHAR and COimp measured by the bio-impedance method in noninvasive and automatic technique. However, the
fifteen subjects (mean r = 0.82, p<0.01). After least squares impedance device is big and expensive, and its accuracy is
linear regression, the precision between COimp and CO often influenced by the change of electrode positions and the
estimated from IHAR (COIHAR) was 1.40 L/min. The total sweat on the skin [2]. Due to the disadvantages mentioned
percentage error of the results was 16.2%, which was well above, these methods are all unquestionably limited to
below the clinical acceptance limit of 30%. The results suggest
bedside use. They are not portable, or wearable, so they are
that IHAR is a promising indicator for wearable and
noninvasive CO monitoring. difficult to incorporate into home health care monitoring
systems. To solve this problem, one of the best ways is to
I. INTRODUCTION derive a new CO indicator from signals provided by the
existing wearable device. Some preliminary studies have
I T is well known that heart rate, blood pressure and cardiac
output (CO) are all essential physiological parameters of
the human cardiovascular system. CO, defined as the blood
implied that the PPG signal could be a candidate for this
application [3]-[6].
The PPG signal, which indicates the blood volume
volume ejected by the heart per minute (unit: L/min, where
changes on site, is an optical signal that could be non-
“L” means liter), is regarded as the ultimate expression of
invasively obtained from body peripheral terminals, such as
cardiovascular performance, since it indicates how well the
ear, finger and toe. The PPG acquisition components,
heart is able to provide enough nutrition and oxygen to the
including a pair of LED emitter and receiver, and related
peripheral organs and tissues. For human beings, in order to
simple circuits, are cheap and small in size, and could be
maintain a normal state of tissue perfusion and oxygen
easily embedded into many existing wearable devices.
delivery condition, the baseline CO should be in the range of
Although PPG is obtained peripherally, many previous
4 L/min to 8 L/min. If CO gets out of this range, it is often a
studies have revealed that the PPG wave contour is primarily
sign of cardiovascular disease, such as hypertension, stroke
influenced by characteristics of the systemic circulation, but
or heart failure. Hence, continuous CO monitoring plays an
not the local perfusion [3]-[6]. Henry Lax and his colleagues
essential role in the evaluation, treatment, and follow-up of
[3] noted that when individuals had cold fingers, this
critically ill patients.
“reduces the over-all amplitude but does not affect the
Ideally, a technology which measures CO should be
configuration of the pulse wave”. Chowienczyk PJ found in
his experiment that the vasodilation drugs which increased
Manuscript received April 7, 2009. This work was supported in part by the local circulation in the upper arm could not change the
the Hong Kong Innovation and Technology Fund (ITF). The authors are corresponding finger PPG wave contour, while, the
grateful to Standard Telecommunication Ltd., Jetfly Technology Ltd.,
Golden Meditech Company Ltd., Bird International Ltd. and Bright Steps systematic insertion of vasodilation drugs, such as GTN,
Corporation of their supports to the ITF projects. could change it substantially [4]. More recently, we have
Y. T. Zhang (phone: 852-2609-8459; e-mail: ytzhang@ee.cuhk.edu.hk), successfully utilized a parameter derived from the PPG wave
Emma Pickwell-MacPherson, Y. P. Liang and L. Wang are with the Joint
Research Centre for Biomedical Engineering, Department of Electronic to trace the change of total peripheral resistance (TPR) after
Engineering, The Chinese University of Hong Kong, Hong Kong SAR. bicycle exercise [5]. McCombie proposed a blind system
Y. T. Zhang is also with the Institute of Biomedical and Health identification method to calculate the cardiac output
Engineering, Shenzhen Institute of Advanced Technology, Chinese
Academy of Sciences, China, and Key Laboratory for Biomedical waveform from two PPG signals measured from different
Informatics and Health Engineering, Chinese Academy of Sciences, China. body locations, but only a qualitative aortic flow curve was
978-1-4244-3296-7/09/$25.00 ©2009 IEEE 1746
obtained [6]. invasive arterial catheter or a bedside Finapres device for
In this paper, we propose a novel indicator of CO, the acquiring the continuous blood pressure measurement.
inflection and harmonic area ratio (IHAR), which is derived
from PPG signals. The ability of IHAR to trace CO changes PPG
was evaluated in a bicycle exercise experiment. Time domain Frequency domain

ABP FFT(f1)

magnitude
FFT(f2)

S1 S2
CO AC TPR

f1 f2 … … fn Hz
frequency
(a) (b)
Fig. 2. (a) is PPG wave in the time domain. S1 and S2 are the areas
under the whole PPG wave and the part of wave after the point of
Fig. 1. Two-element Windkessel model. inflection, respectively. IPA=S2/S1. (b) is PPG wave in the frequency
domain. FFT ( f n ) is the magnitude at the nth harmonic. NHA =
N N
II. METHODOLOGY ∑ FFT ( f ) ∑ FFT ( f ).
n =2
2
n
n =1
2
n

According to the two-element Windkessel model shown However, as shown in equation (1), blood pressure
in Fig. 1, CO could be calculated from the analysis of measurement is not a necessity for obtaining CO, if proper
continuous arterial blood pressure waveform by a so called surrogates of MBP and TPR can be derived from other
pulse contour method. In this model, the cardiovascular signals. According to wave reflection theory, arterial blood
system is analogous to a current source connected with a pulse could be divided into two waves: a first wave
two-element circuit. CO, which is mimicked by the mean produced by heart pumping and a second wave produced by
amount of current passing through the TPR (total peripheral pulse wave reflection. Therefore, the inflection point area
resistance), equals the mean pressure (mean arterial blood ratio (IPA), the area ratio of the second and first peak in the
pressure, MBP) divided by the TPR: PPG wave (see Fig. 2 (a)), is mainly influenced by the
CO = MBP TPR (1). strength of pulse wave reflection. Pulse wave reflection
In this method, if a continuous arterial pressure waveform is results from the impedance mismatch between different parts
obtained, MBP could be calculated beat to beat. TPR is in the arterial system, e.g., the compliance and resistance
firstly initialized by a pair of calibration CO and MBP data, mismatches between the big, elastic arteries and the small
and its value of the current beat is calculated from MBP and arteries. Studies have shown that approximate 90% of the
estimated CO of the previous beat, iteratively. The main TPR is located in the small arteries. Hence, if the small
shortcoming of such technique is that it needs either an arteries contract, the TPR will change, which will change the

5-min. 5-min. 60W 5-min. 60W 5-min. 60W

rest bicycle exercise bicycle exercise bicycle exercise
…… ……

…… ……
1 2 3 4 5 6 7 8 9 10 11 17 18 19 25

3-min. 6-min. 6-min. 6-min.

Resting Post-exercise 1 Post-exercise 2 Post-exercise 3
represented BP measurement by a registered nurse
numbers 1-25 indicate the datasets of simultaneous measurements of CO and PPG
Fig. 3. Experiment procedure.
1747
impedance mismatch, influence the strength of the wave IV. DATA ANALYSIS
reflection, and further change IPA. In our previous For each subject, 25 datasets were obtained, including 1
investigation [5], IPA has shown to be a very good indicator dataset at rest and 8 datasets in each post exercise phase (see
of TPR when it changes after exercise. The pulse wave Fig. 3). The 8 datasets after each exercise were obtained
reflection not only changes the shape of the PPG signal in from the 6-min continuous data divided by the intermittent
the time domain, but also in the frequency domain. Our BP measurements. As shown in Fig. 3, BP was also
previous study [7] has proved that the normalized harmonic measured by a nurse every other dataset, e.g., during datasets
area (NHA), a parameter derived from frequency domain 2, 4, 6, 8 for post exercise phase 1.
analysis (see Fig. 2 (b)), is related to pulse wave reflection For offline analysis, each dataset of PPG signals was re-
and is strongly correlated with systolic and diastolic blood sampled to 100 Hz and divided into beat-to-beat waves. The
pressure. Therefore, IHAR, defined as NHA divided by IPA, start point of each beat is defined as the beginning of its
is proposed as a potential CO indicator and expressed as systolic rise, which is also the end of the diastolic fall wave
follows: of the previous beat. Single PPG wave was transferred to the
⎛ N
⎞ frequency domain by 512 points FFT (ensuring a resolution
⎜ ∑ FFT 2 ( f n ) ⎟
⎜1 − n=2 ⎟ of more than 10 points per Hz). Then, beat-to-beat IHAR
⎜ N ⎟ was calculated using equation (2). Meanwhile, COimp was
⎜ ∑ FFT ( f n ) ⎟
2
obtained by the impedance cardiograph device, beat by beat.
IHAR = ⎝ n=1 ⎠ (2), “Beat by beat” here meant that COimp was acquired and
⎛ S2 ⎞ saved in a “beat to beat” format, but was actually calculated
⎜⎜ ⎟⎟
⎝ S1 ⎠ from the rolling average of eight impedance wave and five
where S1 and S2 are the areas under the whole PPG wave and ECG wave in the algorithm. As a result, it was impossible in
the part of wave after the point of inflection in the time this study to compare IHAR with COimp in a beat-to-beat
domain, respectively (see Fig. 2(a)), and FFT 2 ( f n ) is the way, but only to compare the means of each dataset. For CO
estimation, all datasets of IHAR and COimp were firstly used
square of the magnitude at the nth harmonic of the PPG wave to produce CO = α ⋅ IHAR + β by the least squares linear
in the frequency domain (see Fig. 2(b)).
regression method, and then, the COIHAR results were
The performance of IHAR for CO estimation was
obtained from IHAR by the above equation.
evaluated in a bicycle exercise experiment. An impedance
cardiograph device, Physio Flow PF-05 (Manatec
V. RESULTS
Biomedical, Macheren, France), was used as the benchmark
of CO measurement. The validity of this device has been As shown in Table I, the hemodynamic parameters,
established in both normotensive adults and cardiovascular including heart rate, systolic blood pressure, diastolic blood
patients versus the direct Fick method (lab golden standard pressure and cardiac output, have significantly changed after
of CO measurement) both at rest and exercise conditions [2] each 5 min of bicycle exercise, compared with those
[8]. obtained at rest (p<0.05).
The intra-subject correlation coefficients (r) between
III. EXPERIMENT PROTOCOL IHAR and COimp are listed in Table II. For all subjects,
IHAR was positively correlated to COimp. Except for subject
16 normotensive volunteers (8 males and 8 females), aged
13 (p=0.054), there was a strong correlation between IHAR
from 22 to 34 years, participated in this study. Informed
and COimp in the other 15 subjects (mean r = 0.82, p<0.01).
consent was obtained from each participant. The
experimental procedures are depicted in Fig. 3. Upon arrival, TABLE I
CHANGES OF HEMODYNAMIC PARAMETERS AFTER 1 , 2 , AND 3RD
ST ND
subjects were asked to rest for 5 minutes. Then, the systolic
EXERCISE COMPARED WITH THOSE AT REST
blood pressure (SBP) and diastolic blood pressure (DBP)
1st 2nd 3rd
were measured from the left arm by a registered nurse. After At Rest
Exercise Exercise Exercise
this, PPG and cardiac output were recorded simultaneously Heart Rate
76±10 92±17﹡ 95±17﹡ 100±16﹡
for 1 min. Then, the subjects were asked to ride the bicycle (beat/s)
SBP (mmHg) 106±12 116±14﹡ 117±14﹡ 118±15﹡
(Corival V2, Lode VB, Groningen, Netherland) for 5
DBP (mmHg) 61±7 59±10﹡ 60±10﹡ 61±10﹡
minutes at 60 W. Immediately after exercise, 6-min of PPG CO (L/min) 6.3±1.2 8.5±1.8﹡ 8.7±1.9﹡ 9.2±2.0﹡
and CO data were continuously acquired, meanwhile, at time ﹡
significant difference with the results obtained at rest (p<0.01).
points after exercise of 0 min, 1.5 min, 3 min, 4.5 min and 6
min, blood pressure (BP) was measured. Then, the same 5
The dynamic range of CO estimated from IHAR (COIHAR)
min exercise and post-exercise data acquisition procedures
was very similar to that of COimp (COIHAR: 4.02 L/min-14.72
were repeated twice. PPG was recorded on the right index
L/min; and COimp: 3.66 L/min-14.48 L/min). As shown in
finger by an in-house designed acquisition system. CO was
Table II, the standard deviation (SD) of the estimation
measured by the Physio Flow PF-05 impedance cardiograph
residue is 0.71 L/min. This value does not vary much with
device.

1748
subject variance, which indicates a stable performance of percentage error (PE = 16.2 %) of our technique is well
IHAR based CO estimation. In many previous CO studies, a below the clinically acceptable error of 30%.
percentage error (PE) is calculated to indicate the variance of A potential limitation of this work is the change of PPG
estimation error with respect to reference mean wave as a result of vascular aging or cardiovascular diseases,
( PE = 1.96SD MEANCO ), and a PE less than 30% is e.g., the inflection point may be blurred and difficult to find
ref
as the arterial stiffness increases. Therefore, further
regarded to be clinically acceptable [9] [10]. The PE of the
validation on subjects with a bigger age range and
results in this study is 16.2 %.
cardiovascular patients are needed. Another limitation is the
TABLE II calibration procedure. Since the main purpose of this study
THE SUMMARY OF THE INTRA-SUBJECT CORRELATION COEFFICIENT r is to propose and evaluate a novel CO index, we have not
BETWEEN COIMP AND IHAR AS WELL AS THE REGRESSION COEFFICIENT
fully investigated the calibration procedure. Only a simple
AND STANDARD DEVIATION (SD) OF RESIDUE IN THE LEAST SQUARE
LINEAR REGRESSION least squares linear regression was utilized. However,
COIPHA = α ⋅ IHAR + β SD of individual calibration is an essential procedure for IHAR
Subject r Residue based CO estimation and will be investigated further in a
α β (L/min) separate study.
S01 0.94﹡ 195.1 3.7 0.83
S02 0.94﹡ 189.6 2.9 0.62
S03 0.78﹡ 265.0 0.9 0.61 REFERENCES
S04 0.92﹡ 199.7 2.8 0.64 [1] L. Mathews and K. RK Singh, “Cardiac output monitoring,” Annals of
S05 0.58﹡ 689.0 -3.1 0.87 Cardiac Anaesthesia, vol. 11, Jan.-Jun. 2008, pp. 56-68.
S06 0.80﹡ 272.0 0.0 0.78 [2] R. Richard, E. Lonsdorfer-Wolf, A. Charloux, et al., “Non-invasive
S07 0.91﹡ 347.9 -0.4 0.49 cardiac output evaluation during a maximal progressive exercise test,
S08 256.1 4.8 0.86 using a new impedance cardiograph device,” Eur J Appl Physiol., vol.
0.79﹡
85, Aug. 2001, pp. 202-7.
S09 0.96﹡ 172.1 2.0 0.38
[3] H. Lax, A. Feinberg and B. M. Cohen, “Studies of the arterial pulse
S10 0.89﹡ 319.7 0.9 0.74 wave and its modification in the presence of human arteriosclerosis,”
S11 0.83﹡ 396.6 -0.3 0.85 J Chronic Dis., vol. 3, Jun. 1956, pp. 618-31.
S12 0.96﹡ 282.3 2.4 0.43 [4] P. J. Chowienczyk, R. P. Kelly, H. MacCallum, et al.,
S13 0.39 185.9 4.9 0.78 “Photoplethysmographic assessment of pulse wave reflection: blunted
S14 0.78﹡ 345.1 0.2 0.90 response to endothelium-Dependent Beta2-Adrenergic vasodilation in
S15 0.69﹡ 283.7 4.2 0.83 type II diabetes mellitus,” J Am Coll Cardiol, vol. 34, Aug. 1999, pp.
S16 0.90﹡ 245.8 4.2 0.72 2007-14.
[5] L. Wang and Y. T. Zhang, “A novel photoplethysmogram index for
Mean 0.82 0.71
﹡
total peripheral resistance after bicycle exercise,” Proc. of the 5th
p<0.01. International Conference on Ubiquitous Healthcare, Pusan, Korea,
Oct. 2008, pp. 175-6.
The Bland-Altman plot (see Fig. 4) shows the agreement [6] D. McCombie, A. Reisner and H. Asada, “Identification of vascular
of the estimated CO and the reference CO. The percentage dynamics and estimation of the cardiac output waveform from
of CO estimations within the 95% confidence interval ([- wearable PPG sensors,” Proc. of the 27th IEEE International
Engineering in Medicine and Biology Conference, EMBC 2005,
1.40 L/min -1.40 L/min]) is 95.9%. Shanghai, China, pp. 3490-3.
[7] Y. S. Yan and Y. T. Zhang, “Noninvasive estimation of blood
pressure using photoplethysmographic signals in the period domain,”
3
Proc. of the 27th IEEE International Engineering in Medicine and
Difference of Reference an d

2 Biology Conference, EMBC 2005, Shanghai, China, pp. 3583-4.

E stim ated CO (L /m in )

[8] A. Charloux, E. Lonsdorfer-Wolf, R. Richard, et al., “A new

1 impedance cardiograph device for the non-invasive evaluation of
cardiac output at rest and during exercise: comparison with the
0
95.9% “direct” Fick method,” Eur J Appl Physiol., vol. 82, Jul. 2000, pp.
0 5 10 15 20 313-20.
-1 [9] M. Biais, K. Nouette-Gaulain, V. Cottenceau, et al., “Cardiac output
measurement in patients undergoing liver transplantation: pulmonary
-2 artery catheter versus uncalibrated arterial pressure waveform
analysis,” Anesth Analg., vol. 106, May 2008, pp. 1480-6.
-3 [10] L. A. Critchley and J. A. Critchely, “A meta-analysis of studies using
Average of Reference and Estimated CO (L/min) bias and precision statistics to compare cardiac output measurement
techniques,” J Clin Monit Comput., vol. 15, Feb. 1999, pp. 85-91.

Fig. 4. Bland-Altman plot of cardiac output.

VI. CONCLUSION
In this study, we have tested the performance of a new CO
index, IHAR, in 16 normotensive subjects through a bicycle
exercise study. The strong intra-subject correlation suggests
that IHAR can successfully trace the CO changes over a
wide dynamic range before and after exercise. The small

1749