ROBUST CHARACTERISTIC POINTS FOR ICG

Definition and Comparative Analysis

P. Carvalho, R. P. Paiva, J. Henriques

Department of Informatics Engineering, University of Coimbra, Pólo II, Coimbra, Portugal

M. Antunes

Cardiothoracic Surgery Center of the Hospitais da Universidade de Coimbra, Coimbra, Portugal

I. Quintal

Echocardiography Department of the Centro Hospitalar de Coimbra, Coimbra, Portugal

J. Muehlsteff

Philips Research Laboratories Europe, Aachen, Germany

Keywords: ICG, B-Point and X-Point Detection, Systolic Time Intervals, pHealth.

Abstract: The impedance cardiogram (ICG) is a promising tool for non-invasive and cost effective assessment of the

hemodynamic state, especially in low acuity and home settings. Important diagnostic parameters are related

to characteristic points within the ICG, i.e. the B and the X points that are assumed to mark the opening and

closure of the aortic valve., respectively. Based on synchronized echocardiography-ICG data obtained from

healthy subjects at rest, we compare 4 existing alternatives for the ICG’s characteristic point definitions

associated detection algorithms. We show that those points exhibit considerable biases with respect to the

intended onsets of the systole and diastole compared to the Echocardiography Goldstandard. We introduce

a new approach to determine these characteristic points based on the analysis of the ICG morphology. For

its implementation a computationally simple algorithm, based on high order derivatives, is proposed. This

algorithm is evaluated using simultaneously recorded echocardiographies and ICG signals. The achieved

results show that the proposed method enables the identification of the main characteristic points, B and X,

with significantly smaller errors and much higher correlations compared to current state of the art methods

and existing alternative characteristic point definitions.

1 INTRODUCTION

The impedance cardiogram (ICG) is one of the

reference methods for portable devices in assessing

several key hemodynamic descriptors, such as the

systolic time intervals (STI) and the cardiac output

(CO). This measurement principle is based on the

measurement of the thorax impedance variations

(dZ/dt) that are influenced by airflow through the

lungs, blood flow from the left ventricle to the aorta

and lung perfusion. It is able to track the relative

changes of stroke volume (SV) in healthy subjects

and patients without valve diseases Gotshall et al.

(1989). The assessment of the systolic time intervals

Figure1: Definition of characteristic points for aortic valve

events in the Impedance Cardiogram. Points B and X are

the traditional definitions for opening and closing events

of the aortic valve. B

new

and X

new

correspond to the

proposed definitions. Signal notches related to the

opening and closing of the aortic valve are shown in

circles.

161

Carvalho P., P. Paiva R., Henriques J., Antunes M., Quintal I. and Muehlsteff J..

ROBUST CHARACTERISTIC POINTS FOR ICG - Deﬁnition and Comparative Analysis.

DOI: 10.5220/0003134901610168

In Proceedings of the International Conference on Bio-inspired Systems and Signal Processing (BIOSIGNALS-2011), pages 161-168

ISBN: 978-989-8425-35-5

 2011 SCITEPRESS (Science and Technology Publications, Lda.)

Figure2: Relationship between ICG characteristic points and aortic valve cusps movements. (top) Echocardiography M-

mode; points 1 and 2, respectively, indicate the onset and the end of the movement of the aortic cusps during the opening

movement of the valve; points 3 and 4, respectively indicate the onset and the end of the aortic cusps during the closing

movement. (middle) ECGs captured by the different measurement modalities after synchronization. (bottom) Synchronized

ICG signal with the echo.

Figure 3: Relationship between blood ejection through the aortic valve and the ICG characteristic points. AVO and AVC

represent aortic valve opening and closure, respectively. (top) Echocardiography Doppler-mode. (middle) ECGs captured

by the different measurement modalities after synchronization. (bottom) Synchronized ICG signal with the Doppler Echo

(top image).

as well as SV from ICG require the determination of

the ICG’s characteristic points, which are assumed to

be correlated to the opening and closing of the aortic

valve. The rationale in using ICG to measure the

systolic time intervals is the assumptions that the

notches of the dZ/dt signal (see fig. 1) are due to

conduction changes of blood caused by changes in

orientation of erythrocytes (Visser et al., 1993) that

can be directly related to the opening and closing

events of the aortic valve. Namely, the so-called B

and X points are assumed to mark, respectively, the

opening and the closing of the aortic valve. Using

these reference ICG points, it is observed that the

pre-ejection period (PEP) is defined by the time

interval between the ECG’s Q-peak and the ICG’s B

point, while the left ventricle ejection time (LVET)

is defined as the time interval between the

characteristic points B and X (see fig. 2).

As can be observed in fig. 1, the most common

definitions of the B and X points assume that the B

BIOSIGNALS 2011 - International Conference on Bio-inspired Systems and Signal Processing

162

point corresponds to the local minimum of the notch

to the left of point C, while the X point corresponds

to the time instant where the lowest ICG value

occurs during the negative ICG signal section, to the

right of the C point. Using synchronized

echocardiography (clinical gold standard) with ICG

(see fig. 2) suggests that the traditional definitions of

B and X points do not correlate well to the exact

opening and closing instants of the aortic valve,

respectively. In fact, as was observed by Shyu et al.

(2004) using pressure-volume (PV) loops, the X-

point tends to occur substantially later than the E-

point of the PV loop, which marks the closing of the

aortic valve. The multi-modal graphical data

provided by several authors (e.g. (Wang et al.,

1995)) seem to suggest that the ICG’s B-point

occurs after the onset of the opening movement of

the aortic valve cusps. It should also be mentioned

that some authors (e.g. (Reddy et al., 1988), (Visser

et al., 1993), (Visser et al., 1991)) define the onset

of the aortic valve opening event at the zero-crossing

point of dZ/dt prior to point C (variants exist where

these points are measured at certain distances from

this reference).

There have been several efforts to automate the

detection of the ICG’s characteristic points. Several

algorithms rely on ensemble averaging to overcome

artifacts in the ICG signal (Nagel et al., 1986)

related to baseline drift caused by respiration and

movement artifacts. Because of heart rate variability,

this technique tends to blur less distinctive events

(particularly the B-point) making their detection

more difficult. On the other hand, it comes at the

expense of beat-to-beat variations, since ensemble

averaging filters these out. In order to overcome

these limitations, many authors suggest the use of

filtering and adaptive thresholding techniques. Wang

et al. (1995) use the spectrogram in different

frequency regions to extract a salience measure of

the characteristic points. A review on salience

measure-based methods is reported in (Pandey and

Pandey, 2005). Other authors exploit the wavelet

transform (WT). The advantage of WT is that it

decomposes the signal into different frequency bands

or scales, while preserving and characterizing the

regularities of the signal in those scales. Shyu et al.

(2004) and Shuguang et al. (2005) explored the

zero-crossings and local extremes to find the

characteristic points in ICG in a particular scale.

In this paper we introduce a new definition for

the characteristic points in ICG and an algorithm for

their detection using a computationally simple and

efficient method based on high order derivatives.

The basis for this new definition of the ICG’s

characteristic points is described in section 2.1. The

details of the algorithm are introduced in section 2.2.

A performance comparison with respect to the

current clinical gold standard – the

echocardiography – using state-of-the-art

characteristic point definitions and detection

methods and a commercial system is provided in

section 3. Finally, in section 4, some main

conclusions are presented and discussed.

2 METHODS

2.1 Characteristic Point Definition

There is evidence that the characteristic points

considered in literature for ICG correspond only

approximately to the events used to mark the systole

and the diastole phases of the cardiac cycle, i.e. the

movements of the aortic valve cusps. Heart valve

movements are not instantaneous, but rather

transitory processes that have their intrinsic dynamic.

This is perfectly visible in fig. 2, where the ICG

signal has been co-registered with an M-mode

echocardiography of the aortic valve that enables to

trace the aortic valve cusps movements. As can be

observed, the opening movement of the aortic cusps

is located at the notch of the ICG, to the left of point

C. As depicted in fig. 1, this notch corresponds to a

relatively large region of the ICG. The traditional

definition of the B-point is at the base of the notch

(see fig. 1 and fig. 2). However, from fig. 2 it is seen

that physiologically this point does mark neither the

onset nor the end of the cusps movement during its

opening dynamics at the beginning of the systole.

Synchronized ICG-echocardiographies suggest that

these events occur earlier in time and tend to

correspond, respectively, to the inflection point of

the ICG curve, to the left of the maximum of the

notch, and to a point near the notch’s maximum.

To overcome the uncertainty induced by the

dynamics of the cusps during the opening and

closing of the aortic valve, the left ventricle blood

ejection lobe is usually applied as a reference in

clinical practice, which is measured using

echocardiography in Doppler mode. Fig. 3 depicts

the blood ejection lobe through the aortic valve

measured using the echo-Doppler principle and co-

registered with the ICG signal. As can be observed,

the onset of the ejection lobe tends to be localized

immediately before the ICG notch’s maximum, i.e.,

an event that occurs noticeably earlier than the

ROBUST CHARACTERISTIC POINTS FOR ICG - Definition and Comparative Analysis

163

classical B-point definition. This is the reference for

the B point proposed in this paper (see point B

new

fig. 1).

Regarding the onset of the diastole phase of the

heart cycle, a similar condition is observed. In ICG

literature, the X-point is defined as the point that

corresponds to the lowest ICG value of the first

negative ICG signal section to the right of the C-

Point. Using pressure-volume loops obtained

through invasive measurement, Shyu et al. (2004)

have noticed that this point typically occurs between

10ms and 50ms after the onset of the diastole. This is

in accordance with the synchronized

echocardiography-ICG measurements performed

during this work. As can be observed in fig. 2, the

onset of the aortic cusps closing movement tends to

correspond to the local ICG maximum/start of the

negative slope to the left of the X-point, while the

end of the cusps movement is typically related to the

ICG’s inflexion point located between the mentioned

maximum and the X-point. Not surprisingly, using

the echo-Doppler echocardiographies (see fig. 3), it

is observed that the end of the left ventricle ejection

lobe corresponds to a point between the two

aforementioned points. In this paper, the X-point

will be assumed to be the onset of the descending

part of the notch, to the left of the traditional X-point

(see point X

new

in fig. 1).

2.2 Characteristic Points Detection

Three main steps that target, respectively, the

detection of the C, B and the X points of the signal

form the proposed algorithm for ICG characteristic

point detection using the definition proposed in the

previous section. The algorithm operates on a beat-

by-beat basis. For simplicity, it will be assumed that

Figure 4: Illustration of the initial B0 estimation technique

(vertical axis arbitrary scaled).

Figure 5: Illustration of B and X point detection method

using high-order derivatives in an ICG signal with a subtle

B-point notch. AVO-Echo and AVC-Echo define,

respectively, the detected aortic valve opening and closing

moments using echocardiography. Dashed vertical

markers define B, X and C points detected using the

described approach. Signals are arbitrarily scaled and

translated.

the ICG signal included between two consecutive

ECG R-peaks or Q-peaks is fed into the algorithm.

The C-point corresponds to the most prominent

component of the ICG and can be readily detected as

the time instant where the maximum value of the

ICG signal occurs.

The B-point is usually defined at the base of the

notch to the left of the C-point (see fig. 1 and fig. 2).

Regarding the notch, it is observed that it is not

always well defined and, in many cases, is

completely absent from the ICG signal. Therefore,

the method proposed here first estimates the base of

the notch (Bo). In a second phase the method verifies

if a notch is present in the ICG signal under analysis.

If a notch type salience is found in the ICG, the top

of the notch is searched; otherwise the estimated

base is adjusted according to the signal’s

morphology. The processing steps involve the

computation of high-order derivatives, which

amplify high frequency noise. In order to minimize

high-frequency noise interference, in the proposed

approach the impedance signal is low-pass filtered

using a Butterworth filter with a cut-off frequency of

100Hz. This is motivated by the fact that a typical

ICG signal exhibits spectral components up to this

frequency range (Wang et al., 1995). Zero-phase

filtering is implemented, which involves filtering the

signal in both the forward and backward directions,

to eliminate phase distortion. The following steps

form the actual procedure:

Step 1: In this step, a first estimate of the notch’s

base is performed using an adaptation of the

procedure described by Onu et al. (2004). The first

estimate of the B-point, Bo, is illustrated in fig. 4. It

is obtained based on the intersection of the line fit of

BIOSIGNALS 2011 - International Conference on Bio-inspired Systems and Signal Processing

164

the ICG points between 40% (icg(Ko)) and 80%

(icg(Ku)) of the amplitude of point C with the

horizontal axis (see fig. 4). In order to avoid

underestimating Bo, its value is limited based on the

physiological observation that the aortic valve

opening typically occurs 30ms after the atria-

ventricular (AV) valve closing (Tavel, 1967). As an

approximate reference for the AV closing instant, the

ECG’s R-peak can be applied.

Step 2: This step is to check the existence of a

detectable notch. In order to detect the presence of

the notch, the method resorts to the second

derivative of the ICG signal. Using the backward

discrete derivative definition, the existence of a

notch should lead to a (+,-,+,-)

sign pattern of the

second derivative to the left of the C-Point. If this

second derivative sign pattern is found, then one

might assume that a notch exists in the ICG signal.

In this situation the B-point is defined by the first

minimum of the 3rd derivative to the left of Bo. The

use of this procedure enables the correct detection of

the aortic valve opening moment for well-defined as

well as for subtle notches (see fig. 5). For subtle

notches, it is observed that there is no clear 1st

derivative zero that can be applied to identify the top

of the notch. The 2nd derivative is always negative

due to the concavity of the notch’s region around the

local maximum; hence it is not straightforward to

apply it directly for the notch’s top identification

task. However, the intended point corresponds to the

minimum acceleration of the 1st derivative, i.e. a

minimum of the 3rd derivative (see fig. 5). A similar

approach has been followed by Chan et al. (2007) to

identify characteristic points in the dichrotic notch of

a Photoplethysmogram. If no sign pattern in the

second derivative is detected, then no detectable

notch exists. In this case, the B-point is assumed to

be the first zero-crossing of the first derivative of the

ICG to the left of Bo.

The third important event to be detected in the

ICG is the X-point. In a well-defined ICG signal, it

is observed that, using the traditional definition, the

X-point corresponds to the lowest ICG value of the

first negative ICG signal section to the right of the

C-Point. In practice, due to noise and respiration

artifacts, this might not be the case. Physiologically,

it is observed that the T-wave of the ECG

corresponds to the relaxation of the ventricles of the

heart. Hence, the closing of the aortic valve is

usually observed near the end of the T-wave. As a

first estimate of the X-point,

, the lowest ICG

negative minimum is taken in the interval

≤

1.75

, where

represents the duration

of the R-T ECG segment. As can be observed in fig.

2 and fig. 3, the actual closing of the aortic valve

tends to be localized near the top of the first local

ICG maximum to the left of

. This point can be

readily detected both in well-defined and in less

well-defined ICG signals using the local minimum of

the 3

derivative of the ICG to the left of X

. The

basis for using the 3

derivative in this context is the

same as for B point detection.

3 RESULTS AND DISCUSSION

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 ICG (a Niccomo device from Medis

® was

applied) and echocardiography (Siemens

® Acuson

CV70 device). 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).

The average heart rate during data collection was

72.94 ± 9.87 bpm. All subjects involved in this study

did not have any known congenital or other heart

disease. Regarding the main biometric characteristics

of the population, they exhibited average age of

22.53 ± 3.81 years and average body mass index of

23.27 ± 2.15 Kg/m

Table 1 and table 2 present the comparative

analysis of the achieved results for B and X points

detection, respectively, using five different methods

on the common database. The database is composed

by 564 heart beats with annotated aortic valve

opening and 358 beats with annotated aortic valve

closing moments using echocardiography as

reference. Regarding the echocardiography mode,

213 beats were annotated in M-mode and 351 beats

resorted to Doppler-mode. Values shown in table

and table 2 have been found to be statistically

relevant (p<0.001) using the Wilcoxon test

(Gaussianity was assessed using the Kolmogorov-

Smirnov test).

As can be observed, using echocardiography as

reference, it is clear that the classical definitions of

the systolic events in the ICG signal, i.e. the B and

the X points, do not capture adequately the actual

physiological opening and closing events of the

aortic valve. In fact, all of the conventional ICG

characteristic point identification methods exhibit

significant biases with respect to the

echocardiography reference in detecting both events.

ROBUST CHARACTERISTIC POINTS FOR ICG - Definition and Comparative Analysis

165

Furthermore, it is observed that these biases differ

substantially, leading to significant over or

underestimations of PEP and LVET. This is clearly

shown in table 3 where the impact on LVET

estimation is summarized for all the considered

methods. It should be observed that the statistics on

PEP estimation are the same as the ones reported for

point B (see table 1), given the definition of this

systolic time interval.

From table 1 it is seen that Onu’s algorithm (Onu

et al., 2004) is the most stable method in detecting

the B-point using its classical definition. It exhibits

the highest correlation with respect to the aortic

valve opening and the lowest standard deviation of

the error among the classical methods. Comparing

the results achieved by this method in detecting the

B and the X points, it is seen that the B point is

measured with an average bias of 19.9 ms, while the

X point exhibits a measurement bias of 34.6 ms (see

table 2), hence, as is shown in table 3, these biases

will not cancel out during LVET computation which

is required for stroke volume assessment.

Table 1: Comparative analysis of beat-to-beat B point

detection results with respect to echocardiography.

Average error is defined by the average of B-B

Echo

. The

average absolute error is defined by the average of |B-

Echo

|, whereas SD is the standard deviation of |B-B

Echo

All reported error and SD values are in ms. (

) Niccomo

ICG device from Medis®. (

) Spearman Correlation

(p<0.001).

Proposed

Method

Shyu et

al.

Onu et

al.

Nic

Zero-

Crossing

Av. Er. 5.8 11.4 16.5 9.8 -7.2

Av. Abs. Er.

12.4

8.7

21.1

12.8

19.9

12.5

19.3

13.4

23.96

17.2

Correlation

0.54 0.50 0.70 0.53 0.39

Table 2: Comparative analysis of beat-to-beat X point

detection results with respect to echocardiography.

Average error is defined by the average of X-X

Echo

. The

average absolute error is defined by the average of |X-

Echo

|, whereas SD is the standard deviation of |X-X

Echo

All reported error and SD values are in ms. (

) Niccomo

ICG device from Medis®. (

) Spearman correlation

(p<0.001). (

***

) Pearson correlation (p<0.001).

Proposed

Method

Shyu

et al.

Onu et

al.

Nic

Av. Er. 29.3 -59.9 14.2 -39.5

Av. Abs. Er.

32.2

25.0

61.9

38.0

34.6

22.3

44.1

37.3

Correlation

0.44 0.15 0.16

***

0.22

Table 3: Comparative analysis of beat-to-beat LVET

detection results with respect to echocardiography

reference. The average absolute error is defined by the

average of |LVET-LVET

Echo

|, whereas SD is the standard

deviation of |LVET-LVET

Echo

|. All reported error and SD

values are in ms. (

) Niccomo ICG device from Medis®.

(

) (p<0.001).

Proposed

Method

Shyu et al. Onu et al. Nic

Av. Abs. Er.

34.8

38.0

73.7

43.3

38.6

24.4

54.0

41.8

Correlation

0.36 0.05 0.14 0.33

As can be observed in table 1 and table 2, the

aforementioned biases are significantly reduced

using the new definitions proposed in this work for

the B and X points. Regarding B-point identification,

the proposed method achieves the smallest and most

stable estimation errors among all algorithms, i.e. an

absolute error of 12.4± 8.7 ms is obtained. This is

35.8% smaller compared to the smallest estimation

error achieved by any of the remaining classical

methods. Furthermore, as is depicted in fig. 6 (top),

Figure 6: Bland-Altman bias plots of the estimated beat-

to-beat B and X points with respect to the opening and

closing of the aortic valve measured with

echocardiography. Horizontal dashed lines represent mean

error and mean error ± standard deviation of error,

respectively. (top) B error distribution. Horizontal axis is

defined by (RB+RAVO)/2, where RB and RAVO define,

respectively, the time interval between the ECG’s R-peak

and the estimated B point, and the R-peak and the

measured aortic valve opening moment. (bottom) X error

distribution. Horizontal axis is defined by (RX+RAVC)/2,

where RX and RAVC define, respectively, the time

interval between the ECG’s R-peak and the estimated X

point, and the R-peak and the measured aortic valve

closing moment.

BIOSIGNALS 2011 - International Conference on Bio-inspired Systems and Signal Processing

166

the estimation error dispersion is not dependent on

the actual value of B; it distributes evenly for a large

range of B durations. Regarding the correlation

coefficient, it is observed that the new method

exhibits a smaller correlation compared to Onu’s

algorithm (0.54 vs. 0.70), although the dispersion of

the estimations is significantly smaller for the new

method. The reason for this is due to the fact that the

ICG B-point notch is not always well defined or

present in the signal. For Onu’s method this is not

very significant, since it does not rely on the notch to

detect the B point (the reason for the performance of

Niccomo’s algorithm is unknown). Another reason

for this lower correlation is noise. High-order

derivatives are high-pass filters, which amplify high-

frequency noise.

Regarding the definition of a marker for the

detection of the onset of the aortic valve opening

event, it is observed that the worst results were

achieved using the zero-crossing principle. It

exhibits a significantly larger absolute error as well

as dispersion compared to the other marker

definitions considered. This is due to the fact that (i)

it is known that the zero line may change as a

function of respiration (Reddy et al., 1988) and (ii)

the zero-crossing points tends to occur before the

notch induced by the aortic valve opening event. In

some situations this might induce that no zero-

crossing exists in the dZ/dt signal between the

ECG’s R-peak and the C-point in the ICG.

Regarding the detection of the X point, which

relates to the closing of the aortic valve, it is

observed from the inspection of the results reported

in table 2 that the identification errors are

significantly higher for all methods compared to the

performance achieved for B point detection. The

reason for this is linked to the fact that typical ICG

signals exhibit many minima to the right of the C-

point. In many circumstances these minima are very

similar in amplitude and, therefore, are very prone to

noise interference. Nevertheless, it is seen that the

proposed method has the lowest estimation error

(32.2 ms) and the highest correlation (0.44) with

respect to the annotated echocardiography. These

results clearly show that the traditional definition of

the X point does not correlate well to the closing of

the aortic valve and that the proposed definition and

algorithm significantly improves the estimation of

the aortic valve closing moment. Moreover, fig. 6

(bottom) shows that the performance of the proposed

method is stable for a significant range of systolic

timings.

4 CONCLUSIONS

Based on synchronized echocardiography-ICG data,

we have shown in this work that the classical

definition of B and X points in the ICG exhibit a non

constant bias and a reduced correlation to the

physiological systolic events they are mend to

capture, i.e. the opening and closing of the aortic

valve. This might have a significant impact on the

applicability of ICG in clinical practice, since it

determines the accuracy of the most useful diagnosis

variables - systolic time intervals, contractility and

stroke volume that can be obtained using the signal.

Currently, many researchers also use ICG to

compensate for PEP in pulse transit time

measurement (Payne et al., 2006). The finding

reported herein might also be significant in this

context.

A new definition for the main systolic

characteristic points in the ICG is suggested. These

points minimize the offset with respect to the

physiological events related to the left ventricle

ejection. Based on these new definitions, an

algorithm was introduced that enables a significant

reduction in left ventricle ejection time estimation

errors, while increasing its estimation stability, i.e.

lowering the estimation error dispersion and

increasing the correlation with respect to

echocardiography based measurements. The

algorithm enables the detection of the onset of left

ventricle ejection with an average absolute

estimation error of 12.4 ± 8.7 ms with a correlation

coefficient of 0.54. This is about 35.8% smaller

when compared to state of the art methods using the

classical definition of the B-point. Regarding the

detection of the end of the left ventricle ejection

phase, the proposed algorithm performs significantly

better compared to other state of the art methods.

The detected characteristic points exhibit a much

higher correlation (0.44) to the closing of the aortic

valve than other competing methods (best achieved

correlation 0.22). The average absolute detection

error of this event (32.2 ± 25.0 ms) is higher

compared to the performance achieved for the

detection of the onset of the left ventricle ejection.

Yet, it is smaller (in some cases almost 50% smaller)

and less dispersed compared to state of the art

methods. One might speculate that this might have a

significant impact on the value of ICG in medical

practice e.g. for cardiac output estimation. However,

its clear demonstration probably requires a new

study using thermo dilution as reference.

ROBUST CHARACTERISTIC POINTS FOR ICG - Definition and Comparative Analysis

167

ACKNOWLEDGEMENTS

This work was supported in part by the EU FP7

project HeartCycle (FP7–216695) and SoundForLife

(PTDC/EIA/68620/2006; FCOMP-01-0124-

FEDER-007243) financed by the Portuguese

Foundation for Science and Technology. 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 of Dr. Leitão Marques from CHC

in facilitating the arrangements for the data

acquisition part of the study is also acknowledged.

REFERENCES

Chan G., Middleton P., Celler B., Wang L., Lovell N.,

2007, Automatic detection of left ventricular ejection

time from a finger photoplethysmographic pulse

oximetry waveform: comparison with Doppler aortic

measurement, Physiol. Meas., 28:439–452

Gotshall R., Wood V., Miles D., 1989, Comparison of two

impedance cardiographic techniques for measuring

cardiac output, Ann. Biomed. Eng., 17:495-505.

Nagel J., Shyu L., Reddy S., Hurwitz B., McCabe P.,

Schneiderman N., 1989, New signal processing

techniques for improved precision of noninvasive

impedance cardiography, Ann. Biomed. Eng., 17:517-

534.

Onu T., Miyamura M., Yasuda Y., Ito T., Saito T.,

Ishiguro T., Yoshizawa M., Yambe T., 2004, Beat-to-

beat evaluation of systolic time intervals during

bicycle exercise using impedance exercise, Tohoku J.

Exp. Med., 203:17-29.

Pandey V., Pandey P., 2005, Cancellation of respiratory

artifact in impedance cardiography, Proc. of the Ann.

Conf. of IEEE-EMBC.

Payne R. A., Symeonides C. N., Webb D. J., Maxwell S.

R., 2006, Pulse transit time measured from the ECG:

an unreliable marker of beat-to-beat blood pressure. J

Appl. Physiol., 100:136-41.

Reddy S., Shyu L., Hurwitz B., Nagel J. H., 1988,

Schneiderman N., Improved reliability of impedance

cardiography by new signal processing techniques.

Proc. of the Ann. Conf. of IEEE-EMBC.

Shuguang Z., Yanhong F., Hailong Z., Min T., 2005,

Detection of Impedance Cardiography’s characteristic

points based on wavelet transform, Proc. of the Ann.

Conf. of IEEE-EMBC, 2730-2732.

Shyu L. Y., Lin Y.-S., Liu C.-P., Hu W.-C., 2004, The

detection of impedance cardiogram characteristic

points using wavelets, Computers in Biology and

Medicine, 34:165-175.

Tavel M., 1967, Clinical Phonocardiography and external

pulse recording, Year Book Medical Publishers, Inc.

Visser K. R., Mook G. A., Van der Wall E., Zijlstra W.,

1993, Theory of the determination of systolic time

intervals by impedance cardiography, Biological

Psychology, 36(1-2):43-50.

Visser K., Mook G., Van der Wall E., Zijlstra W., 1991,

Systolic time intervals by impedance cardiography.

Proc. of the Ann. Conf. of IEEE-EMBC.

Wang X., Sun H. Water J., 1995, An advanced signal

processing technique for impedance cardiography,

IEEE Trans. on BioMed. Eng., 42(2):224-230.

BIOSIGNALS 2011 - International Conference on Bio-inspired Systems and Signal Processing

168