Annals of Biomedical Engineering

, Volume 35, Issue 1, pp 45–58

Optical Vibrocardiography: A Novel Tool for the Optical Monitoring of Cardiac Activity

Authors

  • Umberto Morbiducci
    • Department of MechanicsUniversità Politecnica delle Marche
    • Department of MechanicsUniversità Politecnica delle Marche
  • Mirko De Melis
    • Department of MechanicsUniversità Politecnica delle Marche
  • Mauro Grigioni
    • Technology and Health DepartmentIstituto Superiore di Sanità
Article

DOI: 10.1007/s10439-006-9202-9

Cite this article as:
Morbiducci, U., Scalise, L., De Melis, M. et al. Ann Biomed Eng (2007) 35: 45. doi:10.1007/s10439-006-9202-9

Abstract

We present an optical non-contact method for heart beat monitoring, based on the measurement of chest wall movements induced by the pumping action of the heart, which is eligible as a surrogate of electrocardiogram (ECG) in assessing both cardiac rate and heart rate variability (HRV). The method is based on the optical recording of the movements of the chest wall by means of laser Doppler interferometry.

To this aim, the ECG signal and the velocity of vibration of the chest wall, named optical vibrocardiography (VCG), were simultaneously recorded on 10 subjects. The time series built from the sequences of consecutive R waves (on ECG) and vibrocardiographic (VV) intervals were compared in terms of heart rate (HR). To evaluate the ability of VCG signals as quantitative marker of the autonomic activity, HRV descriptors were also calculated on both ECG and VCG time series. HR and HRV indices obtained from the proposed method agreed with the rate derived from ECG recordings (mean percent difference <3.1%). Our comparison concludes that optical VCG provides a reliable assessment of HR and HRV analysis, with no statistical differences in term of gender are present. Optical VCG appears promising as non-contact method to monitor the cardiac activity under specific conditions, e.g., in magnetic resonance environment, or to reduce exposure risks to workers subjected to hazardous conditions. The technique may be used also to monitor subjects, e.g., severely burned, for which contact with the skin needs to be minimized.

Keywords

Cardiac displacementHeart rate variabilityLaser Doppler vibrometryElectrocardiography

Introduction

A great interest has grown during the years on displacement cardiography as enabling methodology for the examination of the cardiovascular dynamics, with several methods proposed for the assessment of cardiac rate. In the past, techniques such as balistocardiography28 and kinetocardiography38 have been proposed, but they have not been used widely. In the 1990s, the seismocardiography has been presented as a novel non-invasive technique (although it is a contact procedure) for recording and analyzing cardiac vibratory activity, 32,33 eligible to be partially or fully alternative to the “gold standard” technique (ECG) for HR assessment. Recently, Augousti and colleagues4 developed a new fiber optic plethysmographic method for monitoring cardiac activity based on the measurement of changes in the torso shape induced by the pumping action of the heart, demonstrating the interest in the development of methods alternative to ECG.

In this optics, wide potentiality could be offered by laser-based techniques that, making use of specific optical sensors, allow non-contact measurements of vibrations to be performed. In the last years, Laser Doppler Vibrometry39 (LDV) has been applied in biomedical areas,6,20 due to its advantageous metrological characteristics. In particular, the high accuracy (about 1% of reading), high resolution (displacement resolution about 8 nm) and the non-contact nature of this technique make these instruments suitable for diagnostic purposes and in vivo tests.20, 30,34

In virtue of the above mentioned facilities, LDV could be an interesting alternative investigational methodology to be tested in the study of cardiac function, the monitoring of heart rate being a typical task in clinical environment for a great variety of patient’s condition. This could be done measuring the movements transmitted to the chest wall by the compression waves generated by the beating heart during its pumping function.

ECG is the most widely used modern method for the monitoring of the cardiac activity (it can be considered the gold standard technique). Careful interpretation of the ECG can reveal not simply the cardiac rate, but subtle and reliable information concerning the condition and functioning of the subject’s heart, by means of HRV analysis. In fact, HRV represents a quantitative marker of autonomic activity, and a powerful tool in the recognition of the relationship between the autonomic nervous system and cardiovascular mortality.2,3,26,36

ECG can be conducted, in most of the cases, without major difficulties, even if in specific cases the use of standard electrocardiography could suffer from limitations.17,21

The monitoring of the ECG requires contact between skin and electrodes, and sensors and cables to be connected. The possibility to obtain information on HR and HRV without contact with the patient represents a future tool in many fields, reaching the goal of a risk reduction: for example preventing secondary exposure of medical personnel to toxic materials under biochemical hazard conditions,23,24 or avoiding potential hazard during magnetic resonance (MR) imaging examinations of monitored patients.14, 15,18. In general, the task of the monitoring of the cardiac activity could not be accomplished when patient’s condition determine impairment of their health if contact with skin or any other surface has to be avoided (e.g. severely burnt subjects31).

In all these situations a method alternative to the typical continuous electrocardiographic recording is desirable to monitor the cardiac rate. In this optics, non-contact devices that could allow from a distance (tens of meters) measurement in general population look to be very useful.

To this aim, we remind that during ventricular contraction, the heart undergoes changes in volume as well as variations in position. The resulting combination of motions is transmitted to the surface of the skin and could be picked up by a laser displacement sensor pointing at a point on the thorax, near the heart.

In this study we investigate the potential of a LDV based method to perform both HR and HRV analysis as the ECG, from measurements of the motion of the external surface of human body (chest wall). In contrast to ECG, the proposed approach can be carried out without electrodes and with the sole laser equipment.

Analysis of HRV was performed considering the time intervals between consecutive fiducial points on ECG and VCG synchronously recorded traces, according to the methods recommended by the task force.37 Time and frequency domain measures of HRV were calculated on both ECG and VCG derived time series, and their equivalence was evaluated.

Methods

Measurement Principle

The movements of the chest wall were measured using a single-point laser Doppler vibrometric system. The LDV technique may accurately measure point-by-point surface velocities using interferometric techniques.39

A laser Doppler vibrometer is based on the principle of the detection of the Doppler shift of laser light, scattered from the vibrating specimen: surface motion induces a Doppler frequency shift on the impinging laser beam, and this shift is linearly related to the velocity component in the direction of the laser beam. The relationship of the Doppler frequency shift fD introduced into the measurement beam, with the vibrational velocity v, is expressed by
$$ f_{\hbox{D}} \, = \,2\frac{{v(t)}} {\lambda } $$
(1)
with λ being the wavelength of the laser radiation. Therefore, vibrational velocity can be obtained from equation (1). As the Doppler shifts are usually very small when compared to the laser fundamental frequency (1 part out of 108, typically), the only way to appreciate such small quantities is to use interferometry, so that high frequency oscillations are reduced to much lower values that can be dealt with by standard electronics. In the present study we used a single-point, direct beam laser Doppler vibrometer (Polytec GmbH, Germany) incorporating a Mach-Zender interferometer, which allows the measurement of both the vibrational velocity and displacement. Figure 1 shows the diagrammatic sketch of the optics of a laser vibrometer, in which a coherent laser beam at frequency f0 is splitted into two equal beams (passing through a beam splitter). One of the beams, i.e. the measuring beam, is focused on the vibrating object, the other one stays in the laser head and it is used as reference beam. After being reflected from the surface of the (moving) object, the measuring beam re-enters the laser head and is recombined with the reference beam. Surface displacement modifies the optical path difference between the two laser beams and this results into a phase lag changing with vibration velocity: due to the Doppler effect, it is frequency shifted as f0 ± fD before interfering with the reference beam. Demodulating the Doppler signal it is possible to extract the amplitude of v (see equation (1)) The information on the direction of the surface motion can be extracted by means of an acousto optic modulator (center frequency fm), the Bragg cell: in this way, from the original laser beam two beams at frequency difference fm are obtained. Due to the low pass filter of the interferometry, which cut off the very high frequency f0, a signal with shifted heterodyne frequency fm ± fD can be finally revealed at the photo detectors.
https://static-content.springer.com/image/art%3A10.1007%2Fs10439-006-9202-9/MediaObjects/10439_2006_9202_Fig1_HTML.gif
Figure 1.

Sketch of the Mach-Zender interferometer incorporated into the laser vibrometer.

Experimental Set up

The LDV system performs measurements of velocity, i.e. the ones carried out in the present investigation, with a resolution up to 0.5 μm/s. The laser head was placed at about 1.5 m from the subject chest wall. In order to simplify the measurement procedure and in order to optimize the quality of the signal (increase of the S/N ratio of the vibratory signal) a small (about 2 mm2, weight <1 g) adhesive retro-reflective tape was placed on the chest wall.

Short-term recordings (5 min) were carried out on 10 healthy human subjects (five males aged from 24 to 30 years; five females aged from 23 to 30 years) at rest, lying supine on a bed. Figure 2 shows both the experimental set up, and the laser head positioning and the measurement point on the chest wall for VCG signal recording. The LDV system has a maximum velocity range of 10 m/s, a 0–350 kHz maximum bandwidth, a resolution of about 1 μm/s and an accuracy in the order of 1–2% of RMS reading. No filtering was set for the chest vibratory signals recording.
https://static-content.springer.com/image/art%3A10.1007%2Fs10439-006-9202-9/MediaObjects/10439_2006_9202_Fig2_HTML.gif
Figure 2.

Sketch of the experimental set-up. Laser head positioning and measurement site for VCG signal recording are also shown.

Laser power is less than 1 mW, so that no special safety measures are required, but nevertheless also with such low power levels working distances of some tens of meters are possible. On each individual investigated (as explained in the following) ECG and VCG traces were simultaneously recorded. The ECG (obtained from the II-lead output) is connected as in the classical configuration for the recording of the three fundamental leads. An analog-to-digital 12-bit acquisition board with anti-aliasing filters, together with a custom-made software program developed in a LabVIEW® environment (National Instruments, USA), have been used to store the signals. The analog inputs were sampled at 1 kHz. A PC (Pentium IV) was used both for setting of the A/D acquisition board and for providing the storage and processing of the experimental data.

Data Analysis

From the recordings on individuals undergoing measurements, we investigated the physiologic relationship of the ECG to the vibratory signal VCG in terms of heart rate variations. To do this, we applied methods proposed by the guidelines on international standards of heart rate variability.17

The measurement method is based on the assumption that the peak of the vibratory signal, which measures chest wall motion, takes place in consequence of the cardiac muscle contraction triggered by the electrical signal measurable by the use of the electrocardiograph. ECG and VCG recordings were filtered with an eighth-order Butterworth low pass filter with a cut-off frequency of 100 Hz. We used heartbeat fiducial timing point provided by ECG recordings, i.e., the time of occurrence of the major local extremum of a QRS-complex (the time of the R-wave maximum). As fiducial point in the VCG signal, we selected the first local maximum value (labeled V peak) in the vibratory trace that follows the R-wave maximum in the ECG trace (the rationale is that VCG signal is responsive to changes in myocardial contraction induced by electrical activation). The results of automatic labeling of R and V fiducial points were reviewed and manually edited for error correction.27

From synchronous recordings, ECG derived RR, and VCG derived VV, interval time series were generated. Before this, ECG signals were filtered with two median filters to remove the baseline wander: in particular, each ECG signal was firstly processed with a median filter of 200-ms width to remove QRS complexes and P-waves, the resulting signal was then processed with a median filter of 600 ms width to remove T-waves. The signal outcoming the second filter operation contained the baseline of the ECG signal, which was then subtracted from the original signal to produce the baseline corrected ECG signal.9

On the calculated RR and VV time series HRV descriptors were calculated, for linear and non-linear quantitative description, both in the time and the frequency domain.

Statistical Analysis

The agreement between ECG and VCG methods measures was evaluated from two aspects: (i) agreement when these techniques are used for cardiac rate monitoring, assessing intra-individual variations on RR and VV time series; (ii) agreement when on the RR and VV time series HRV indices are calculated. To examine the former agreement, 10 beats-to-10 beats VCG rate (VV) and heart rate (RR) for 5 min segments were compared within each subject, and the agreement between corresponding values were evaluated with the Bland and Altman test5: for the cardiac rate measured with the two methods, the difference against their mean value is plotted. This approach allows to evaluate if two methods for clinical measurement are interchangeable.

To examine the latter agreement, HRV descriptors calculated on intra-individual RR and VV synchronous time series were compared in terms of percent differences (the RR time series being the reference values), i.e., if XVV is the quantity computed on VV series, and XRR the corresponding quantity on the RR series, the percent difference is:
$$X(\% )\, = \,100 \cdot \frac{{\left( {X_{{\hbox{VV}}} - X_{{\hbox{RR}}} } \right)}} {{X_{{\hbox{RR}}} }} $$
(2)
The positive or negative value in the percent difference tells if the quantity computed on VV time series is greater or lower than the corresponding quantity on RR time series.

Statistical analysis was performed by means of nonparametric test due to the limited number of individuals, which prevented analysis of normality: differences between ECG and VCG derived indices (for each HRV index, values extracted by ECG or VCG from all individuals represent an independent sample containing mutually independent observations, i.e., the individuals themselves) was checked by using the Kruskal–Wallis one-way analysis of variance nonparametric test. Significant level was set at < 0.05.

Spectral Analysis

The Fast Fourier Transform (FFT) algorithm was applied for the calculation of the power spectral density (PSD). RR and VV time series were resampled at 2 Hz,8 linearly interpolated, and normalized (by subtracting their mean value and then dividing for the latter). A Hanning window was applied in the time domain to reduce leakage. The square of the FFT was computed, and the power spectral density (PSD) was obtained multiplying each frequency component by 2.66, to correct for Hanning window.22 From the PSD of both RR and VV tachograms total spectral power and LF/HF spectral powers ratio (LF, 0.04–0.15 Hz; HF 0.15–0.40 Hz) were calculated and compared.22,26. Moreover, we calculated also spectral entropy:12,29 PSD from ECG and VCG recordings extracted tachograms were normalized with respect to the total spectral power, then Shannon channel entropy was calculated on them, to have an estimate of spectral entropy of the process, as:
$$ E\, = \,\sum\limits_{i = 1}^N {p(f_i )} \log \left( {\frac{1} {{p\left( {f_i } \right)}}} \right) $$
(3)
where p(fi) is the normalized PSD value at frequency fi, and N the number of spectral components. As stated by Acharya et al.,1 the entropy can be heuristically interpreted as a measure of uncertainty about the event at frequency f.

Nonlinear Dynamics Analysis

It is known that the normal heart rate is not regular, but varies from beat to beat in an irregular manner.10 Among many tools for the studies of nonlinear dynamics of heart rate, the Poincaré plot of interval time series X deserves special attention: it is a technique portraying the nature of interval fluctuations in a time series, consisting in a scatter-graph built up plotting samples Xi+1 vs. Xi. Poincaré plot analysis is a powerful tool in the investigation of HRV,13 with the shape of the plot categorizable into functional classes that indicate the degree of the heart failure in a subject.19,41 In normal subjects RR intervals Poincaré plot typically appears as an elongated cloud of points oriented along the line-of-identity. For short time recordings, the dispersion of points perpendicular to the line-of-identity reflects the fast beat-to-beat variability in the data.40

On the Poincaré plots of RR and VV intervals fast beat-to-beat variability was quantified by computation of the standard deviation (SD1) of the distances Di of the points from the line-of-identity,1,40 expressed as:
$$ D_i \, = \,\frac{{\left| {\left[ {y_i - (mx_i + q)} \right].\left[ {\frac{{y_i - q}} {m} - x_i } \right]} \right|}} {{\left( {\left[ {y_i - \left( {mx_i + q} \right)} \right]^2 + \left[ {\frac{{y_i - q}} {m} - x_i } \right]^2 } \right)^{1/2} }} $$
(4)
where (xi, yi) are the coordinates of a point in a Poincarè map, and = 1 and = 0 are parameters describing the line-of-identity (angular coefficient and x axis intercept, respectively).

Results

Figure 3 is an example of simultaneous ECG and VCG recordings. VCG peaks measured with the laser Doppler vibrometer occur during systole, and the R-wave interval measures inter-beat period. Figure 3 shows also the preliminary step for data analysis, peak detection (as mentioned above). Moreover, it can be noticed the effect of the amplitude modulation in the VCG signal, due to the breathing activity.
https://static-content.springer.com/image/art%3A10.1007%2Fs10439-006-9202-9/MediaObjects/10439_2006_9202_Fig3_HTML.gif
Figure 3.

Example of 10 s unaveraged continuous recording of the VCG and ECG. Common features can be noticed in the magnification of two beats. R wave intervals on ECG, together with the corresponding V waves intervals on VCG are also shown.

Figure 4 shows a typical VCG beat together with the II-lead output of the ECG, synchronously recorded. Diastolic and systolic ECG points and events are identified. Reminding the basic hypothesis that the chest wall movement is responsive to changes in myocardial contraction induced by electrical activation, events on VCG recording will be delayed, with respect to ECG. Under this realistic assumption, after a visual inspection of the traces, the point selected to test chest wall vibratory signal capability to evaluate HR and HRV is the point labeled V in Fig. 4: this fiducial point is the first local maximum value in the VCG trace that follows the R-wave maximum in the ECG trace.
https://static-content.springer.com/image/art%3A10.1007%2Fs10439-006-9202-9/MediaObjects/10439_2006_9202_Fig4_HTML.gif
Figure 4.

Example of a one beat vibrocardiogram (VCG), together with the simultaneously recorded ECG II-lead output (time duration 0.80 s). Individual components of the cardiac cycle are labeled on the ECG trace. On VCG, the point chosen to test chest wall vibratory signal capability to evaluate HRV, was labeled (V).

Although there are minor differences among subjects, resting VCGs show common morphologic features, in particular the shape of the “so called” W complex, following QRS in the synchronous ECG recording: this can be noticed (circled) in Fig. 5, where two segments (time duration 400 ms, 160 ms before, 240 ms after R wave) randomly selected from the 5 min recordings of two male and two female subjects are shown. Notwithstanding the presence of inter-variability, there is a recognizable VCG morphology: all the components of the vibratory signal are present for all the monitored subjects, both females and males, and common features to all can be noticed. These features can be identified by: a local maximum (1) labeled H followed (2) by a steep transition to a local minimum K; (3) a rapid raise to the local maximum V (i.e., the fiducial point), followed (4) by a transition to the local minimum J coming before the slow rise (in the time interval between the end of the QRS complex and the onset of the T wave on the ECG trace) of the VCG signal.
https://static-content.springer.com/image/art%3A10.1007%2Fs10439-006-9202-9/MediaObjects/10439_2006_9202_Fig5_HTML.gif
Figure 5.

Two segments (time duration 400, 160 ms before, 240 ms after R wave) randomly selected from the 5 min recordings of two males (number 1 and 2) and two females (number 6 and 8) subjects. There is a recognizable “W shape” VCG morphology (circled) common to all. These features can be identified by a local maximum (H) followed by a steep transition to a local minimum (K), a rapid raise to the local maximum (V) (fiducial point), then by a transition to the local minimum (J) coming before the slow rise of the VCG signal (between the end of the QRS complex and the onset of the T wave on the ECG trace).

Figure 5 also shows the most significant differences in VCG traces due to gender: females exhibit H peak values always higher than the peak labeled V, the opposite being in males recordings.

Noticeable, great similarity in the morphology of the recorded VCG signals can be observed with another method for displacement cardiography, i.e., the contact procedure named seismocardiography.33

As an example of the time series extracted from synchronous ECG and VCG recordings, Fig. 6 shows the RR and VV tachograms relative to one subject: notably, there is high coincidence in the trend exhibited by the two time series.
https://static-content.springer.com/image/art%3A10.1007%2Fs10439-006-9202-9/MediaObjects/10439_2006_9202_Fig6_HTML.gif
Figure 6.

Example of time series (relative to the monitored subject labeled with number (1) built up from the time intervals between consecutive R peaks (RR), and VCG time intervals from the thorax (VV). Right side depicts a detail (50 beats) to show the close relationship of the two time series.

To compare the methods for cardiac rate assessment, i.e., the gold standard ECG and VCG, we applied the Bland–Altman test on the 10 beats-to-10 beats VCG rate (VV) and heart rate (RR) segment values, averaged over the entire 5 min recording length. Figure 7 shows the Bland and Altman plots for two individuals investigated (a male and a female). Table 1 summarizes the results of the Bland and Altman test in terms of bias and of the Pearson’s product moment correlation, R, which measures the strength of linear relationship between the two sets of data: provided differences within 2 standard deviations from the mean difference interval are not clinically important, for all individuals. Table 1 confirms the absence of a consistent bias (0.015 bpm mean bias for females, 0.006 bpm mean bias for males), no proportional error in cardiac rate assessed by VCG, with respect to ECG (linear relationship between means and differences assessed by very low R values both for females and males), and no dependence of the difference between the two methods by the beat duration. Results from Table 1 and Fig. 7 state that we could use the two measurement methods interchangeably, for cardiac rate monitoring.
https://static-content.springer.com/image/art%3A10.1007%2Fs10439-006-9202-9/MediaObjects/10439_2006_9202_Fig7_HTML.gif
Figure 7.

Bland–Altman plots relative to two monitored subjects (2 males, 9 females), to check differences in the assessment of the cardiac rate between ECG and VCG methods. The test was carried out on 10 beats-to-10 beats VCG rate (VV) and heart rate (RR) segment values, averaged over the entire 5 min recording length. Unit measure for both the axis is bpm. (continuous line) mean; (dotted line) ± 2SD; (dashed line) linear regression line.

TABLE 1.

Results of the Bland–Altman test in terms of bias and of the Pearson’s product moment correlation, R, which measures the strength of linear relationship between the two sets of data.

Subject number

Gender

BIAS (bpm)

R

1

M

0.0118 ± 0.118

0.087

2

M

0.003 ± 0.040

0.127

3

M

0.001 ± 0.087

−0.229

4

M

0.010 ± 0.053

0.260

5

M

−0.005 ± 0.084

−0.206

Mean valuea (M)

0.006

/

6

F

−0.033 ± 0.162

−0.159

7

F

0.000 ± 0.044

0.234

8

F

−0.000 ± 0.046

0.019

9

F

0.001 ±  0.041

0.138

10

F

0.041 ±  0.149

−0.114

Mean valuea (F)

0.015

/

We remind that for each individual we applied the Bland–Altman method on the 10 beats-to-10 beats VCG rate (VV) and heart rate (RR) segment values. We can observe absence of a consistent bias, no proportional error in cardiac rate assessed by VCG, with respect to ECG, and no dependence of the difference between the two methods by the beat duration.

aThe mean value of the percent differences is calculated on absolute values.

Table 2 summarizes the comparison (relative to all the monitored subjects) between the RR and VV time series in terms of: mean, standard deviation (SDNN), coefficient of variation (CV), and the root mean square of successive differences (Xi − Xi+1) in the time series, labeled RMSSD. As reported in the guidelines,37 these descriptors reflect all the cyclic components responsible for the variability in the period of recording. The results summarized in Table 2 state that time descriptors of HRV calculated on the VV time series furnish the same information as the RR time series from ECG, being their percent differences very low. As for the mean values of the time series, values less than 0.02% (0.01% mean value for male subjects, 0.01% for females) were obtained; for SDNN, values less than 0.63% (0.38% mean value) were obtained in male subjects, values less than 2.55% (1.25% mean value) for females. On RMSS we computed percent difference less than 5.53% (1.05% mean value for male subjects, 2.80% for females), while values lower than 2.6% were obtained for the coefficient of variation CV (1.27% and 0.38% mean value for females and males, respectively).
TABLE 2.

Mean values of RR and VV time series, for all monitored subjects.

Subject number

Gender

RR mean (ms)

VV Mean (ms)

MEAN % difference

SDNN RR (ms)

SDNN VV (ms)

SDNN % difference

RMSSD RR

RMSSD VV

RMSSD % difference

CV RR

CV VV

CV % difference

1

M

692.54

692.39

0.02

54.49

54.59

−0.19

37.05

37.53

−1.28

7.87

7.88

−0.21

2

M

810.33

810.26

0.01

45.15

45.43

−0.63

39.77

40.20

−1.13

5.57

5.61

−0.63

3

M

743.63

743.59

0.00

74.78

74.43

0.47

70.02

68.97

1.50

10.06

10.01

0.47

4

M

1014.20

1014.1

0.01

85.35

85.19

0.19

81.56

80.74

1.01

8.41

8.40

0.18

5

M

987.92

988.05

−0.01

95.46

95.86

−0.42

94.47

94.79

−0.33

9.66

9.70

−0.40

Mean valuea (M)

849.72

849.68

0.01

71.05

71.10

0.38

64.58

64.45

1.05

8.31

8.32

0.38

6

F

775.88

776.13

−0.03

122.55

122.22

0.27

123.31

122.95

0.29

15.79

15.74

0.30

7

F

719.90

719.91

0.00

38.26

38.71

−1.17

26.30

26.51

−0.80

5.31

5.38

−1.17

8

F

881.70

881.68

0.00

49.66

50.48

−1.65

34.05

35.94

−5.53

5.63

5.72

−1.65

9

F

726.82

726.81

0.00

38.82

39.06

−0.61

25.55

26.23

−2.68

5.34

5.37

−0.61

10

F

694.35

694.02

0.00

50.88

52.18

−2.55

31.83

33.32

−4.70

7.33

7.52

−2.60

Mean valuea (F)

759.73

759.71

0.01

60.03

60.53

1.25

48.21

48.99

2.80

7.88

7.95

1.27

Standard deviations (SDNN), coefficient of variations (CV) and the root mean square of successive differences in the time series (RMSSD) are also summarized, together with their percent differences. For each quantity, the mean value is presented. (F = female; M = male).

aThe mean value is calculated on absolute values of the percent differences.

As for the spectral analysis on HRV, PSDs of RR and VV time series were calculated for each individual: Fig. 8 shows PSDs relative to a male and a female subject, as an example. Table 3 summarizes the values of HRV indices calculated from the spectral analysis on PSDs of VCG and ECG derived time series: total spectral power (PS), Shannon channel entropy E (equation 3), and LF/HF spectral powers ratio were calculated for all the monitored individuals. On male subjects, percent differences less than 4.80% (3.71% mean value) were obtained for LF/HF spectral power ratio, differences less than 3.32% (1.09% mean value) were obtained for PS, and difference less than 1.30% (0.65% mean value) were obtained for estimated spectral entropy E. Female subjects exhibited percent differences less than 4.61% (2.90% mean value) for LF/HF spectral power ratio, differences less than 5.88% (2.82% mean value) for PS, and difference less than 4.00% (1.25% mean value) for E. A positive value in the percent difference means that the index computed on the VV time series is greater than the corresponding one computed on the RR time series; a negative value means that the index computed on the VV time series is lower than the corresponding one computed on the RR time series.
https://static-content.springer.com/image/art%3A10.1007%2Fs10439-006-9202-9/MediaObjects/10439_2006_9202_Fig8_HTML.gif
Figure 8.

Spectral analysis: example of Power spectral density (Hz−1) relative to a female and a male subject (labeled with numbers 1, 7, respectively). For each individual, PSD derived from RR and VV time series are shown. The y-axis scale is not equal for the two subjects, in order to make the results more readable.

TABLE 3.

Spectral indices of HRV calculated on RR and VV time series. The table summarizes the values of Total spectral power, Shannon channel entropy E, and LF/HF spectral powers ratio calculated on all the monitored subjects, For each quantity, the mean value is presented, together with percent differences . (F = female; M = male)

Subject number

Gender

LF/HF RR

LF/HF VV

LF/HF % difference

Total spectral power (SP) RR

Total spectral power (SP) VV

Total spectral power (SP) % difference

Spectral entropy (E) RR

Spectral entropy (E) VV

Spectral entropy (E) % difference

1

M

1.13

1.11

1.53

2.39 e−03

2.39 e−03

−0.14

−904.49

−901.92

0.28

2

M

1.85

1.94

−4.46

1.83 e−03

1.85 e−03

−0.68

−386.29

−383.29

0.78

3

M

1.41

1.47

−4.05

10.00 e−03

9.89 e−03

1.15

−398.72

−403.87

−1.29

4

M

1.74

1.81

−3.73

2.64 e−03

2.63 e−03

0.15

−339.88

−342.5

−0.77

5

M

1.10

1.04

4.80

3.44 e−03

3.56 e−03

−3.32

−311.81

−311.33

0.15

Mean valuea (M)

1.45

1.47

3.71

4.06 e−03

4.06 e−03

1.09

−468.24

−468.58

0.65

6

F

1.55

1.49

3.44

2.30 e−03

2.30 e−03

−0.31

−316.72

−315.94

−0.25

7

F

10.73

13.31

−4.61

2.06 e−03

2.11 e−03

−2.77

−515.19

−518.11

−0.57

8

F

3.42

3.45

−0.88

2.11 e−03

2.17 e−03

−3.22

−468.05

−466.18

4.00

9

F

3.57

3.61

−1.09

2.32 e−03

2.36 e−03

−1.90

−450.56

−447.36

0.71

10

F

11.09

10.60

4.47

3.23 e−03

3.42 e−03

−5.88

−504.14

−507.92

−0.75

Mean valuea (F)

6.07

6.49

2.90

6.46 e−03

6.84 e−03

2.81

−450.93

−451.10

1.25

aThe mean value is calculated on absolute values of the percent differences.

TABLE 4.

Fast variability index SD1 calculated on Poincaré plots from ECG and VCG derived time series, for all the monitored subjects (labelled from 1 to 5 for males, and 6 to 10 for females).

Subject number

Gender

RR SD1 (ms)

VV SD1 (ms)

SD1 % difference

1

M

19.024

19.31

−1.52

2

M

16.44

16.98

−3.30

3

M

38.09

37.47

1.64

4

M

43.51

42.80

1.63

5

M

53.91

54.18

−0.49

Mean valuea (M)

34.20

34.15

1.72

6

F

83.93

83.75

0.21

7

F

10.83

10.93

−0.95

8

F

15.41

16.02

−4.00

9

F

11.04

11.06

−0.23

10

F

14.71

14.83

−0.83

Mean valuea (F)

27.18

27.32

1.24

For each quantity, the mean value is presented together with percent differences (F = female; M = male).

aThe mean value is calculated on absolute values of the percent differences.

Figure 9 shows the Poincaré plots (Xn+1 vs. Xn) both for RR and VV time series, relative to two monitored subjects, a female (subject 7) and a male (subject 1). Being Poincaré plot analysis a quantitative-visual technique, it is evident from a visual inspection of Fig. 9 that Poincaré maps from VCG and ECG assume the same, overlapped shape, i.e., an elongated cloud of points oriented along the line-of-identity. As for the characterization and quantification of Poincaré maps, the SD1 index (equation 4) was calculated, this last being a synthetic descriptor of the non-linear content in the time series. Table 4 summarizes the percent differences in SD1 between VCG and ECG derived sequences, for all the monitored subjects: very small percent differences (RR reference values, as mentioned above for Table 1) in the computed fast variability index between Poincaré plots from ECG and VCG (less than 3.30% for males, less than 4.00% for females) can be observed. The results in Table 4 state that the non-linear, chaotic content of the VV time series extracted from VCG was the same as the RR time series from ECG.
https://static-content.springer.com/image/art%3A10.1007%2Fs10439-006-9202-9/MediaObjects/10439_2006_9202_Fig9_HTML.gif
Figure 9.

Examples of Poincaré plots (Xn+1 vs. Xn) both for RR and VV time series, relative to two monitored subjects. The identity line is also showed, used for the quantification of the Poincaré plots (the SD1 index describing short term variability was computed using the identity line as a reference).

No statistical difference was found between ECG and VCG derived indices, over the 10 subjects investigated: MEAN, SDNN, RMSSD, SD1 (p < 0.88), CV (p < 0.82), LF/HF (p < 0.94), SP (p < 0.76), E (p < 0.94).

Difference in the results due to gender was also checked: no statistical significant difference was found between the mean values of the percent differences found for HRV descriptors relative to males and females (p < 0.172).

The herein shown results clearly demonstrate the relationship between the selected events, i.e. the relationship between the R peak on the ECG, and the V peak on the vibratory signal from the thorax. The calculation of HRV descriptors (both linear and non-linear, in the frequency and in the time domain) on the time series from synchronous VCG and ECG recordings confirm their equivalence: VCG may take the place of electrocardiographic HRV analysis without information loss.

Discussion

Although measurement of RR intervals of ECG is the standard for HRV analysis, this method has practical limitations.21 Particular emphasis we wish to put in the limitations inherent to the use of standard ECG instrumentation for those cases in which electrodes and cables cannot be easily used. Moreover, the risk of interference with other biomedical instrumentation (such as defibrillators, electrical surgical units, magnetic resonance instrumentation) operating together with ECGs is always possible.

Due to the challenges and limitations in the employment of standard ECG in specific conditions, and being HRV a very powerful tool in the assessment of cardiovascular disease, several functional information could be achieved by non-contact cardiac monitoring, once its capability to investigate the simpato-vagal control from the neural structure of a patient is demonstrated. To overcome, fully or partially, the inherent limitations of electrocardiographic technique, analysis of cardiac rate from pulse wave signal was proposed as a potential surrogate of HRV analysis by ECG,7 in virtue of the wide use of pulse wave equipments both in hospital cares and clinical homecare practices. However, it was assessed that the smoothed morphology of pulse waves precludes accurate measurement of pulse-to-pulse intervals such as those used for measuring RR intervals in ECG recordings. Recently, Matsui and collegues23,24 proposed a non-contact method using a microwave radar to monitor the heart and respiratory rates of a healthy person placed inside an isolator or of experimental animals exposed to toxic materials.

Laser based vibration measurement could be of primary interest in this field. Due to the non-contact nature of the optical probe, laser techniques have a series of undoubted advantages, offering interesting perspectives of progress for vibration measurements in terms of innovative applications all the times a non-contact monitoring is the best or the needed choice.

In this paper, the authors propose a novel optical measurement procedure, based on the remote measurement of chest wall movement, to be used as a valid substitute of standard ECG in the assessment of the cardiac rate and in the analysis of heart rate variability, both in daily life and in “polluted environments”. An advantage of this approach is that it could provide a non-contact method to measure the compression waves that, generated by the heart during its movement excited by electric depolarization waves, are transmitted to the chest wall. The herein presented study moves in the same direction of non-invasive/non-contact methods recently proposed 4,11,23,24 for monitoring the cardiac activity.

Is Optical VCG Suitable for HR and HRV Analysis?

In clinical comparison between ECG and optical VCG techniques for the assessment of the vital sign monitoring (the actual cardiac rate), it is necessary to see whether they agree sufficiently for the new to be equivalent to the old: the results of the test of Bland–Altman put in evidence that significant differences are not present, from a clinical viewpoint.

In this study it was also demonstrated that, at present, optical VCG is suitable for HRV analysis. To do this, time- and frequency-domain measures of HRV were calculated as recommended by the Task Force of the European Society of Cardiology,21 and compared on ECG and optical VCG simultaneous recordings, as by Migliaro et al. 25

It is opinion of the authors that a comparison with basic ECG recording or RR intervals time series only is not sufficient to evaluate the sensitivity of optical VCG to perform HRV, which represents an added value for the proposed technique. In fact, in the analysis of method comparison data, neither the correlation coefficient between RR and VV time series nor techniques such as regression analysis are completely appropriate.5 In the same way, a comparison based only on mean values and standard deviations of the RR and VV time series is not sufficient to test the equivalence of the methods in terms of HRV analysis. Being this true, in general, for ECG based HRV analysis (where SDNN was found to be insensitive to specific physio-pathological conditions, i.e., two subjects with the same SDNN may have different clinic reference frames), we think that the same in deep analysis has to be performed when a new method for HRV has to be tested, even if this implies the calculation of several HRV descriptors, covering the quantification of both linear and non linear content, in the time and in the frequency domain. In fact, it may be possible that RR and VV time series have identical mean, standard deviation and ranges, but different autocorrelation functions and therefore different power spectra: this was clearly showed by Kaplan.16

The HRV measured using VCG agreed, with high coincidence, with the HRV derived from an ECG record. In fact, results showed mean percent differences of VCG derived descriptors, with respect to ECG ones, that do not threshold the 4.80% (3.03% mean value) for the LF/HF index, the 4.00% (1.48% mean value) for the fast variability index, the 4.00% (0.93% mean value) for the spectral entropy, the 2.55% (0.81% mean value) for the SDNN index, the 0.03% (0.01% mean value) for the mean value of the time series, the 2.60% for the CV (0.83% mean value), and the 5.53% for the RMSSD (1.93% mean value). The highest percent difference was found in subject 10, that exhibited a 5.88% difference in the spectral power values, but for this HRV descriptor a very low 1.95% difference mean value was found.

The similarity for the times series from ECG and VCG, is due to the capability of the laser based method to catch features of the cardiac displacement that are directly related to the electric activity recorded by ECG. This is clearly testified by the VCG and ECG synchronous recordings from which the time series VV and RR are extracted. Our analysis reveals small differences causing small variations in the indices calculated. However, these small differences in the indices can be considered to specify the same synoptic picture for each subject investigated: HRV analysis between normal and pathologic subjects shows sharp-cut differences in the quantitative indices.

No significant differences between ECG and VCG analysis were obtained, in terms of gender. Males and females showed similar very low percent differences in the values of HRV descriptors calculated by means of VCG and ECG, even if higher differences were found for females, with respect to males. This is probably due to the specific site on the thorax chosen in the present study: in general different local chest wall movements could be expected for females in regions of the chest walls where anatomic features differences between genders are generally pronounced.

These observations indicate that the VCG provides a reliable assessment of cardiac rate and HRV, suggesting that this methodology is potentially useful for assessing cardiovascular variability and dynamics.

Fields of Application of the Technique

Optical VCG seems to overcome the inherent limitations of previously proposed displacement cardiography methods for the examination of the cardiovascular dynamics, the main limits to the diffusion of such methods being the feasibility of the technique (kinetocardiography, balistocardiography), the low sensitivity of the device and the influence of the artefacts on the acquired signal (apexcardiography), the influence of environmental noise (ballistocardiography, apexcardiography), and the contact nature of the sensing technique (balistocardiography, kinetocardiography and seismocardiography). Therefore, a potential field of application for the proposed technique could be the ones were displacement cardiography methods were applied.

The procedure can easily be operated without applying electrodes, avoiding any contact between sensor and patient, therefore resulting suitable to monitor the cardiac activity of subjects under specific conditions.

For example, VCG could be used to monitor HRV without touching the patient, thus preventing secondary exposure of medical personnel to toxic materials under biochemical hazard conditions (e.g., toxic vapors and infectious bacteria). Alternative to the apparatus recently proposed by Matsui and collegues23,24 (a non-contact method using a microwave radar to monitor the heart and respiratory rates of healthy people), the technique appears promising for future pre-hospital monitoring of nerve gas victims, septic patients or in predicting multiple organ dysfunction syndrome patients, thus reducing the risk of secondary exposure in the case of large-scale disasters.

Moreover, being the optical sensor intrinsically insensitive to electromagnetic interferences, optical VCG could offer the benefit of immunity to this kind of noise, as well as avoiding the generation of spurious artefacts in an MR scanning environment, as suggested by Agousti et al.4

Concerning MRI, in general practice ECG monitoring is considered safe if the setup is MR compatible and the electrodes are fixed in a proper way. However, the literature on ECG monitoring during MRI contains little information other than the guideline to avoid loops in the leads and conductive loops formed by conductors with skin contact.14 Recently, Kugel et al.18 demonstrated that there is a potential hazard during examinations of patients with attached or implanted long conductors: high voltages can be induced in straight conductors without loops as ECG cables by coupling with the electric component of the HF field in the MR bore. Local heating or sparking can cause an open flame at the position of the electrodes, and this danger exists even with ECG equipment that is specifically marked as MR compatible.

Limitations in monitoring ECG in interventional and intraoperative magnetic resonance imaging are also pointed out by Kettenbach et al.,17 which state that is difficult to achieve true lead I, II, and III waveforms owing to the close placement of the electrodes that is necessitated by their use in the magnetic field.15 Furthermore, the magnetohydrodynamic effect of blood flowing through the heart creates an ECG artifact.

The optical VCG technique could be alternative to ECG in MRI practice, thus avoiding both the artefacts and risks mentioned above, that are primarily due to the presence of cables in the MR bore.

The technique may be used also to monitor serious thermal and all electrical burns, because the placement of ECG leads may be difficult when there are extensively burned areas. In fact, as ECG leads do not adhere to burned tissues, they must be placed in non-traditional areas (e.g., anywhere not burned, such as the back or legs), and if there are no unburned areas of the body, ECG leads can be placed in the customary positions and held in place with a wrap of gauze around the chest. Alternatively, in the emergency departments a more invasive technique such as stapling ECG leads in place, may be used, with appropriate systemic analgesics to be administered before this procedure31. Also in this case, the possibility to perform the non-contact monitoring of the cardiac activity could be relevant for patients health.

Limitations of the Study

A limitation to the application of the optical VCG may be caused by the physiological differences between VV and RR interval. Theoretically, the variability of the vibratory rate is the sum of the variability existing in RR interval and in the velocity of propagation of the mechanical pulse through the chest. Although the present study indicates usefulness of the VCG for assessing cardiac rate and good agreement in HRV analysis in short recordings at rest, its usefulness as a surrogate of ECG for assessing autonomic functions and mortality risk need to be examined more in deep.

Notwithstanding the use of adhesive retro-reflecting tape applied on the chest wall, we think that the dimension of the tape we used in our study (3 × 2 mm2, weight <1 g) does not represent a limitation in the application of the method. To confirm this, it was recently demonstrated35 that the effect of the skin surface on the vibratory signal is an amplitude reduction with respect to the optical VCG traces measured from an “optimal” surface, i.e. the retro-reflective tape applied on the skin. Moreover, optical VCG traces measured directly from skin are affected by drop-out, but a strategy for vibrometric signal filtering including the implementation of tracking filters allows drop-out effects avoidance without any loss of information.35

The main limitation to the application of the optical VCG, if performed using commercial laser Doppler vibrometry equipments, lies in the fact that it is an expensive technology. The authors choose to make use of a commercial laser Doppler vibrometer for their investigation, because of its very high sensitivity for vibratory velocity and displacement measurements at long operative distances (some meters). We considered this a conservative choice, the aim of the study being the feasibility of an optical, non-contact, from a distance HRV analysis, which needs signals with SNR the highest, to be performed. However, a posteriori considerations on the quality of the vibratory signal recorded from the chest wall35 allow us to affirm that lower cost optical sensors are available for optical VCG, with costs comparable to those of commercial ECG devices.

Future Aims

Further improvement for the proposed VCG application is in the interpretation of the vibratory signal in terms of cardiac mechanics, in order to correlate each morphologic feature in the VCG to specific events in the cardiac cycle for a wide range of physio-pathological conditions. In future, goals of the authors will be the evaluation of other optical sensors, and the evaluation of the hypothesis that the sole optical VCG signal could furnish the same informative content that is now obtained by three distinct physiological measurements, i.e., breathing, arterial pressure and electrical activity of the heart. In fact it could be thought that several functional information could be achieved from the optical signal we carried out from the vibrometer, and relevant patient’s condition related to the pathology could be obtained by means of biomechanical studies.

Conclusions

We proposed a new technique (VCG) capable to perform non-contact monitoring of the cardiac rate using an optical method. The purpose of this study was to evaluate the feasibility of the technique, and to test its sensitivity for HR monitoring and HVR analysis.

Our analysis revealed that:
  • - cardiac rate assessed by VCG agreed with heart rate measured by ECG;

  • - HRV descriptors calculated on the VV time series, both in the time and frequency domain, show values equivalent to those given by HRV analysis on the RR time series.

VCG is fast and easy to perform and recordings are easily reproducible. VCG holds promise for being a useful and powerful tool for non-contact monitoring of cardiac activity of subjects under specific conditions. Clinical studies are currently in progress to evaluate the usefulness of VCG in cardiac resynchronization therapy. Finally, the authors think that using this novel non-contact application in fields such as MRI, or in “polluted environments”, a risk minimization can be reached both for patients and operators.

Acknowledgments

The authors wish to thank Professor Enrico Primo Tomasini and Dr. Giorgio Corbucci for the useful comments and the endless support to the research activity.

Copyright information

© Biomedical Engineering Society 2006