Arterial pulse waves measured with EMFi and PPG sensors and
comparison of the pulse waveform spectral and decomposition analysis in healthy subjects
Matti Huotari1, Antti Vehkaoja2, Kari Määttä1, Juha Röning1
1 Oulu University, Oulu, Finland, 2 Tampere University of Technology, Tampere, Finland
Matti Huotari, Oulu University, Oulu, FINLAND. Email: matti.huotari@ee.oulu.fi.
Abstract
The purpose of this study is to show the time domain and frequency domain analysis of signals recorded with Elec‐
tromechanical Film (EMFi) and Photoplethysmographic (PPG) sensors in arterial elasticity estimation via pulse wave decomposition and spectral components obtained from left forefinger, wrist, and second toe arteries. ECG and pulse waves from the subjects were recorded from 7 persons (30‐60 y) in supine position. Decomposition of the pulse waves produces five components: percussion, tidal, dicrotic, repercussion, and retidal waves. Pulse wave decomposition parameters between EMFi and PPG are compared to detect variables for information on person’s arterial elasticity. Results show that elasticity information in the form of pulse wave decomposition from PPG and EMFi waves is obtainable and shows clear shortening between percussion wave and tidal wave peak time in PPG waveforms with age. The spectral information obtained with frequency domain analysis could also be valuable in assessment of the arterial elasticity. In addition, both PPG and EMFi measurements are absolutely non‐invasive and safe. In PPG measurement, the sensors are on the opposite sides of the finger tip, however, EMFi measure‐
ment needs the good skilled operator attaching the sensor on the patient’s wrist by touching gently to obtain accu‐
rate waveforms.
Keywords: arterial elasticity measurement, electromechanical film (EMFi), photoplethysmography (PPG), pulse wave decomposition
Introduction
Arterial pulse wave is defined as a heart‐beat driven wave of blood that propagates via each artery into vein through capillaries. Various kind of medical information can be obtained by the detection of pulse wave and its detailed analysis. This information is not obtainable from the blood pressure or electrocardiographic measure‐
ments. It could be possible to make early diagnosis on the basis of the pulse wave analysis. Especially atherosclero‐
sis is the main cause of circulatory diseases. Its quantitative assessment is essential for making an early diagnosis of such diseases. In addition, persons with other cardiovascular diseases (CVD) may have decreased arterial elasticity compared with those free of CVDs. Change of arterial elasticity is one of the early markers of accelerated arterial aging and can correlate with many coronary risk factors. Especially arterial elasticity reflects the arterial and aortic expanding during left ventricular contraction. Arterial elasticity can be measured indirectly provided the measure‐
ment method is relevant and accurate enough. In this study, biophysical function and structure of the arteries have been measured by photoplethysmographic (PPG) and electromechanical film (EMFi) sensors. For the measurement data, we applied pulse wave decomposition (PWD), which reflects clearly the elasticity of the aorta and its periph‐
eral arteries. The combined PPG & EMFi measurements can establish aortic and arterial elasticity based on PWD of the both signals during a heart cycle. We do not necessarily need distance measurement required for pulse wave velocity estimation, which can be rather inaccurate in the case of the arterial tree [1]. The EMFi and PPG technolo‐
gies require a few electromechanical or opto‐electronic components: a EMFi film sensor connected to an amplifier, and a light source to illuminate the tissue (e.g. finger), and a photodetector to measure the small variations in light absorbance amplified also by a transimpedance amplifier. The obtained pulse wave is the peripheral pulse that can be decomposed into five logarithmic normal components. Despite its simplicity, the origins of the different com‐
ponents of the PPG or EMFi signal are not fully understood [2]. We believe that these parallel pulse waves can provide valuable information about the circulatory system with patient‐friendly and safe means.
In the elderly and in the stiffened arteries, the forward and reflected pulse waves travel faster, i.e., pulse wave velocity (PWV) is higher than in a young person’s arteries which are elastic. The arterial waves reflected from the periphery of the arterial tree, return earlier merging with the systolic part of the incident wave causing augmenta‐
tion of the workload of the heart. Favorable softness between coupling of the left ventricle and the arterial tree is thus progressively lost. This loss can be greatest in the aorta, and least in the upper limbs. The wave reflection of the pulse wave due to increase in PWV also increases with age, but can be largely prevented by physical activity and proper diet. The amplitude spectrum of the ECG, EMFi, and PPG are changing by so‐called integral pulse fre‐
quency modulation (IPFM) [3]. This modulation is caused by the autonomic control mechanisms of cardiac func‐
tions which are involved in short‐term fluctuations in the time interval between the consecutive heart cycles. The IPFM reflects cardiac function which is also detected in the periphery of the arterial tree. Healthy modulation in coupling of the left ventricle and the arterial tree is, however, progressively lost.
In our study we measure EMFi, two PPGs, and ECG signals, and we apply Fast Fourier Transform on the signals in addition to logarithmic normal function decomposition of the EMFi and PPG pulse waveforms.
Methods
In this study, the PPG sensors are based on LEDs and photodetector, EMFi sensor is based on a plastic EMFi film, and ECG sensors are standard electrodes. Changes in light absorption, in the pressure, or in electrical potentials are acting on each sensor generating a measurable voltage. The EMFi sensor acts as a sensitive pressure sensor and the PPG sensor as a sensitive absorbance sensor. Signals from ECG, PPG, and EMFi sensors were recorded with the
PC from 7 healthy persons (30‐60 y) using a data acquisition card with the sampling frequency of 500 Hz. ECG sig‐
nal was used as reference in detecting features from the PPG and EMFi related signals. Pulse wave series from EMFi, PPG finger and toe were processed by Origin software as follows. Firstly, the alternating baseline on each signal bottom (minimum) was searched and then subtracted. Secondly, the peak of each signal (maximum) was searched, and the average of the peak values was calculated. The signal was divided by the average to obtain the baseline removed and normalized waveforms which has amplitude from the zero to round about one. Processing continues a pulse by a pulse in the PWD.
Results
An example pulse wave from each sensor is shown in Figure 1 on which analysis in time domain and FFT in fre‐
quency domain was performed. Each pulse wave component from EMFi, PPG finger, and toe were processed by the software. Decomposed pulse waves from the left wrist (EMFi), and from the left forefinger and the left second toe (PPG) are decomposed in time domain and transformed in frequency domain. In Figure 1 it is shown EMFi (solid), PPG finger (dash dot), PPG toe (dashed) pulse waves, and electrocardiogram (ECG) (dot). They have the start time as zero and then decomposed with their residual and confidence intervals, respectively, in the Figure 2.
0,0 0,5 1,0 1,5 2,0 2,5 3,0
-0,2 0,0 0,2 0,4 0,6 0,8 1,0 1,2
t[s]
Amplituderel B EMFi_TTSS_30
D PPG2 (toe) F PPG1 (finger) H ECG
Figure 1. The baseline removed normalized pulse waves: EMFi (left wrist, solid), PPG1 (left forefinger, dash dot), PPG2 (left second toe, dashed), and ECG (dot) (Male 30).
0,0 0,1 0,2 0,3 0,4 0,5 0,6 0,7 0,8 0,9 1,0 0,00
0,25 0,50 0,75 1,00 1,25
0,0 0,1 0,2 0,3 0,4 0,5 0,6 0,7 0,8 0,9 1,0
-0,03 -0,02 -0,010,000,010,020,030,04
0,128
0,288
0,464
0,656
0,8
t [s]
0,6 0,7 0,8 0,9
0,0 0,2
0,6 0,7 0,8 0,9
-0,03 -0,02 -0,010,000,010,020,030,04
0,656
0,8
t [s]
(0,032 s)/2=0,016 s 0,160 s_0,032 s_2,250
0,176 s 0,192 s
0,128 s+0,016 s= 0,144 s Data: Data24_B TTSS_30-EMFI-2 Model: LogNor mal Equation:
y = y0 + A/(sqrt(2*PI)*w*x)*exp(-(ln(x/xc)) ^2/(2*w^2)) Weighting:
yNo weighting Chi^2/DoF= 0.00007 R^2= 0.99899
y00±0
xc10.16787 ±0.00041 w10.57342 ±0.00164 A10.19773 ±0.00031 xc20.29496 ±0.00254 w20.11755 ±0.00795 A20.0053 ±0.00027 xc30.47814 ±0.00355 w30.17313 ±0.00546 A30.07034 ±0.00109 xc40.6713 ±0.00765 w40.12885 ±0.02126 A40.01904 ±0.00192 xc50.806±0.00518 w50.06443 ±0.01196 A50.00423 ±0.00153
t [s]
EMFirel
Resudual
t [s]
(0,032 s)/2=0,016 s 0,160 s_0,032 s_2,250
0,176 s 0,192 s
0,128 s+0,016 s= 0,144 s
Data: Data24_B TTSS_30-EMFI-2 Model: LogNormal Equation:
y = y0 + A/(sqrt(2*PI)*w*x)*exp(-(ln(x/xc))^2/(2*w^2)) Weighting:
y No weighting Chi^2/DoF = 0.00007 R^2 = 0.99899
y0 0 ±0
xc1 0.16787 ±0.00041 w1 0.57342 ±0.00164 A1 0.19773 ±0.00031 xc2 0.29496 ±0.00254 w2 0.11755 ±0.00795 A2 0.0053 ±0.00027 xc3 0.47814 ±0.00355 w3 0.17313 ±0.00546 A3 0.07034 ±0.00109 xc4 0.6713 ±0.00765 w4 0.12885 ±0.02126 A4 0.01904 ±0.00192 xc5 0.806±0.00518 w5 0.06443 ±0.01196 A5 0.00423 ±0.00153
EMFi rel
a
Resudual
0,0 0,1 0,2 0,3 0,4 0,5 0,6 0,7 0,8 0,9 1,0
-0,1 0,0 0,1 0,2 0,3 0,4 0,5 0,6 0,7 0,8 0,9 1,0 1,1
0,0 0,1 0,2 0,3 0,4 0,5 0,6 0,7 0,8 0,9 1,0
-0,050 -0,025 0,000 0,025 0,050
0,14217
0,26854
0,4739
0,64766
0,77403
t[s]
0,0158 s * 2=0,0316 s 0,12637 s_-0,0158 s_1,88887
0,20536 s 0,17376 s
0,12637 s EMFip-PPG1P=0,14217 s - 0,128 s=-0,01417 s
Data: Data34_B TTSS_30PPG1_2 Model: LogNormal Equation:
y = y0 + A/(sqrt(2*PI)*w*x)*exp(-(ln(x/xc))^2/(2*w^2)) Weighting:
y No weighting Chi^2/DoF = 0.00023 R^2 = 0.99707
y0 0 ±0
xc1 0.26002 ±0.00431 w1 0.79597 ±0.00635 A1 0.34821 ±0.00451 xc2 0.27935 ±0.00414 w2 0.2149 ±0 A2 0.02029 ±0.00115 xc3 0.4971 ±0.0093 w3 0.21556 ±0.01328 A3 0.11562 ±0.00534 xc4 0.6575 ±0.01008 w4 0.11953 ±0.02158
A4 0.03082 ±0.0073
xc5 0.77568 ±0.00501 w5 0.06467 ±0.00868 A5 0.01035 ±0.00307
PPG1 rel
b
t[s]
Residual
Figure 2a) A single EMFi pulse wave (solid, measured) which is here decomposed into components: percussion (dash), tidal (dot), dicrotic (dash dot), repercussion (short dash), and retidal wave (short dot). (Insert the two last wave components). The confidence interval (99%) is marketed short dot dot. The residual curve is shown in the lower panel. b) A single PPG1 (finger) pulse wave decomposed, respectively. (Male 30).
0,0 0,1 0,2 0,3 0,4 0,5 0,6 0,7 0,8 0,9 1,0 1,1 1,2 1,3 0,0
0,1 0,2 0,3 0,4 0,5 0,6 0,7 0,8 0,9 1,0 1,1
0,0 0,1 0,2 0,3 0,4 0,5 0,6 0,7 0,8 0,9 1,0 1,1 1,2 1,3 -0,02
-0,01 0,00 0,01 0,02 0,03
0,14478
0,29356 0,48485
0,76115
1,16498
t [s]
0,14878 s_0,004 s_2,028
0,19129 0,2763
0,40383 PPG1p-EMFip=0,18637-0.14478=0,04159
Data: Data30_B Model: LogNormal Equation:
y = y0 + A/(sqrt(2*PI)*w*x)*exp(-(ln(x/xc))^2/(2*w^2)) Weighting:
y No weighting Chi^2/DoF = 0.00005 R^2 = 0.99943
y0 0 ±0
xc1 0.21327 ±0.00091 w1 0.61405 ±0.00213 A1 0.26783 ±0.00078 xc2 0.29341 ±0.00055 w2 0.14128 ±0.00202 A2 0.02649 ±0.00038 xc3 0.49686 ±0.00153 w3 0.19327 ±0.00264 A3 0.0898 ±0.00087 xc4 0.80032 ±0.00409 w4 0.20676 ±0.00389 A4 0.0672 ±0.00069 xc5 1.1614 ±0.00319 w5 0.01339 ±0.00291
A5 0.00052 ±0.0001
EMFi rel
a
t [s]
Residual
-0,1 0,0 0,1 0,2 0,3 0,4 0,5 0,6 0,7 0,8 0,9 1,0 1,1 1,2 -0,1
0,0 0,1 0,2 0,3 0,4 0,5 0,6 0,7 0,8 0,9 1,0 1,1 1,2 1,3 1,4
-0,1 0,0 0,1 0,2 0,3 0,4 0,5 0,6 0,7 0,8 0,9 1,0 1,1 1,2 -0,04
-0,02 0,00 0,02 0,04
0,18241
0,28597
0,5138
0,78305
0,96946
t [s]
-0,04142
0,10356 s_-0,07885 s_1,568
0,22783 0,26925
0,18641
Data: Data2_B PL_60 Model: LogNormal Equation:
y = y0 + A/(sqrt(2*PI)*w*x)*exp(-(ln(x/xc))^2/(2*w^2)) Weighting:
y No weighting Chi^2/DoF = 0.00017 R^2= 0.9982
y0 0 ±0
xc1 0.28169 ±0.00702 w1 0.6794 ±0.00901 A1 0.30214 ±0.00715 xc2 0.29168 ±0.00323 w2 0.20137 ±0 A2 0.03543 ±0.00244 xc3 0.55412 ±0.0201 w3 0.29211 ±0.02518 A3 0.2456 ±0.01681 xc4 0.80841 ±0.03208 w4 0.17232 ±0.03695 A4 0.09407 ±0.03278 xc5 0.98826 ±0.00982 w5 0.10954 ±0.01568 A5 0.0349 ±0.02001
t [s]
PPG1 rel
b
Residual
t [s]
Figure 3a) A EMFi pulse wave (solid, measured) which is decomposed into components: percussion (dash), tidal (dot), dicrotic (dash dot), repercussion (short dash), and retidal wave (short dot). The confidence interval (99%) is marketed short dot dot. The residual curve in the lower panel. b) A single PPG1 (finger) pulse wave decomposed, respectively. (Male 60).
0,0 0,1 0,2 0,3 0,4 0,5 0,6 0,7 0,8 0,9 1,0 0,000
0,001 0,002 0,003 0,004 0,005 0,006 0,007 0,008 0,009 0,010
0,0 0,2 0,4 0,6 0,8 1,0
0,000 0,005 0,010 0,015 0,020 0,025 0,030 0,035
abs [ResidualEMFi]
t [s]
C abs[ResidualEMFi]
t[s]
Integral of residuals
Integral of Data6_C EMFi Integral of Data16_C PPGfinger Integral of Data27_C
Integral of Data37_C Integral of Data48_C Integral of Data58_C
Figure 4. Integral of residuals (EMFi solid) and PPG1 (finger, dash). Insert: Absolute value of residual of EMFi wave‐
form, (Male 30).
-0,1 0,0 0,1 0,2 0,3 0,4 0,5 0,6 0,7 0,8 0,9 1,0 -0,1
0,0 0,1 0,2 0,3 0,4 0,5 0,6 0,7 0,8 0,9 1,0 1,1
-0,1 0,0 0,1 0,2 0,3 0,4 0,5 0,6 0,7 0,8 0,9 1,0 -0,04
-0,02 0,00 0,02 0,04
t [s]
t [s]
Data: Data15_B TTSS_30_PPG2 Model: LogNormal Equation:
y = y0 + A/(sqrt(2*PI)*w*x)*exp(-(ln(x/xc))^2/(2*w^2)) Weighting:
y No weighting
Chi^2/DoF = 0.0001 R^2 = 0.99891
y0 0 ±0
xc1 0.22511 ±0.00103 w1 0.56858 ±0.00198 A1 0.25135 ±0.00199 xc2 0.23982 ±0.00162 w2 0.20243 ±0.00819 A2 0.01611 ±0.00075 xc3 0.60279 ±0.00223 w3 0.20447 ±0.00478 A3 0.03438 ±0.00058
PPG2rel
Residual
Figure 5. The whole single PPG2 pulse wave (black, measured) which is here decomposed into components respec‐
tively as in Figure 2, but here only three components are found. (Male 30, Fig. 1).
0,1 1 10 0,00
0,05 0,10
FFT2_r[4146]: X = 2,991; Y = 0,0328 FFT2_r[4152]: X = 3,357; Y = 0,0379FFT2_r[4149]: X = 3,174; Y = 0,0505
FFT2_r[4135]: X = 2,319; Y = 0,0357
FFT2_r[4128]: X = 1,892; Y = 0,0238 FFT2_r[4132]: X = 2,136; Y = 0,0725 FFT2_r[4111]: X = 0.8545 Y = 0.0271 FFT2_r[4117]: X = 1.2207, Y = 0.026
FFT2_r[4114]: X = 1.038, Y = 0.1178
Frequency (Hz)
Amplitude
TTSS_30_EMFi
-200-150 -100100150200-50500
0,1 Frequency (Hz) 1 10
Angle(deg)
0,1 1 10
0,00 0,05 0,10 0,15
FFT1_r[4146]: X = 2.991, Y = 0.0260 FFT1_r[4152]: X = 3.357, Y = 0.0294FFT1_r[4149]: X = 3.174, Y = 0.0400 FFT1_r[4128]: X = 1.892, Y = 0.0233 FFT1_r[4135]: X = 2.319, Y = 0.0317FFT1_r[4132]: X = 2.136, Y = 0.0700
FFT1_r[4114]: X = 1.0376, Y = 0.1546 FFT1_r[4117]: X = 1.221, Y = 0.0303
FFT1_r[4111]: X = 0.854, Y = 0.0388
FFT1_r[4101]: X = 0.244, Y = 0.0254
Frequency (Hz)
Amplitude
TTSS_30_PPG1
-200-150 -100100150200-50500
0,1 Frequency (Hz) 1 10
Angle(deg)
Figure 6. The pulse wave of EMFi’s amplitude spectrum (up), and of PPG1’s amplitude spectrum (down) with the IPFM parameter values for the three first components for 20 s record. In the PPG1’s amplitude spectrum contains the breath rate frequency value (Male 30).
0,1 1 10 0,000
0,025 0,050 0,075 0,100 0,125 0,150 0,175
FFT1_r[8201]: X = 0,244; Y = 0,0177 FFT1_r[8265]: X = 2,197; Y = 0,0088
FFT1_r[8233]: X = 1,221; Y = 0,0074
FFT1_r[8217]: X = 0,732; Y = 0,0142 FFT1_r[8250]: X = 1,740; Y = 0,0092 FFT1_r[8290]: X = 2,96020508; Y = 0,0426239233
FFT1_r[8257]: X = 1,953125; Y = 0,074377257
FFT1_r[8225]: X = 0,977; Y = 0,1695
Frequency (Hz)
Amplitude
AAVV_31_EMFi
-200-150 -100100150200-50500
0,1 Frequency (Hz) 1 10
Angle(deg)
0,1 1 10
0,00 0,05 0,10 0,15
FFT2_r[8201]: X = 0,244; Y = 0,0171 FFT2_r[8265]: X = 2,197; Y = 0,0050
FFT2_r[8250]: X = 1,740; Y = 0,0088
FFT2_r[8233]: X = 1,221; Y = 0,0054
FFT2_r[8217]: X = 0,732; Y = 0,0181 FFT2_r[8289]: X = 2,9296875; Y = 0,0274294923
FFT2_r[8257]: X = 1,953125; Y = 0,0580455913
FFT2_r[8225]: X = 0,977; Y = 0,1722
Frequency (Hz)
Amplitude
AAVV_31_PPG1
-200-150 -100100150200-50500
0,1 Frequency (Hz) 1 10
Angle(deg)
Figure 7. The EMFi pulse wave signal’s amplitude spectrum (up), and of PPG1’s amplitude spectrum (down) with the IPFM parameter values for the three first components. In the PPG1’s and EMFi’s amplitude spectrum contain also the breath rate frequency value for 20 s signal record as a sample length (Male 31).
0,1 1 10 0,00
0,05 0,10
FFT3_r[8301]: X = 3,296; Y = 0,0261
FFT3_r[8274]: X = 2,472; Y = 0,0410
FFT3_r[8247]: X = 1,648; Y = 0,0562
FFT3_r[8220]: X = 0,824; Y = 0,1213
Frequency (Hz)
Amplitude
PPLL_60_EMFi
-200-150 -100100150200-50500
0,1 Frequency (Hz) 1 10
Angle(deg)
0,1 1 10
0,000 0,025 0,050 0,075 0,100 0,125 0,150 0,175
FFT5_r[8301]: X = 3,296; Y = 0,0170
FFT5_r[8274]: X = 2,472; Y = 0,0298
FFT5_r[8247]: X = 1,648; Y = 0,0502
FFT5_r[8220]: X = 0,824; Y = 0,1535
Frequency (Hz)
Amplitude
PPLL_60_PPG1
-200-150 -100100150200-50500
0,1 Frequency (Hz) 1 10
Angle(deg)
Figure 8. The EMFi pulse wave signal’s amplitude spectrum (up), and of PPG1’s amplitude spectrum (down). In the PPG1’s and EMFi’s amplitude spectrum contain also the breath rate frequency value (Male 60).
The comparison of the PPG1 pulse waves shows that the tidal wave comes closer to the percussion wave peak value when person’s age increases becoming shorter than the percussion wave defined from the start to the wave maximum (Fig. 2b and 3b). The integrals of residual errors for each sensor and in different measurement show good relation. At the first, the integral of each residual overlaps and then EMFi and PPG based residuals differ after the systolic phase (Figure 4). In Figure 5 it is shown a typical toe PPG pulse wave which contains at least three components according to the decomposition. The comparison of both the EMFi’s and PPG1’s amplitude spectra shows that the modulation frequency disappears or comes close to the carrier frequency as the person’s age in‐
creases (Fig. 6, 7, and 8). This study shows also that both the EMFi and PPG pulse waveforms can be decomposed to their component waves, namely, percussion wave, tidal wave, dicrotic wave, repercussion, and retidal waves.
Also in frequency domain the amplitude spectra of the respective pulse waves contain at least five or six frequency components.
Discussion
Clinical research is necessary to quantify the ageing effects in relation to the obtained variables and also to explore pulse waveform changes with subject age. However, based on these optical and mechanical measurement meth‐
ods it was shown the changes in the pulse waveform. Measurement reliability and repeatability is very good pro‐
vided the EMFi measurements are done by a skilled operator. Pulse waveform analysis of both the PPG and EMFi offers an alternative means of non‐invasive cardiovascular monitoring, but further both software and hardware development is required to enable user‐friendly clinical and preclinical measurement and analysis system. In PPG, the sensors are on the opposite sides of the finger tip, however, EMFi needs the good skilled operator (A.V.) at‐
taching the sensor on the patient’s wrist by touching gently to obtain waveforms. The elasticity is obtained quanti‐
tatively from both EMFi and PPG pulse signals both in time and frequency domain because of the components time interval changes or the spectral changes clearly inspected. However, from time domain, decomposition into loga‐
rithmic normal function, differs from that obtained from frequency domain because the latter contains many pulse waves which are IPFM modulated and they frequency components overlap. This information from the pulse wave propagation analysis can’t be received. In studies, EMFi and PPG theory should be described when analyzing arte‐
rial pulse wave signals because they contain effects of respiration, autonomous nervous activity, gastric mobility, and also arterial properties, i.e., arterial diseases.
References
[1] Alametsä J, Palomäki A. Comparison of local pulse wave velocity values acquired with EMFi sensor. Finnish Journal of eHealth and eWelfare 2012;4(2):89‐98.
[2] Allen J. Photoplethysmography and its application in clinical physiological measurement. Physiological Meas‐
urement 2007;28:R1–R39.
[3] Middleton PM et al. Spectral analysis of finger photoplethysmograpic waveform variability in a model of mild to moderate haemorrhage. Journal of Clinical Monitoring and Computing 2008;22:343‐353.
