Photoplethysmographic signal rising front analysis for the discrimination of subjects with increased arterial ageing

This study analyses the photoplethysmographic (PPG) signal waveform rising front in order to discriminate subjects with premature increase in arterial ageing. As a reference measure for the evaluation of the arterial system, the aortic augmentation index (AIx@75) was calculated using a SphygmoCor device. The study was carried out on 24 healthy subjects and 20 diabetes patients. Negative correlation relationships (r = – 0.68 and r = – 0.77) were found between the age and the slopes of the PPG signal. Significant differences were found between the healthy controls and diabetes patients. The highest difference between the two groups was found using the advanced signal processing algorithm for the slope calculation. The sensitivity, specificity, and accuracy of the method were 85%, 88%, and 86%, respectively. We conclude that the PPG technology with the advanced signal processing algorithm can be used for the discrimination of subjects with increased arterial ageing. This method gives the possibility to diagnose cardiovascular diseases in an early stadium.


INTRODUCTION
* Increased arterial stiffness has been reported as one of the markers of cardiovascular damage [1].Non-invasive methods and devices have been developed to evaluate the arterial system and to diagnose a cardiovascular disease in the early stage [2,3].The SphygmoCor system can be used to estimate the aortic augmentation index (AIx) [4].The AIx has been considered as an independent predictor of mortality and cardiovascular events [5,6].In addition, AIx has been found to increase through ageing [7].The values are higher on the patients with type I diabetes [8] and hypercholesterolemia [9], which can be explained by premature arterial ageing [10,11].However, the SphygmoCor system is operator dependent and the measurements are often time consuming.A simple, inexpensive and operator independent method is needed to assess cardiovascular risk in the screenings of large populations.
The photoplethysmographic (PPG) technique has been considered as a simple and inexpensive technology for pulse wave recording.The analysis of the finger PPG waveform has been used as a non-invasive method to assess arterial stiffness [12,13].PPG is a ubiquitous optical technique, based on the measurement of changes in blood circulation in the vascular bed of a tissue.PPG sensors are simple and inexpensive devices, consisting of a light emitting diode (LED) and a photodetector (PD) [14].In the case of a finger clip sensor, light is emitted from the LED to the tissue and a small fraction of emitted photons is detected by the PD, placed on the opposite side of the finger.The AC component of the PPG signal is synchronous [15] with the heart rate and depends on the changes in the pulsatile pressure and blood volume.
The changes in the AC component waveform of the finger PPG signal, related to the arterial stiffness and cardiovascular ageing, have been investigated by different sophisticated methods in the time and frequency domain [12,[16][17][18].In addition, the systolic and diastolic part of the PPG signal is successfully analysed through the waveform decomposition into Gaussian functions [19].The slope of the PPG signal raising front has been analysed during the cold pressor test in order to monitor cardiovascular and nervous system reactivity [20].In our pilot study we analysed the slope changes, caused by cold and warm stimulation [21].In addition, it was found that the slope of the raising front of the PPG signal depends on the subject's age.It was found that slope of the raising front decreases in proportion to the increase of the subject's age.Similar results were found in the exhaustive study on population of healthy subjects, where the dependence of the slope of the PPG signal raising front on the age was investigated [22].
Focus here is on the analysis of the possibilities to use the slope of the PPG signal waveform for the discrimination of the subjects with accelerated arterial ageing, which is related to the premature increase in arterial stiffness.Compared to the study by Zahedi et al. [22] the signal processing algorithm for the raising front slope calculation is improved for better discrimination of the subjects.In addition, in this study the group of diabetes patients is included in order to test the methodology.The SphygmoCor system was used as a reference method for cardiovascular system age evaluation through aortic AIx calculation.

Subjects
The study was performed in accordance with the Declaration of Helsinki and formally approved by the Tallinn Ethics Committee on Medical Research.The experiments were conducted in the North Estonia Medical Centre.The signals were recorded from 24 healthy subjects (mean age 41 ± 13 years) and 20 type II diabetes patients (mean age 44 ± 14 years).The subjects in the group of healthy controls were not on permanent medication and they were engaged in some physical activity at least once a week.The subjects were in supine position at least 10 min prior to and during the recording of the signals.During the experiments the room temperature was kept constant at around 23 ± 1 °C.

Instrumentation
As a reference, the aortic AIx was estimated with the SphygmoCor (AtCor Medical, Australia) device.The pressure waves were recorded from the left hand radial artery by the sensitive wide band piezoelectric sensor [4].On all the subjects, the operator index above 85 was achieved.The aortic pressure waveform was derived from the radial artery waveform using the validated generalized transfer function [23], illustrated in Fig. 1.The aortic AIx was estimated by SphygmoCor device and the augmentation pressure was divided with the pulse pressure [24], calculated as where s P is systolic pressure, d P is diastolic pressure, and i P is the pressure value at the inflection point, which is detected from the fourth derivative of the pressure waveform [7].According to Wilkinson et al. [25], the aortic AIx depends on the heart rate and the SphygmoCor normalizes the AIx to the heart rate of 75 bpm.The normalized AIx is denoted as AIx@75.
During the experiments, the pressure waveforms were recorded firstly with the SphygmoCor device.After that the physiological signals were recorded by the experimental measurement complex, described in [26].The PPG signal was recorded from the index finger using the commercially available Envitec F-3222-12 finger clip sensor (Honeywell, Germany).The infrared LED of the sensor at a peak value of 880 nm was used.The sensor was connected to the lab-built PPG module, where the current of the LED was controlled and the photocurrent of the PD was converted to the voltage signal.The PPG module was connected to the PCI MIO-16-1 data acquisition card (National Instruments, USA) and the analogue PPG signal was digitized with a sampling frequency of 1 kHz.One minute long PPG signals were recorded from the left hand index finger of the subject.

Analysis of the signals
The recorded PPG signals were post-processed in MATLAB (Mathworks, USA).The PPG signals were filtered with high-and low-pass filters.The cut-off frequencies were 0.5 Hz (order of the filter 4000) and 15 Hz (order of the filter 500), respectively.The filter was designed using the window method, with the Hamming window function.
The slope of the PPG signal rising front was calculated on the basis of the study by Zahedi et al. [22].The first derivative signal [ ]  x n ′ (n is the number of the sample) was calculated from the PPG signal [ ].
x n For each recurrence, the maximal point m was detected from the first derivative signal.In addition, the amplitude of the PPG signal PPG A was calculated for each recurrence (Fig. 2).The slope of the raising front of the PPG signal r S was calculated for each recurrence as For each PPG signal, the average slope rav S and the standard deviation were calculated.
The raising front of the PPG signal depends on the heart rate.With higher heart rate the waveform front tends to increase.The previously described and used algorithm [22] is not taking into account the changes in the heart rate.In this study we used the new PPG signal processing algorithm in order to compensate the effect of the heart rate on the slope calculation.The recorded PPG signals were postprocessed in MATLAB using the algorithm described in [27].The algorithm resamples the signal so that each recurrence in the signal contains an equal number of samples.This means that all the recurrences are normalized to the same length, i.e., to the same heart rate.All the recurrences were normalized to the length of 1 s, which corresponds to the heart rate of 60 bpm.Further, the signal was low-pass filtered using the Parks-McClellan low pass filter.According to our previous study [27], the following filter parameters were used: edge frequency 6 Hz, transition band 1 Hz, pass and stop band maximum allowable errors 0.001.As a result, all the recurrences of the PPG signal were limited equally with the same number of harmonic components.The slope of the raising front, denoted as n , S was calculated for each recurrence using Eq. ( 1).Similarly, the average slope norm S and the standard deviation were calculated to compare the results.

RESULTS
The average heart rates of healthy controls and diabetes patients are 66 ± 6 bpm and 78 ± 15 bpm, respectively.Figures 3a and c show the relationship between the age and the slopes of rav S and norm .

S
The data points are given with standard deviation bars.Similarly, Fig. 3e presents the relationship between the age and AIx@75.For each parameter, the regression lines and models are given.In addition, Pearson's correlation coefficient was calculated for each parameter using data points from healthy subjects.
For clarity, the Bland-Altman plots for each parameter are compared in Figs 3b, d, and f.A paired t-test was carried out in order to analyse the group differences between the healthy controls and diabetes patients.The t-test was carried out in MS Excel (Microsoft, USA) assuming unequal variances with 0.05.

α =
A significant difference was found between the groups of healthy controls and diabetes patients ( 0.0001) P < at AIx@75 and slope norm .S The differences found between the two groups have a 1.4% probability of occurring by chance alone at slope rav .
S Based on the Bland-Altman plot, the sensitivity, specificity, and accuracy for each parameter were calculated and the results are presented in Table 1.

DISCUSSION
Figure 3a and c reveal negative linear relationship between the age and the slopes of the raising front of the PPG signal.This is in accordance with the results from our previous pilot study with a smaller number of subjects [21] and with earlier study by Zahedi et al. [22].
In addition, the positive correlation relationship was found between the age and the reference parameter AIx@75.Similar results were found in [7], and the changes in the AIx due to the age are explained partially through the sclerotic processes in the arterial wall.Figures 3a and c show that the absolute slope values of rav S are higher than those of norm .S According to the average heart rates, in most of the cases, to calculate norm S the recurrences were stretched during the normalization process.This causes a decrease in the slope of the raising front.Partially, the difference in the absolute values of rav S and norm S can be caused by the differences in the bandwidth of the PPG signals.The bandwidth was wider at rav .S Figure 3b shows that the mean difference of the diabetes patients group, who have probable premature increase in arterial stiffness, has noticeably lower norm S values than the group of healthy controls.According to Table 1, the slope of norm S has higher sensitivity, specificity and accuracy compared to the slope rav .
S This shows that to improve the discrimination between healthy subjects and subjects with increased arterial stiffness, the advanced signal processing can be used in the calculation of the raising front slope.Furthermore, the sensitivity, specificity and accuracy of norm S are even higher at the reference methodology calculated parameter AIx@75 (Table 1).

CONCLUSIONS
The possibilities to use the slope of the PPG signal waveform for the discrimination of the subjects with accelerated arterial ageing were investigated.As a reference measure for the evaluation of the arterial system age, the aortic augmentation index was calculated using a SphygmoCor device.It was found that slope of the PPG signal raising front correlates negatively with subject's age.In addition, it was found that the slope is lower in case of diabetes patients, which is related to the accelerated arterial ageing.Two groups in this study are relatively small.However, the results show a possibility to use the inexpensive PPG technology with the advanced signal processing algorithm for the discrimination of subjects with increased arterial ageing.

Fig. 1 .
Fig. 1.Aortic pressure waveform with detected characteristic points P d , P i , and P s for the AIx calculation.

Fig. 2 .
Fig. 2. The PPG signal x[n] (a) and first derivative of the PPG signal x′ [n] (b) with detected sample number m, which marks the detected maximal point of the first derivative of the PPG signal.

Fig. 3 .
Fig.3.The analysis of linear relationship between the calculated parameters and the subject's age.Linear relationships with Pearson's correlation coefficients between the age and the calculated parameters: (a) S norm , (c) S rav , and (d) AIx@75 with the regression models y norm , y rav , and y AGI@75 , respectively.Bland-Altman plots with standard deviation of models: (b) SD norm , (d) SD rav , and (f) SD AIx@75 and the average difference of diabetes group Mean Dia .

Table 1 .
The sensitivity, specificity, and accuracy of the parameters