In this study, PPG datasets of 93 healthy individuals were recorded from the fingertip using TAS9 VIEW (or Canopy9 RSA, IEMBIO Co. Ltd., Chuncheon, Korea). Measurements were taken for 20 seconds at a sampling rate of 1 kHz while the participants were seated at rest. APG signals were obtained by double differentiating the PPG signal. In the PPG and APG signal feature extraction algorithm, 10 APG and 14 PPG indices based on the peripheral signals were simultaneously obtained: a_peak, b_peak, c_peak, d_peak, e_peak, age index, c_position, d_position, bc_slope, and bd_slope for APGs and area A1, area A2, A1_time, A2_time, crest time, delta time, early systolic peak, late systolic peak, inflection point area, reflection index, relative crest time, relative delta time, stiffness index, and systolic amplitude or PPGs. All of the subjects provided written informed consent before participating in the investigation. They exhibited normal blood pressure, i.e., <150 mmHg systolic or <90 mmHg diastolic, regardless of any drug administration. The participants were normotensive (mean systolic blood pressure, 123.10 mmHg; range, 110–148 mmHg) and had no known cardiovascular, neurological, or respiratory disease. Prior to the experiment, the physical characteristics of the subjects were acquired. Physical data were measured for reference and are summarized in Table 1. Figure 1 shows the age group distribution.

We applied two statistical methods to evaluate the relationships between the 25 total indices and age, including the combined index of the PPG and APG features, defined as the inflection point area plus the d_peak (IPAD). The methods included calculation of the Pearson correlation coefficients (r) and a scatter plot regression analysis using the MedCalc program (MedCalc Software, Ostend, Belgium), which is considered a useful tool for the statistical analysis of medical data. A 95% confidence interval for the correlation coefficient (r) was calculated, and regression was performed to assess the independent association between age and each feature. Statistical significance was defined as p < 0.05. Scatter plots were generated to confirm the linear relationship according to the correlation coefficients.

The PPG is a remarkably simple device consisting of a light source (LED) and a light detector (photo diode). The detector was placed directly across from the light source for plethysmography transmission from the fingertip. The plethysmographic waveform displayed in Figure 2 is a filtered signal based on the red signal (approximately 660 nm). This wavelength is absorbed when light is transmitted through the peripheral vascular bed. The magnitude of the PPG signal depends on the amount of blood ejected from the heart into the peripheral vascular bed with each cardiac cycle, the optical absorption of the blood, the composition and color of the skin and underlying tissues, and the wavelengths used to illuminate the blood. Because PPG signals are often very small and typically contain unwanted noise together with the meaningful vascular information, it is important to detect the inflection point of the PPG waveform reflecting vascular aging during the cardiac cycle. The pulsatile component, or the AC (alternating current) portion of the signal, is displayed by removing the static component, or the DC (direct current) portion of the signal, representing the absorption of light by surrounding tissue by means of a high-pass filter. The PPG waveform is regulated by an auto-gain algorithm designed to maximize the size of the waveform [7]. One of the most useful plethysmographic features is the waveform amplitude. Amplitude changes are obtained by turning off the auto-gain because the amplitude of the plethysmograph signal is directly proportional to the vascular distensibility [8]. If the vascular compliance is low during episodes of increased sympathetic tone, PPG amplitude is also low. However, PPG amplitude is decreased during a significant increase in arterial blood pressure due to increased sympathetic tone. The pulse distorts the arterial wall, and the distortion travels at a speed determined by the stiffness or elasticity of the wall. Softer arteries have slower return waves with longer peak-to-peak times. The reverse is true for harder arteries. A mathematical equation can be used to determine the stiffness index (SI) of the arteries based on the distance/timing of these two peaks. Regarding the SI, large arterial stiffness is highly age-related and is a good biomarker of aging [9]. The time between the systolic and diastolic peak points, or the delta time (DT), is related to the time taken for the pressure wave to propagate from the heart to the periphery and back. The crest time (CT) is calculated as the time from the foot of the PPG waveform to its peak:

The reflection index (RI) has been shown to be a noninvasive indicator for vascular assessments [10]. RI can be calculated as

The augmentation index (AI) is defined as the ratio of late systolic peak (LSP) to early systolic peak (ESP) as follows [11]:

Dividing the pulse area into two areas at the dicrotic notch, area A1 and area A2, has been associated with vascular healthiness [12]. The ratio of the two areas is shown in Figure 2. This ratio is called the inflection point area (IPA) ratio and is defined as

The relative CT (RCT) and the relative DT (RDT) are defined as

The second derivative was applied to the filtered PPG to analyze the APG waveform, which is more commonly analyzed than the PPG waveform itself. This waveform is called the acceleration plethysmogram because it is an indicator of the acceleration of the blood in the finger. As shown in Figure 3, the APG waveform includes four systolic waves and one diastolic wave, namely, wave_a (early systolic positive wave), wave_b (early systolic negative wave), wave_c (late systolic re-increasing wave), wave_d (late systolic redecreasing wave) and wave_e (early diastolic positive wave), representing the dicrotic notch. In APG analyses related to age, the ratios of the height of each wave to that of the wave_a (b/a, c/a, d/a, and e/a) are typically used for wave analyses [13]. Other authors have demonstrated that the b/a ratio and the three other ratios (the c/a ratio, the d/a ratio, and the e/a ratio) reflect increased arterial stiffness and decreased arterial stiffness, respectively [14]. Additionally, they found that the (b-c-d-e)/a, defined as the aging index, increases with age and may be useful for evaluating vascular aging and for screening for arteriosclerotic diseases. Recent investigations have also evaluated the usefulness of APG and compared data on blood pressure, serum biochemistry, and the aortic calcification index in chronic kidney disease patients [15]. It has been reported that the APG aging index and the aortic calcification index increase with aging and that the aging index is significantly correlated with the aortic calcification index. In this study, the detection of waves in the APG signals was used to accurately measure heart rate. The bc_slope and the bd_slope were calculated as shown in Figure 3.

The circuit topology used in this study is shown in Figure 4. For an operational voltage of 3.3 V (V_cc), a 64-Ω resistor between the LED and V_cc in series was used to create an LED bias of 32 mA. In the next stage, the high-pass filter was a simple first-order RC (register-capacitor) circuit with a lower frequency bound of (2*pi*R₀*C₀)⁻¹. The low-pass filter was also implemented as an active filter to facilitate amplification. The high-pass filter had a cut-off of (2*pi*R₂*C₂)⁻¹ and a gain of (R₁+R₂)/R₁. To increase the effectiveness of the bandpass filter, a Sallen-Key bandpass filter was implemented [7]. The transfer function (TF) of the circuit shown in Figure 4 is expressed as

PPG signals, obtained with a preprocessing circuit of this TF, showed detailed changes of pulsatile blood volume including dicrotic notches. The APG waveform enabled a more accurate recognition of inflection points from such PPG signals.

A linear regression analysis was performed to examine the relationship between physical age and APG indices among all subjects. The APG-related parameters are shown in Table 2. The slope between age (x-axis) and the other indices (yaxis) was analyzed for data correlation. The correlation coefficient of the d_peak among the ten APG indices was the highest, showing that there is moderate association between the d_peak and age. As seen in Figure 5A, the scatter plot of a negative linear relationship shows a somewhat linear trend, although the trend is not clear. The e_peak with a correlation coefficient of r = 0.140 (p = 0.1794) is not associated with age. The age index has a correlation coefficient of r = 0.296 (p = 0.0390), indicating that there is not a statistically strong relationship with age. Figure 5B shows that the age index does not necessarily represent a strong association with age. Takazawa et al. [14] demonstrated that the second derivative of the wave aging index increased with age (r = 0.80, p < 0.001). The index of vascular age increases as age increases, reflecting age-related thickening of the vascular vessel. The c_peak and d_position were correlated with age, with coefficients of r = −0.288 (p = 0.0052) and r = −0.276 (p = 0.0075), respectively. The correlation coefficient between the bd_slope and age (r = −0.265, p = 0.0103) was smaller than that of the c_peak and the d_position. The bc_slope-related correlation coefficient was r = −0.135 (p = 0.1974).

Figure 6A shows the relationship between age and the IPA (r = −0.345, p = 0.0007). The PPG-related parameters are shown in Table 3. The correlation was stronger than the correlation between age and the APG age index (r = 0.296, p = 0.0039). The correlation of the late systolic peak with age corresponded to the next highest r value (r = −0.327, p = 0.0014), as shown in Figure 6B. In addition, age was correlated with the reflection index (r = 0.325, p = 0.0015), the A2 area (r = −0.329, p = 0.0013), and the A1_time (r = −0.247, p = 0.017). Millasseau et al. [9] showed that the SI increases with age because the time delay between the systolic and diastolic peaks decreases with age as a consequence of increased large artery stiffness. However, in this study, the SI and age exhibited negligible correlation (r = −0.031, p = 0.7713).

We analyzed the combined APG and PPG index, which corresponded to the highest correlation coefficient. The IPA from the PPG signals and the d_peak from the APG signals were combined and were the IPAD. The correlation coefficient of the IPAD was r = 0.56 (p < 0.001). Figure 7 shows that there was good correlation between age and the IPAD.