2.4 Definition of PTT

PTT is usually defined as the time between the R-peak of the electrocardiogram (ECG) and a reference point on systolic PPG signal segment. The reference point can be derived from different PPG features (e.g., end-of-diastolic valley, systolic peak, see Figure 3), which leads to different PTT definitions.

Illustration of two types of PTT, i.e., preprocessed PTT and filtered PTT. Here, the preprocessed PTT refers the interval between the ECG R-wave and the end-of-diastolic valley of the preprocessed PPG signal, while the filtered PTT is the interval between the ECG R-wave and the end-of-diastolic valley of the filtered PPG signal.

This work selected eight different PPG pulse waveform characteristics to define eight PTTs. The eight PPG pulse waveform characteristics are (see Figure 4).

The characteristic points on (A) the PPG signal and its (B) first and (C) second derivatives. Points 1-8 denote O1D, VP, M2D, M1D, V2D, FP, MP, and V1D, respectively. Adopted from (Mejia-Mejia et al., 2022). (D) and (E): Corresponding PPG waveform characteristics points of both signals (preprocessed and filtered) were excluded from the analysis if any of them falls in the first 0.5s (D) or last 0.75s of the recording (E).

The derivatives were approximated using backward difference calculated from adjacent sampling points. Therefore, the first and second derivatives of the PPG signal started from the second and third sampling points, respectively. Considering the high sampling frequency (i.e., 2,500 Hz), the error caused by the approximation was very limited (< 4×10−4 s for the timing of any characteristic point). The characteristic points were detected from the extrema (i.e., peak and valley points) of PPG and its derivatives, as well as the decomposition of forward and backward pulse waves. The details of defining and detecting characteristic points can be found in our early works (Liu et al., 2021; Lin et al., 2022).

To prevent inaccurate readings at the immediate start and end of a PPG recording, any characteristic point was excluded from the analysis if it or its ‘counterpart’ (i.e., any of the preprocessed or filtered one) fell in the first 0.5 s (e.g., in Figure 4D, the pair of peak points are excluded) or the last 0.75 s (Figure 4E). To exclude the missing or erroneous feature points, any detected feature point was excluded if there was no ‘counterpart’ point within ±0.3 s of the detected feature point. The time axis was unchanged (i.e., no shift of any signal) during signal processing.

As to ECG signals, the R wave peak was detected as the maximal value in a cardiac cycle using the Pan Tompkins method (Sathyapriya et al., 2014). To prevent inaccurate readings at the immediate start and end of an ECG recording, similar as in PPG preprocessing, any R peak point was excluded if it or its counterpart was in the first 0.5 s or the last 0.5 s. When processing the noisy PPG signals in some cardiac cycles, only the valleys within 100–500 ms after the ECG R-peak (i.e., 100 ms≤ PTT ≤500 ms) were selected for analysis. For each PPG signal (preprocessed or filtered), the PTT was calculated as the mean value of PTTs of all included cardiac cycles.