Figure 1
a. and b. display the ECG and PPG waveform morphology, respectively. The ECG is divided into distinct waves (a, I–V), of which the R-wave (a, II) is used for heart beat extraction. With the PPG wave, the systolic peak (b, I) is used. The plot in c. shows the relationship between ECG and PPG signals.
Figure 2
The ECG signal (a.) shows a strong QRS complex together with little amplitude variation. The PPG signal measured simultaneously while the patient is at rest in a hospital bed (b.) shows some amplitude variation but relatively stable morphology. When measuring PPG in a driving simulator using low-cost sensors (c.), strong amplitude and waveform morphology variation is visible.
Figure 3
Figure showing the process of peak extraction. A moving average is used as an intersection threshold (II). Candidate peaks are marked at the maximum between intersections (III). The moving average is adjusted stepwise to compensate for varying PPG waveform morphology (I). (IV). shows the detection of the onset and end of clipping, and the result after interpolating the clipping segment.
Figure 4
Image showing how the dynamic threshold is fitted using SDSD. The last image (III.) shows that even missing a single beat will lead to a large increase in SDSD compared to the optimal fitting. BPM is also taken into account when fitting.
Figure 5
The plotted RR-intervals with thresholds (I.), and the resulting rejected peaks (II.).
Figure 6
Plot from PPG dataset with low-confidence sections marked. These are ignored in the computation of output measures.