Novel computation of pulse transit time from multi-channel PPG signals by wavelet transform

2.1 Hardware

To gain access to the PTT, two pick-up points for the pulse wave are needed at different distances from the heart along the same artery. We implement this using a dual photo plethysmography (PPG) node, encompassing a conceptually traditional, transmissive finger clip (though with adaptive gain and filtering) and a more intricate reflective sensor. These two are guaranteed by design to be absolutely synchronous. The 12 bit samples are acquired with a 500 Hz rate. The reflective sensor is placed either a top the radial or ulnar artery in close proximity to the wrist, dependent on the emerging signal quality. The finger clip can be positioned on whichever finger, as long as the arterial length in-between is roughly known (typ. 15–25 cm). This setup would, in theory, suffice for calculating the PTT. Nonetheless, for the sake of robustness, a wirelessly (Bluetooth) synchronized minimal ECG is recorded to be able to assess the different PAT at the two sites. These subsystems are part of a bigger robust body-sensor-network (rBSN) [8], [9].

The standard used for validation is the data retrieved from a Finapres Medical Systems Portapres system, which has been synchronously recorded via the Portapres analog output module and an rBSN universal data recorder. The finger cuff is attached to a different finger than the transmissive PPG clip, so as to avoid interference, and the hydrostatic pressure compensation of the Portapres has been deactivated. The Portapres is by no way a gold standard – unfortunately arterial catheter or cuff sphygmomanometer measurements were not viable because of their invasivity and low sample rate respectively.

2.2 Methodology

The experiment involves raising and holding the measured limb at various angles to the ground while standing: –90, –45, 0, 45 and 90 degrees respectively [10]. This effects different relative height to the heart and emulates a change in blood pressure. The limb can either be passively held in place by appropriate means, or actively held up, yielding two measurement scenarios.

2.3 Processing

The resulting signals (one ECG – preferably Einthoven II, and two PPGs) are processed in Matlab, at first using quasi-standardized algorithms: the R-peak detection is done via a Pan-Thompkins-like algorithm [11], albeit slightly adapted for noisy environments. The DC component of the PPG signals is discarded, as it bears no significance for the further proceedings, and different fiducial points thereof are computed; for example pulse wave onset and peak are extracted with a modified Zong algorithm [12].

From here on, two methods will be pursued: based on the aforementioned fiducial points, or employing the continuous wavelet transform. Since the expected fluctuation in PAT is very small, the task of reliably accessing this information is jeopardized by signal disturbances. Therefore the chosen reference point for the pulse wave arrival is halfway between onset and first maximum – much more immune to noise and artifacts due to its location on the sharply ascending upslope.

The second method relies on the continuous wavelet transform to identify the pulse wave location within the signal.

The advantage of this approach, though more computationally complex, is its relative imperviousness to noise and uncorrelated artifacts. To increase the temporal resolution of the results, a ten-fold up sampling is performed, followed by the CWT with an appropriate analysing wavelet. In the performed analysis, Mexican Hat employing rather large scaling factors has empirically been proven most appropriate so far. As pulse wave morphology can vary very strongly between individuals and even at different pick-up locations (Figure 3) of the same one, the effect of wavelet shape is difficult to evaluate. After each R-peak, the maximum of the resulting coefficients is an indicator for the most probable wave location. The resulting indexes from the two methods can now be subtracted to gain the PTT between wrist and finger. As obvious from Figure 3, the morphology of the pulse wave, or rather the change thereof, plays a role in the detected peak locations, and therefore the calculated PAT and PTT. This is, however, not problematic, but should much rather could have the added benefit of assessing two features at the same time – PTT and morphology variation related to BP changes. Nonetheless, further research needs to be undertaken to investigate the link.

3.1 Remarks

As stated, during the 90 and sometimes 45 degree raising of the arm, the PPG quality spontaneously degenerates, yielding an unsatisfactorily accurate signal for determining the pulse wave locations. This is due to multiple factors, amongst which is the experimental setup itself. By producing such a big height difference and, therefore, static pressure differential between heart and hand, the pulse strength is strongly diminished. Furthermore, due to the uncomfortable and somewhat unnatural pose, tremors and contractions are provoked in the subject, juddering the measurement apparatus and introducing disturbances.

The PTT does not, however, tell the whole truth. One would likely expect to find a simultaneous fluctuation of the PAT at both locations (i.e. wrist and finger) – interestingly, this is not the case. While the PAT at the fingertip does visibly scale with the height of the hand, the PAT at the wrist seems less influenced, remaining flat almost throughout all phases, as depicted in Figure 5. This would suggest that the big arteries (aorta, brachial and radial/ulnar) do not experience much pressure change at all, even when the BP as a whole does vary. This is consistent with the actuality that most of the vascular resistance occurs in the small blood vessels [13], making the changes in BP more visible therein.

Finally, it has been observed that the PAT in the fingertips (and accordingly the PTT) does not return to base level as soon as the limb is lowered in its incipient position, but does linger in a heightened state for some time. This is probably due to the latency of the capillary system’s reaction to abrupt state changes. Also, PAT at the wrist seems to rise only for the 90 degree test, but returns to normal instantaneously after release, hinting at a possible mechanical clamping of the arteries at the shoulder.