Framework for Automatic Delineation of Second Derivative of Photoplethysmogram: A Knowledge-based Approach

Abstract

Knowledge-based rules for delineating the second derivative of photoplethysmogram (SDPTG), which is widely used as an indicator of arterial stiffness, are proposed in this study. The SDPTG facilitates the distinction of five sequential waves, namely the initial positive wave (IPW), the early negative wave (ENW), the late upsloping wave (LUW), the late downsloping wave (LDW), and the diastolic positive wave (DPW). An analysis of these waves indicates that the SDPTG is a slowly time-varying signal and that the difference between two adjacent pulses cannot go beyond a certain range. It also indicates that the diastolic positive wave can be accurately estimated from the envelope of the SDPTG signal even with a noisy signal. To delineate the SDPTG, pulse waveforms are first divided into pulse segments, each of which contains one period of the SDPTG signal, using the slope sum function with an adaptive thresholding scheme, which simplifies detecting pulse onsets by enhancing the upslope of the pulse signal and suppressing the remainder of the signal. After pulse segmentation, IPW is first identified by picking the maximum positive peak in the segment. DPW is then extracted using a knowledge-based rule that uses the envelope of the SDPTG signal. In the range from IPW to DPW, ENW, LUW, and LDW are sequentially determined using knowledge-based rules. The proposed method is evaluated using the HMS database, which includes 1,386 pulses. A positive predictive value of 99.71% and a false negative rate of 1.02% are obtained, and thus the proposed rules are expected to facilitate pulse diagnosis using SDPTG signals.