Variability of morphology in beat-to-beat photoplethysmographic waveform quantified with unsupervised wave-shape manifold learning for clinical assessment

TL;DR

This study uses unsupervised manifold learning to quantify beat-to-beat morphological variability in PPG waveforms, aiming to enhance clinical assessment despite PPG's susceptibility to artifacts.

Contribution

It introduces a novel application of Dynamic Diffusion Map to analyze PPG waveform variability for clinical insights.

Findings

PPG waveform morphology variability correlates with patient condition.

Unsupervised manifold learning effectively captures waveform variations.

Method validated on real clinical data.

Abstract

We investigated the beat-to-beat fluctuation of the photoplethysmography (PPG) waveform. The motivation is that morphology variability extracted from the arterial blood pressure (ABP) has been found to correlate with baseline condition and short-term surgical outcome of the patients undergoing liver transplant surgery. Numerous interactions of physiological mechanisms regulating the cardiovascular system could underlie the variability of morphology. We used the unsupervised manifold learning algorithm, Dynamic Diffusion Map, to quantify the multivariate waveform morphological variation. Due to the physical principle of light absorption, PPG waveform signals are more susceptible to artifact and are nominally used only for visual inspection of data quality in clinical medical environment. But on the other hand, the noninvasive, easy-to-use nature of PPG grants a wider range of biomedical application, which inspired us to investigate the variability of morphology information from PPG waveform signal. We developed data analysis techniques to improve the performance and validated with the real-life clinical database.