Steady cardiovascular monitoring can play a key position in precision well being. Nevertheless, some basic cardiac
biomarkers of curiosity, together with stroke quantity and cardiac output, require invasive measurements, e.g., arterial strain
waveforms (APW). As a non-invasive various, photoplethysmography (PPG) measurements are routinely collected
in hospital settings. Sadly, the prediction of key cardiac biomarkers from PPG as a substitute of APW stays an open
problem, additional difficult by the shortage of annotated PPG measurements. As an answer, we suggest a hybrid ap-
proach that makes use of hemodynamic simulations and unlabeled medical knowledge to estimate cardiovascular biomarkers immediately
from PPG alerts. Our hybrid mannequin combines a conditional variational autoencoder educated on paired PPG-APW knowledge
with a conditional density estimator of cardiac biomarkers educated on labeled simulated APW segments. As a key consequence,
our experiments display that the proposed strategy can detect fluctuations of cardiac output and stroke quantity
and outperform a supervised baseline in monitoring temporal modifications in these biomarkers.
- †ETH Zurich, Switzerland
- ** Work achieved whereas at Apple
