TY - JOUR

T1 - Bayesian source separation of electrical bioimpedance signals

AU - Pichler, Christof

AU - Ranftl, Sascha

AU - Heller, Arnulf

AU - Arrigoni, Enrico

AU - von der Linden, Wolfgang

PY - 2021/5

Y1 - 2021/5

N2 - For physicians, it is often crucial to monitor hemodynamic parameters to provide appropriate treatment for patients. Such hemodynamic parameters can be estimated via electrical bioimpedance (EBI) signal measurements. Time dependent changes of the measured EBI signal occur due to several different phenomena in the human body. Most of the time one is just interested in a single component of the EBI signal, such as the part caused by cardiac activities, wherefore it is necessary to decompose the EBI signal into its different source terms. The changes of the signal are mostly caused by respiration and cardiac activity (pulse). Since these fluctuations are periodic in sufficiently small time windows, the signal can be approximated by a harmonic series with two different fundamental frequencies and an unknown number of higher harmonics. In this work, we present Bayesian Probability Theory as the adequate and rigorous method for this decomposition. The proposed method allows, in contrast to other methods, to consistently identify the model-function, compute parameter estimates and predictions, and to quantify uncertainties. Further, the method can handle a very low signal-to-noise ratio. The results suggest that EBI-based estimation of hemodynamic parameters and their monitoring can be improved and its reliability assessed.

AB - For physicians, it is often crucial to monitor hemodynamic parameters to provide appropriate treatment for patients. Such hemodynamic parameters can be estimated via electrical bioimpedance (EBI) signal measurements. Time dependent changes of the measured EBI signal occur due to several different phenomena in the human body. Most of the time one is just interested in a single component of the EBI signal, such as the part caused by cardiac activities, wherefore it is necessary to decompose the EBI signal into its different source terms. The changes of the signal are mostly caused by respiration and cardiac activity (pulse). Since these fluctuations are periodic in sufficiently small time windows, the signal can be approximated by a harmonic series with two different fundamental frequencies and an unknown number of higher harmonics. In this work, we present Bayesian Probability Theory as the adequate and rigorous method for this decomposition. The proposed method allows, in contrast to other methods, to consistently identify the model-function, compute parameter estimates and predictions, and to quantify uncertainties. Further, the method can handle a very low signal-to-noise ratio. The results suggest that EBI-based estimation of hemodynamic parameters and their monitoring can be improved and its reliability assessed.

KW - Bayesian probability theory

KW - Electrical bioimpedance

KW - Impedance cardiography

KW - Source separation

KW - Uncertainty quantification

UR - http://www.scopus.com/inward/record.url?scp=85102499037&partnerID=8YFLogxK

U2 - 10.1016/j.bspc.2021.102541

DO - 10.1016/j.bspc.2021.102541

M3 - Article

VL - 67

JO - Biomedical Signal Processing and Control

JF - Biomedical Signal Processing and Control

SN - 1746-8094

M1 - 102541

ER -