Breathing removal using Gaussian process regression for improved parametric impedance estimation of human respiratory system

Abstract

Respiratory Oscillometry is a promising technique to provide information to medical practitioners on the respiratory system of a patient in a non-invasive fashion. It focuses on identifying the respiratory impedance between the air pressure and flow signals. However, for conscious patients, breathing acts as a disturbance to the parameter estimation process. Therefore, it increases the variance of the estimated parameters of the impedance. To solve this problem, this paper proposes a method to separate the breathing contribution from the controlled excitation contribution in the measured signals. The breathing contribution is modelled as a Gaussian Process in the frequency domain. This allows to remove the breathing contribution altogether and leads to a reduced variance in the impedance estimation.