A system identification approach to non-invasive central cardiovascular monitoring

Abstract

This thesis presents a new system identification approach to non-invasive central cardiovascular monitoring problem. For this objective, this thesis will develop and analyze blind system identification and input signal reconstruction algorithms for a class of 2-channel IIR and Wiener systems. In particular, this thesis will present blind identifiability conditions for a class of 2-channel IIR and Wiener wave propagation systems and develop the associated blind identification algorithms. It will be shown that the blind identifiability conditions can be achieved in many real-world applications by appropriate selection of channel lengths, sensor locations, and sampling frequency which are the specifications that the system design can exploit for blind identifiability In addition, this thesis will develop a novel input signal reconstruction algorithm that is applicable to general class of multi-channel IIR and Wiener systems. Furthermore, this thesis will rigorously analyze and evaluate three analytic measures for determining the system order and other key parameters of the black-box dynamics as well as for quantifying the quality of the identified gray-box dynamics, without any direct use of unknown input signal: persistent excitation, model identifiability and asymptotic variance. The blind identification and input signal reconstruction algorithms will first be applied to 2-sensor central cardiovascular monitoring problem using two distinct peripheral blood pressure measurements, where the cardiovascular wave propagation dynamics is blindly identified and the aortic blood pressure and flow signals are reconstructed by exploiting black-box and physics-based gray-box model structures of the cardiovascular system.

(cont.) The validity of the 2-sensor central cardiovascular monitoring methodology will be illustrated by experimental data from swine subjects and simulation data from a full-scale human cardiovascular simulator across diverse physiologic conditions. The 2-sensor central cardiovascular monitoring methodology will then be extended to address noninvasive, 1-sensor cardiovascular monitoring problem, where the specific challenges involved are 1) identifying the cardiovascular wave propagation dynamics and reconstructing the aortic blood pressure signal by exploiting the measurement from a single peripheral sensor, and 2) identifying the scale for calibrating the blood pressure signal. In order to address these challenges, this thesis will propose a heuristics-based system order estimation algorithm and a model-based blood pressure calibration algorithm, which will be combined with the blind identification of the cardiovascular wave propagation dynamics to realize the non-invasive 1-sensor central cardiovascular monitoring. The non-invasive 1-sensor central cardiovascular monitoring methodology will be illustrated by experimental data from swine subjects, simulation data from a full-scale human cardiovascular simulator, and experimental data from human subjects across diverse physiologic conditions.