Elsevier

Control Engineering Practice

Volume 17, Issue 11, November 2009, Pages 1318-1328
Control Engineering Practice

Modeling and 2-sensor blind identification of human cardiovascular system

https://doi.org/10.1016/j.conengprac.2009.06.006Get rights and content

Abstract

This paper presents modeling and blind identification of human cardiovascular system. In contrast to the population-based methods widespread in current practice, the proposed method does not require any a priori knowledge of the cardiovascular system. This paper develops a human cardiovascular system model, analyzes its identifiability, and identifies the model using two diametric blood pressure measurements. The aortic blood pressure reconstructed using the identified model can eliminate the use of invasive aortic blood pressure measurement for cardiovascular monitoring. Results based on the data from a realistic human cardiovascular system simulator demonstrate the validity of the proposed model and identification method.

Introduction

Among many signals of interest in the cardiovascular (CV) system, the central signals, such as the aortic blood pressure (BP) and flow (BF), are important to understand cardiac physiology and global circulation, in comparison with extremity signals. Peripheral circulatory measurements, e.g. arterial BP measured at distal extremity locations, cannot be used as direct surrogates for their central counterparts, because the morphology of the central CV signals are distorted in distal locations (Nichols & O’Rourke, 1998). However, direct measurement of central signals, such as pulmonary artery or central aortic catheterization, entails costly and risky invasive procedures.

Therefore, there has been substantial interest in developing indirect methods to recover central CV signals from peripheral measurements by utilizing the dynamic relationship between the two, i.e. the CV dynamics. Most of the methods available to date involve a population-based model, employing parameter values derived from prior experimentation (Fetics, Nevo, Chen, & Kass, 1999; Gallagher, Adji, & O’Rourke, 2004; Hope, Tay, Meredith, & Cameron, 2003; Stok, Westerhof, & Karemaker, 2006). In many cases, this approach yields useful estimates (Fetics et al., 1999; Gallagher et al., 2004). In theory, however, the use of a population-based model for different subjects under diverse physiologic conditions will be invalid to some extent, i.e. there will be states when an average relationship between the aortic and peripheral signals is less applicable (Hope et al., 2003; Stok et al., 2006).

In order to reduce the reliance on a priori knowledge and population-based assumptions regarding the relationship between central and extremity CV signals, efforts have been made to use multiple extremity circulatory measurements (Hahn, Reisner, & Asada (2006), Hahn, Reisner, & Asada (2009); Swamy, Ling, Li, & Mukkamala, 2007; Zhang & Asada, 2004). It has been shown that, having sensors placed at multiple peripheral locations to measure BP waves, clinically important signals and phenomena may be monitored by identifying its dynamics using the multi-channel blind system identification (ID) methodology (Gurelli & Nikias, 1995; Xu, Liu, Tong, & Kailath, 1995).

Some of these efforts are based on the black-box models (Swamy et al., 2007; Zhang and Asada, 2004), where the order of the models has to be manually tuned by trial and error for different physiologic conditions to obtain high-fidelity models. To overcome this drawback, the feasibility of a physics-based approach was developed and demonstrated (Hahn, Reisner, & Asada (2006), Hahn, Reisner, & Asada (2009)), which was successfully applied to experimental swine subjects. In an attempt to extend its initial success to human CV health monitoring, this paper develops a physics-based model of human CV dynamics, and presents a method for its ID based on two distinct extremity BP measurements. The validity of the model and the approach will be illustrated using the central aortic and the peripheral BP data obtained from a realistic human CV system simulator (Ozawa, Bottom, Xiao, & Kamm, 2001). Theoretical analysis addressing the blind identifiability provides a sound basis for evaluating and validating the proposed method, altogether yielding an accurate and reliable method for characterizing the human CV dynamics.

This paper is organized as follows. Section 2 develops a physics-based model of human CV system. Blind ID method for a class of two-channel Wiener systems is presented and analyzed in Section 3, which is applied to characterize the human CV system model developed in Section 2. Section 4 discusses the human CV system simulator data used to validate the proposed method and the results obtained based on those data. Conclusion and future work are given in Section 5.

Section snippets

Modeling of two-channel human cardiovascular system

In this paper, the CV system is viewed as a two-channel wave propagation system, where its upper- and the lower-limb branches are described by distinct wave propagation channels in which BP and BF waves travel back and forth. In this context, the use of lumped parameter models such as the Windkessel model (Fogliardi, Di Donfrancesco, & Burattini, 1996) and the chamber-based model (Smith, Geoffrey Chase, Shaw, & Nokes, 2005) is not relevant due to the nonzero wave propagation time, which can

Blind ID of two-channel human cardiovascular system

In this section, a blind ID method for a class of two-channel Wiener systems (which subsumes the human CV system model developed in this paper) is presented, and this method is applied to identify the human CV system model and reconstruct the unknown aortic BP.

Human cardiovascular system simulator

A realistic human CV system simulator was used to investigate the validity of the human CV system model and its ID method presented in this paper. In particular, a distributed CV system model developed by Ozawa et al. (2001) was used in order to create BP data for several human subject models under different physiologic conditions. This model consists of a distributed arterial system and boundary conditions to simulate the left ventricle, bifurcations and peripheral vessels. The CV system model

Conclusion

In this paper, a method to characterize the human CV system and reconstruct the central/abdominal aortic BP for indirect central CV health monitoring applications was presented. A physics-based model of the two-channel human CV system was developed, which, using the blind ID method for two-channel Wiener systems presented in this paper, can be characterized from the observation of arterial BP at two distinct extremity locations. Using the eight human subject models based on a realistic human CV

Acknowledgments

This research was supported in part by the Sharp Corporation. We also appreciate Professor Roger D. Kamm and Dr. Xinshu Xiao at Massachusetts Institute of Technology for their kind assistance and suggestions in the use of the human CV system simulator.

References (22)

  • D. Gallagher et al.

    Validation of the transfer function technique for generating central from peripheral upper limb pressure waveform

    American Journal of Hypertension

    (2004)
  • B.W. Smith et al.

    Experimentally verified minimal cardiovascular system model for rapid diagnostic assistance

    Control Engineering Practice

    (2005)
  • K. Campbell et al.

    Time-domain formulation of asymmetric T-tube model of arterial system

    American Journal of Physiology

    (1990)
  • B. Fetics et al.

    Parametric model derivation of transfer function for noninvasive estimation of aortic pressure by radial tonometry

    IEEE Transactions on Biomedical Engineering

    (1999)
  • R. Fogliardi et al.

    Comparison of linear and nonlinear formulations of the three-element Windkessel model

    American Journal of Physiology

    (1996)
  • M. Gurelli et al.

    EVAM: An eigenvector-based algorithm for multichannel blind deconvolution of input colored signals

    IEEE Transactions on Signal Processing

    (1995)
  • J.O. Hahn et al.

    A blind approach to reconstruction of aortic blood pressure waveform using gray-box identification of multiple pressure transfer channels

    Proceedings of American Control Conference

    (2006)
  • Hahn, J. O., Reisner, A., & Asada, H. (2009). Blind identification of 2-channel IIR systems with application to central...
  • S. Hope et al.

    Use of arterial transfer functions for the derivation of aortic waveform characteristics

    Journal of Hypertension

    (2003)
  • J. Kong

    Electromagnetic wave theory

    (2000)
  • W. Nichols et al.

    McDonald's blood flow in arteries

    (1998)
  • Cited by (15)

    • Aortic pressure waveform reconstruction using a multi-channel Newton blind system identification algorithm

      2021, Computers in Biology and Medicine
      Citation Excerpt :

      Thus, in recent years techniques involving adaptive transfer function have been proposed to avoid the aforementioned problems with GTFs. These adaptive transfer functions have been based on various physical models, such as the Windkessel model [32,33], the 1D model [34,35] and the tube-load model [36–38]. The ATF methods based on the Windkessel model have limited accuracy and the single-element Windkessel model cannot describe pulse wave propagation.

    • Performance of variations of the dynamic elastance model in lung mechanics

      2017, Control Engineering Practice
      Citation Excerpt :

      In particular, two different approaches are proposed and investigated to determine the minimal E(P) by using different extensions to a single compartment lung model (Fig. 1). It is hypothesized that models that can determine behavior across multiple PEEP levels can capture patient specific information that could eventually help clinicians provide individualized care for ARDS patients (Hahn et al., 2009; Le et al., 2010; Smith et al., 2005). This comparison study was based on the retrospective clinical data of Bersten (1998), where 12 patients with different levels of lung injury or ARDS were included ventilated in square wave profile volume controlled mode at 3 or 4 different PEEP levels.

    View all citing articles on Scopus
    View full text