Elsevier

Computers in Biology and Medicine

Volume 89, 1 October 2017, Pages 355-367
Computers in Biology and Medicine

Time-varying assessment of heart rate variability parameters using respiratory information

https://doi.org/10.1016/j.compbiomed.2017.07.022Get rights and content

Highlights

  • Time-varying HRV parameters are estimated by ARMAX modeling and Kalman filtering.

  • Respiratory information is explicitly included for enhanced LF and HF band discrimination.

  • The proposed joint model outperforms classical AR modeling especially for low respiration rate.

  • PTT and PPGamp can help in more efficient separation of the HRV components when the respiration signal is not available.

  • PTT outperforms PPGamp during normal respiration, and vice versa during deep respiration.

Abstract

Analysis of heart rate variability (HRV) is commonly used for characterization of autonomic nervous system. As high frequency (HF, known as the respiratory-related) component of HR, overlaps with the typical low frequency (LF) band when the respiratory rate is low, a reference signal for HF variations would help in better discriminating the LF and HF components of HR. The present study proposes a model for time-varying separation of HRV components as well as estimation of HRV parameters using respiration information. An autoregressive moving average with exogenous input (ARMAX) model of HRV is considered with a parametrically modeled respiration signal as the input. The model parameters are estimated using smoothed extended Kalman filtering. Results for different synthetic data show that our proposed joint model outperforms the classical AR modeling in estimation of HRV parameters especially in the case of low respiration rate. In addition, the possibility of using pulse transit time (PTT) and the amplitude of photoplethysmogram (PPGamp) as surrogates of the input respiratory signal has been investigated. To this end, electrocardiogram (ECG), PPG and respiration have been recorded from 21 healthy subjects (10 males and 11 females, mean age 27.5 ± 4.1) during normal and deep respiration. Results show that indeed PTT and PPGamp offer good potential to be used as references for respiratory-related variations of HR, thus avoiding additional devices for recording respiration.

Introduction

Heart rate variability (HRV) analysis has been widely used to assess the function of autonomic nervous system (ANS) in regulation of cardiovascular system [1], [2]. ANS activity has been investigated through analysis of HRV —also known as R-R interval (RRI) variability— or studying the interaction between heart rate (HR), blood pressure (BP) and respiration variations [3], [4]. HR is affected by BP through the baroreflex system, while respiration mediates cardiovascular parameters through both ANS and mechanical paths [5].

In general, two frequency components in HR spectrum are of significance: low frequency (LF) component (with a typical band of 0.04–0.15 Hz) which is thought to be affected by both sympathetic and parasympathetic tones, and high frequency (HF) component (0.15–0.4 Hz) which is mostly mediated by respiration and known as respiratory sinus arrhythmia [1], [6]. HF component is mainly of parasympathetic (vagal) origin. The ratio between the powers of these two components (LF/HF) is considered as a measure of sympatho-vagal balance [1].

Some indices of HRV in stable conditions (e.g., time-domain indices including statistical and geometrical indices and frequency-domain indices like LF/HF ratio) have been established for clinical diagnosis of ANS function [7]. Furthermore, time-varying analysis of HRV is of importance in order to assess HRV parameters continuously in healthcare monitoring systems or during non-stationary conditions. Different parametric and non-parametric methods have been considered for this purpose [1], [4], [8], [9], [10], [11], [12], [13]. The main shortcoming of most HRV studies is that they do not consider respiration information. When the respiration rate is very low, the HF component of HRV interferes with the typical LF band. This leads to an unreliable estimation of the LF and/or HF powers [14], [15], [16]. To address this problem, some studies have considered adaptable frequency bands depending on the respiration frequency [17]. Commonly, a band with the center frequency equal to the respiration rate is considered for the HF component. In these approaches, the respiration frequency is measured from a respiratory-related signal [18], [19], [20], [21], [22], [23].

While some studies benefit from respiration information in HRV analysis for better definition of the HF band, some others have tried to remove respiratory-related oscillations from HR. It is thought that using the non-respiratory related part of HR signal in HRV or HR-BP coupling analysis results in derivation of more relevant ANS or baroreflex indices [6]. To this end, different methods have been used, e.g., adaptive filtering, orthogonal subspace projection and multiscale principal component analysis (PCA) [6], [24] and autoregressive moving average with exogenous input (ARMAX) analysis with the respiration-related signal as the input of the model [25], [26]. The performance of these different methods have been investigated in [6] and it has been concluded that ARMAX and orthogonal subspace projection work better than other approaches in separating RRI components.

Our study specifically addresses the instantaneous separation of RRI components and time-varying estimation of HRV parameters simultaneously using respiration information. An ARMAX model for the RRI series with the respiratory signal as the exogenous input is considered for this sake. As mentioned before, the efficiency of ARMAX in separation of respiratory-related component of HR has been reported previously [6]. In this study, we specifically employ the ARMAX model in the context of time-varying estimation of HRV parameters by parametrically modeling the respiratory signal as a sinusoid. This joint modeling enables us to estimate time-varying HRV parameters without the need to consider pre-defined LF and HF bands. Smoothed extended Kalman filtering (SEKF) is used for time-varying estimation of the model parameters. Kalman filter is a powerful tool for state estimation even when the precise nature of the model is unknown. SEKF avoids the lag error presented by the normal Kalman filtering [9].

HRV analysis using respiratory information normally needs an additional device to be connected to the body like strain-gauge belts, nasal thermistors or capnogram to record the respiration, which is uncomfortable and may be contraindicated for certain patients. Therefore, we also investigate the possibility of using pulse transit time (PTT) and the amplitude of photoplethysmogram (PPGamp) as the input of our model. These signals, especially PPGamp, have been used in other related contexts, e.g., estimation of respiration rate [18], [27], [28]. Herein, we study their efficiency as a reference for respiratory-related oscillations in our HRV analysis framework.

The paper is organized as follows. Section II describes the synthetic and real data and the proposed model. In addition, time-varying estimation of the model parameters using SEKF is explained. Section III presents the results on the separation and power estimation for synthetic and real data. Section 4 Discussion, 5 Conclusion are devoted to discussion and conclusion respectively.

Section snippets

Synthetic data

Different synthetic signals are generated using a physiologically based model. The model includes parts for the baroreflex and cardiovascular couplings and has been described and validated elsewhere [29], but we briefly describe it here for the sake of completeness. The cardiovascular system is represented by a 2-element Windkessel model. Physiologically based equations explain the baroreflex effects on HR through separate sympathetic and parasympathetic ways. The ANS-stimulating signal

Step change in the frequencies of LF and HF components (scenario 1)

Fig. 2 shows the respiration frequency estimated by the Kalman smoother algorithm (fr(n)) together with the actual respiration rate (fHF in eqn. (1)) as the reference. Fig. 3a shows the constructed respiration signal by (6) using the estimated parameters. The LF and HF time series of RRI estimated by the first and second parts of (3), respectively, are shown in Fig. 3b together with the real synthetic RRI signal. Figs. 3c and d show the frequency spectra of the estimated LF and HF components of

Discussion

As Fig. 4, Fig. 7 show, our proposed joint model performs better than AR model in estimating the powers of LF and HF components of HR for the different synthetic signals. In Fig. 4, when the frequencies change at the time point 250 s and HF spectrum gets into the typical LF band, the AR model underestimates and overestimates the HF and LF powers, respectively. For the chirp varying frequency signal, AR is unable to estimate the powers of LF and HF components properly before the approximate time

Conclusion

This study has presented a method for time-varying separation of HRV components and estimation of HRV classical parameters simultaneously based on a modified ARMAX model for the RRI series jointly used with a parametric model of respiratory-related signal. AR modeling is commonly used in the context of HRV parameter estimation which gives good frequency and reasonable time resolutions. Compared to AR, our joint model is especially effective in the case of interference of respiration rate with

References (43)

  • U.R. Acharya et al.

    Heart rate variability: a review

    Med. Biol. Eng. Comput.

    (2006)
  • L.T. Mainardi

    On the quantification of heart rate variability spectral parameters using time–frequency and time-varying methods

    Phil. Trans. Roy. Soc.Lond. Math. Phys. Eng. Sci.

    (2009)
  • M.P. Tarvainen et al.

    Time-varying analysis of heart rate variability signals with a kalman smoother algorithm

    Physiol. Meas.

    (2006)
  • E.M. Dantas et al.

    Spectral analysis of heart rate variability with the autoregressive method: what model order to choose?

    Comput. Biol. Med.

    (2012)
  • E. Toledo et al.

    Wavelet analysis of instantaneous heart rate: a study of autonomic control during thrombolysis

    Am. J. Physiol. Regul. Integr. Comp. Physiol.

    (2003)
  • J. Echeverria et al.

    Application of empirical mode decomposition to heart rate variability analysis

    Med. Biol. Eng. Comput.

    (2001)
  • N. Wessel et al.

    Is the normal heart rate chaotic due to respiration?

    Chaos Interdiscip. J. Nonlinear Sci.

    (2009)
  • L. Sidorenko et al.

    Standard heart rate variability spectral analysis: does it purely assess cardiac autonomic function?

    Europace

    (2016)
  • G.E. Billman

    The LF/HF ratio does not accurately measure cardiac sympatho-vagal balance

    Front. Physiol.

    (2013)
  • S. Nemati et al.

    Data fusion for improved respiration rate estimation

    EURASIP J. Adv. Signal Process.

    (2010)
  • W. Karlen et al.

    Multiparameter respiratory rate estimation from the photoplethysmogram

    IEEE Trans. Biomed. Eng.

    (2013)
  • View full text