A novel online method to monitor autonomic nervous activity based on arterial wall impedance and heart rate variability

Abstract

This paper proposes a new method of evaluating autonomic nervous activity using the mechanical impedance of arterial walls and heart rate variability. The cardiovascular system is indispensable to life maintenance functions, and homeostasis is maintained by the autonomic nervous system. Accordingly, it is very important to be able to make diagnosis based on autonomic nervous activity within the body’s circulation. The proposed method was evaluated in surgical operations; the mechanical impedance of the arterial wall was estimated from arterial blood pressure and a photoplethysmogram, and heart rate variability was estimated using electrocardiogram R–R interval spectral analysis. In this paper, we monitored autonomic nervous system activity using the proposed system during endoscopic transthoracic sympathetic block surgery in eight patients with hyperhidrosis. The experimental results indicated that the proposed system can be used to estimate autonomic nervous activity in response to events during operations.

Appendix

In the program of MemCalc, a time series is assumed to be composed of underlying variation and fluctuating parts; the underlying variation is expressed as the function x _uv (t), which can be given by a linear combination of sine and cosine functions

$$ x_{uv} \left( t \right) = a_{0} + \sum\limits_{n = 1}^{Np} {\left[ {a_{x} \sin \left( {2\pi f_{n} t} \right) + b_{x} \cos \left( {2\pi f_{n} t} \right)} \right]} $$

(5)

where f _n is the frequency of the nth component, a _n and b _n are the amplitudes of the nth periodic component, Np is the total number of components, and a ₀ is a constant that indicates the mean value of the time series. The value of f _n is determined by the peaks in the power spectral density. Its estimate, P(f), can be expressed as

$$ P\left( f \right) = {\frac{{\Updelta tP_{m} }}{{\left[ {1 + \sum\nolimits_{k = - m}^{m} {\gamma_{m,k} \exp \left( { - i2\pi fk\Updelta t} \right)} } \right]^{2} }}} $$

(6)

where P _m is the output power of the prediction error filter of the order m, and γ _m,k is the corresponding filter coefficient, m = 0, 1, 2,…, M (M = optimum filter order). P _m and γ _m,k are determined by Yule–Walker equations using Burg’s algorithm. Ohtomo and Tanaka [24] demonstrated that Eq. 5 gives a basis for determining the filter order, and the optimum order should be determined by the condition; the filter order is >1/f _min, where f _min is the minimum amongst the central frequencies of the components. In this program, the filter order was much higher than those obtained from the first minima of conventional information criteria. MemCalc does not cause distortion of the power calculation even if the underlying variation is changed.