New insight into arrhythmia onset using HRV and BPV analysis
Abstract — In this paper Heart Rate Variability (HRV) and Blood Pressure Variability (BPV) were analyzed before the onset of cardiac arrhythmia in order to derive markers for short-term forecasting. The (a) coherence between systolic blood pressure (SBP) and cardiac oscillations in low-frequency (LF) and high-frequency (HF) band; (b) fluctuations of phase; (c) HRV and BPV as a LF power and HF power in frequency and time-frequency domain; (d) transfer function analysis of cardiovascular signals were analyzed. Arrhythmia was preceded by: a) lower coherence; b) increase in fluctuations of phase between signals; c) higher spectral energy associated with respiratory frequency in blood pressure signal; d) raise of sympathetic outflow to the heart; e) decreased HRV. Cardiac arrhythmia was characterized mainly by an increase in LF power of blood pressure, cardiac signal and transfer function. During self-termination of arrhythmia a larger increased in total BPV and HRV was recorded. These results suggest that important information about both neuronal cardiovascular control and risk for spontaneous arrhythmia can be provided by combined analysis of frequency, phase, and time-frequency analysis of blood pressure and cardiac oscillation.
Keywords : – cross-spectral algorithm, wavelet analysis, arrhythmia, blood pressure and heart rate variability, wavelet coherence
I. INTRODUCTION
In the last decades, several studies have suggested the role of the autonomic nervous system (ANS) both in the genesis and maintenance some of arrhythmias [1-4]. Time and frequency domain analysis of heart rate variability (HRV) has been proven effective in describing alterations of ANS
control mechanisms and in identifying patients with increased cardiac and arrhythmic mortality [4-9].
Studies on HRV analysis have shown a relationship between sympathovagal dysfunction and the occurrence of atrial fibrillation (AF) [3-7] or ventricular tachycardia (VT) [8-9], suggesting a potential role of HRV as an earlier marker of disease. However, studies analyzing variability of the length of individual heart cycles in implantable cardioverter defibrillators (ICD) stored data led to different results [10-12].
Moreover, inappropriate or unnecessary ICD discharges remain an important clinical problem as they cause pain, psycho-social and sometimes proarrhythmic effects [13]. The discrepancies might be caused by different methods for HRV analysis, but also by the study design – the majority of the methods describing the ANS control mechanisms or cardiac function using only cardiac electrical signals analysis.
Evidences showing that monitoring arterial pressure and heart rate may allow for early diagnosis of arrhythmia long before manifestation of its clinical symptoms are now emerging [14-16]. These multivariable studies adjusted for age, hypertension requiring treatment, and mean arterial pressure have suggested also that pulse pressure can predict future development of AF. The purpose of this study was to evaluate the relationship between blood pressure and cardiac electrical activity of the heart, by cross spectral algorithm, HRV and blood pressure variability (BPV) through spectral analysis, before, during, and after a self-terminating cardiac arrhythmia.
In this study, our team has developed a rabbit model for electrophysiological research related with ANS role on atrial fibrillation inducibility. In this model, atrial fibrillation has been provoked by a 50Hz pacing of atria, or by vagal and sympathetic electrical stimulation. During experimental work, one rabbit (from 22) developed a cardiac arrhythmia for about 7 minutes, before experimental procedure for ANS stimulation and high frequency pacing. The present work discuss the ability and importance of wavelet analysis, cross-spectral and transfer function algorithm to follow the short-term cardiovascular function changes before, during, and after self-terminating cardiac arrhythmia in this rabbit.
II. METHODS
A. Study design
The details of animal model preparation to study the mechanism of atrial fibrillation (AF) are published elsewhere [5]. Briefly, the New Zealand white rabbit was anaesthetized, paralyzed, and artificially ventilated. After a medial thoracotomy, a bipolar electrode was placed on the right cervical vagus nerve for parasympathetic electrical stimulation whereas a concentric bipolar electrode was inserted in the intermediolateral cell column of the spinal cord, at T1 level, for sympathetic stimulation. A set of recording electrodes was also placed on the atrial epicardium and in the area surrounding the pulmonary veins. Blood pressure, ECG and atrial electrograms were recorded throughout the experiment.
ECG and blood pressure were acquired at 1kHz (PowerLab, AD Instruments). Data were analyzed before, during and after the arrhythmia in the following way: 1 minute of baseline in the beginning of recording, 2 minutes before arrhythmia onset, 2 minute during the arrhythmia and 2 minute after the termination of arrhythmia. Intervals between changes in cardiac electrical activity were detected using dynamically varying thresholds in order to identify R wave or AF or ventricular tachycardia (VT) I wave. Frequency and time-frequency analysis of HRV and BPV power spectra were estimated for R-R or I-I intervals and systolic blood pressure signals. The subtraction of a linear trend from the data was realized using a multiresolution wavelet decomposition algorithm. The signals were decomposed using db10 mother wavelets and reconstructed without coefficient of decomposition corresponding to band frequencies between 0-0.05 Hz. The fast Fourier transform (FFT), continuous wavelet decomposition (CWT) and complex wavelet transform were used to estimate the power spectra.
B. Cross spectrum
Cross-spectrum were calculated between the blood pressure variables (SAP) and the R-R or I-I intervals.
The power spectrum for blood pressure signal xn and pressure variation leads the interval variation at this frequency; for a positive phase the reverse holds. If the coherence is low for a certain frequency, the phase at this frequency cannot be estimated reliably. In the phase spectra, lines between successive values were suppressed if the difference was larger than 180 because in these cases the phase has passed the -180 border to reappear at +180, or vice versa. If no such method is used, confusing vertical lines may appear in the phase spectra.
When the observation is based entirely on frequency analysis, information on dynamically varying or “short time” dependence between the signals or the temporal structure of coherence, which is useful in the study of cardiovascular dynamics, is not obtained. Wavelets combine high temporal resolution with good frequency resolution and offer a reasonable balance between these parameters. The continuous wavelet transform (CWT) can decompose a signal into a set of finite basis functions. Wavelet coefficients Wx (a, ) are produced through the convolution of a mother wavelet function (t) with the analyzed signal xn where a and denote the scale and translation parameters. Daubechies mother wavelet was applied to SAP, R-R or I-I intervals in order to estimate time-frequency changes in blood pressure and cardiac signals before, during and after arrhythmia using CWT. The CWT has edge artifacts because the wavelet is not completely localized in where 0 is the wavelet central pulsation.
The Gaussian’s second order exponential decay of the Morlet function gives a good time and frequency localization. We choose the complex WT (cMWT) as it provide the signal amplitude and phase simultaneously and permits investigation on the coherence/synchronization between two signals. The cross wavelet transform of two linearities, loss of time invariance or other system inputs signals. Less organized phase spectrum of cardiovascular signals recorded during arrhythmia is observed both in frequency and time-frequency domain representation (Figs. 2 and 3).
Figure 1. Coherence and phase representation of cross spectrum between SAP and cardiac signal recorded before initiation of arrhythmia in the frequency domain. From top to bottom can be observed decrease in coherence and increase variation of phase spectrum.
Figure 4. Wavelet coherence and phase of blood pressure and cardiac signal recorded during initiation of arrhythmia. Rapidly changes in phase angle and directionality in LF and VLF bands can be observed.
IV. CONCLUSION
This work shows the power of the cross spectrum and time- frequency analysis in the study of features of arrhythmia. As far as we know, this is the first study that used the analysis of BPV and transfer function for arrhythmia characterization and forecasting. Our results show that BPV and cardiovascular coupling characterization may be used to predict the onset of arrhythmic episodes. In addition, using the proposed algorithm, valuable information on mechanisms of certain cardiac arrhythmia with reference to respiration and blood pressure changes can be better understood.