IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 49, NO. 7, JULY 2002 651 Wavelet Decomposition of Cardiovascular Signals for Baroreceptor Function Tests in Pigs Urban Wiklund*, Member, IEEE, Metin Akay, Senior Member, IEEE, Stuart Morrison, and Urban Niklasson Abstract—In this paper, the discrete wavelet transform (DWT) was applied to analyze the fluctuations in RR interval and sys- tolic arterial pressure (SAP) recorded from eight -chloralose anesthetized pigs. Our aim was to characterize the autonomic modulation before and after cardiac autonomic blockade and during baroreflex function tests. The instantaneous power of decomposed low-frequency (LF) and high-frequency (HF) com- ponents was used for a time-variant spectral analysis. Our results suggested that transient events and changes in autonomic modu- lation were detected with high temporal resolution. A nonlinear relationship between RR interval and SAP during pharmacolog- ically induced changes in blood pressure was found, when the superimposed effect of respiratory sinus arrhythmia was removed. In addition, the baroslopes were nearly linear when both the LF and HF components were removed using DWT decomposition. Index Terms—Baroreflex, heart rate variability, HRV, spectral analysis, wavelet transform. I. INTRODUCTION T HE autonomic control of the cardiovascular system con- cerns the modulation of the cardiac pacemaker and the reg- ulation of the myocardial performance and the vascular system. An analysis of beat-to-beat variations in heart rate and blood pressure provides an important tool for understanding cardio- vascular regulation. One application of particular interest is the assessment of the baroreceptor function based on an analysis of the changes in RR interval and their corresponding changes in systolic arterial pressure (SAP). Several approaches for estimation of baroreflex sensitivity (BRS) have been suggested. The first method observes reflex changes in RR intervals following manipulation of systemic ar- terial pressures by injection of a vasoactive drug (e.g., phenyle- phrine or sodium nitroprusside) [1]. Baroreceptor function can then be estimated by the gain in the open-loop transfer func- tion using linear regression of SAP on the respective RR in- tervals during the stimulus [2]. A second approach is by spec- tral and cross-spectral analysis of the spontaneous fluctuations in RR intervals and SAP, e.g., by computing the ratio of the RR variability spectrum over the SAP spectrum [2]–[5]. A third Manuscript received January 18, 2001; revised February 12, 2002. Asterisk indicates corresponding author. *U. Wiklund is with the Department of Biomedical Engineering and Infor- matics and the Department of Clinical Physiology, University Hospital, SE-901 85 Umeå, Sweden (e-mail: Urban.Wiklund@vll.se). M. Akay is with the Thayer School of Engineering, Dartmouth College, Hanover, NH,03755 USA. S. Morrison was with the Department of Anesthesiology, University Hospital, SE-901 85 Umeå, Sweden. U. Niklasson is with the Department of Clinical Physiology, St. Göran Hos- pital, SE-112 81 Stockholm, Sweden. Publisher Item Identifier S 0018-9294(02)05772-5. technique, called the sequence method, applies linear regression over data sequences of at least three consecutive beats showing contemporaneous changes (increases or decreases) in SAP on RR interval [6], [7]. Baroreflex gain has also been estimated using autoregressive moving average (ARMA) analysis [8]. Spectral analysis of cardiovascular signals is often carried out using the fast Fourier transform (FFT) or autoregressive (AR) algorithms [9]. The problem with these methods is the assumption that signals are stationary, which in general only holds for short-term segments. Therefore, new techniques have been introduced for analysis of nonstationary cardiovascular signals [10]–[14]. Adapted wavelet transforms, which include wavelet packets and the discrete wavelet transform (DWT), have recently been used to analyze heart rate fluctuations. One approach has been to analyze the wavelet coefficients in the transform domain [15]–[18]. The DWT has also been used to decompose the cardiovascular fluctuations into the very low-frequency (VLF, centered near 0.04 Hz), low-frequency (LF, near 0.10 Hz) and high-frequency (HF, above 0.15 Hz) components [19], followed by parametric or Volterra series modeling [20] of each component. In this study we use the wavelet transform method to es- timate barorereflex sensitivity on data recorded from -chlo- ralose anesthetized pigs during controlled ventilation. In addi- tion, we discuss how the instantaneous power for decomposed components can be used to detect transient events or changes in cardiovascular signals after provocative maneuvers and cardiac autonomic blockade. II. METHODS In this section, we summarize the theory and implementation of the wavelet transform method used in this paper. A. Discrete Wavelet Transform Method The decomposition of the signal with the DWT is based on a partition in the frequency plane using a quadrature mirror filter (QMF) bank [21]. The filter bank consists of pairs of dig- ital high-pass and low-pass filters organized in a tree structure. The signal is decomposed at each scale into its detail (high-pass component) and approximation (low-pass) signals and down- sampled. The detail signal is then stored and the decomposition continues by filtering the approximate signal as the input signal for the next scale. At each scale, , the frequency axis is recur- sively divided into halves at the ideal cutoff frequencies (1) 0018-9294/02$17.00 © 2002 IEEE