Doi:10.1016/j.colsurfb.2005.01.014

Colloids and Surfaces B: Biointerfaces 42 (2005) 131–135 Extracting and analyzing sub-signals in heart rate variability a College of Bio-information, Chongqing University of Posts and Telecommunications, Chongqing 400065, PR China b Department of Biomedical Engineering, Chongqing University of Medical Sciences, Chongqing 400016, PR China Received 29 November 2004; accepted 24 January 2005 Abstract
A new statistical signal processing method, which was called independent component analysis (ICA) was used to extract sub-signals of heart rate variability (HRV). Ten healthy volunteers (4F, 6M) were involved in this study. Electrocardiogram (ECG) recording was consistedof 6 min when the volunteer was lying and another 6 min when the volunteer was standing. HRV was extracted from ECG. According totime-delay, HRV was divided into five groups as mixed signals. Five signals were reconstructed into two groups by ICA and the rebuilt twosignals were transformed by Fourier transformation. Results showed that one group signal component centralized in low frequency (calledIC1); the other did in high frequency (called IC2). The power of IC1 was significantly increased (P < 0.05) while that of IC2 had no significantchange (P > 0.05) and the ratio of IC1 to total power was significantly increased (P < 0.01) from lying to standing. Comparing the two posturalresults, it shows that IC1 may express sympathetic activity, and IC2 represents parasympathetic activity. Sympathetic and parasympatheticnervous function can be evaluated respectively and quantificationally by data and graphs from the two decomposed components. As anelectro-physiological method, it can assist the investigation about the tension of autonomic nervous, myocardial bielectricity activity, as wellas myocardial cell membrane characters.
2005 Elsevier B.V. All rights reserved.
Keywords: Independent component analysis; Heart rate variability; Autonomic nervous system; Nervous tension; Excitability; permeability; Membrane 1. Introduction
ability of Ca2+. Researching the activity of ANS, especiallythat of CSNA and CPNA can derivate the features of my- The heart rate was controlled by two antagonistic parts ocardium, such as membrane surface tension, ionic concen- of the autonomic nervous system (ANS), which was named tration, myocardial cell membrane permeability, endothelial as the cardiac sympathetic (CSNA) and the parasympathetic cell excretion effect of arrhythmia drug The cell (CPNA) nervous Experimental evidence showed the membrane feature study could refer to papers of professor association between propensity for lethal arrhythmia in heart rate and signs of modified activities of ANS indicated the Therefore, information about the dynamics of CSNA and central role of CSNA and CPNA in cardiovascular regulation CPNA could be useful for analysis and diagnosis in clinical phenomena. It has been reported that several diseases may application.HRV is the time digital sequence of R wave to R wave (R–R) intervals of electrocardiogram (ECG) that a high sympathetic activity during myocardial infarction HRV has plenty of the information of ANS Usually greatly increases the probability of fatal cardiac arrhythmia it is assumed the spectral analysis of HRV can provide such an observer Spontaneous fluctuations in HRV On the other hand, the change of ANS excitability would have been separated into three spectral analysis bands: very result in the alteration of myocardial cell activity and perme- low-frequency (VLF) oscillations (range: 0.008–0.04 Hz),low-frequency (LF) fluctuations (range: 0.04–0.15 Hz), and high-frequency (HF) components (range: 0.15–0.4Hz). HF is Corresponding author. Tel.: +86 23 62460025.
E-mail address: cquptlibiao@yahoo.com.cn (L. Biao).
generally recognized to reflect parasympathetic modulation, 0927-7765/$ – see front matter 2005 Elsevier B.V. All rights reserved.
doi:10.1016/j.colsurfb.2005.01.014 L. Zhangyong et al. / Colloids and Surfaces B: Biointerfaces 42 (2005) 131–135 LF does not have any physiological significance and VLF Electrocardiogram acquisition: The subjects were calm is more controversial.In order to solve this controversy we and lay down on the back (supine position). ECG from CM5 have presented a novel method that was based on blind lead have been recorded continuously for 6 min and sampled source separation (BSS) to extract and analyze sub-signal at 300 Hz on Pentium III computer with an A/D board (analog from HRV. Vetter used linear BSS to separate heart rate (RR) to digital). After the 6 min lying down recording, the subject and arterial blood pressure (ABP) into two independent would stand up. When ANS came to a new equilibration, signals, which express the modulation of CSNA and CPNA However, HRV affected ABP by heart output, CSNA HRV acquisition: A software package developed by our and CPNA affected ABP also by hear output, the regulation Lab was used to identify R wave peak then R–R of CSNA and CPNA can not be simulated as linear model.
intervals were extracted. The time digital sequence made BSS did not fit the linear model. Since CSNA and CPNA from R–R intervals was HRV signal.
can regulate HRV separately in a very short time, we usedthe sequential HRV signal acquired by time delay as input 2.2. Independent component analysis (ICA) signal to eliminate the nonlinear effect when volunteerswere in stable status.Independent component analysis (ICA) is an attractive method developed rapidly in recent years. Itis a new statistic signal processing technique developed with x = As
blind source separation. ICA (Comon, 1994) was originallyproposed to solve the BSS problem The purpose of where both the sources s and the mixing matrix A are un-
ICA is to recover independent source signals that are linearly known and need to be extracted from the only known mea- mixed through an unknown process. The source signals surement x. It is usually further assumed that the dimensions
must be independent for ICA processing. The assumptions of x and s are equal. If A can be estimated, we can compute
of CSNA and CPNA regulating HRV are ensured linearly its inverse, say W, and obtain the independent component
mixed and independent model. ICA is a highlight in BSS and has made great progress in communications, speech recogni- s = Wx
tion, sonar, bio-signal processing, vibration and control.Theinnovative contributions of this paper are twofold: first is that Many algorithms have been proposed n fact, the aim a novel method to decompose HRV into two independent of all the algorithms is to maximize the non-Gaussianity of components was applied; second is that two components wTx, thus we can have one of the independent components. In
have been confirmed which reflects the activity of CSNA this paper, we use fast ICA algorithm developed by hyv¨arinen and CPNA. The paper is organized as follow: Section extract the independent components in ECG. Fast ICA briefly describes the signal acquisition and ICA principle.
algorithm is employing Eqs. let w converge,
The results are analyzed and discussed in Sections 2. Methods
w+ = E{xg(wTx)} − E{g (wTx)}w,
w+
Volunteers: 10 healthy adults, 4 females and 6 males, age w =
||w+||
Table 1ICA Data of 10 volunteers and results in statistics a The maximum value position of IC2.
L. Zhangyong et al. / Colloids and Surfaces B: Biointerfaces 42 (2005) 131–135 According to the method of acquiring signals, every vol- unteer had two HRV signals, which was in lying and standingpostures separately. Each HRV signal was divided into fivegroups according to time delay and every group had 60 RR in-tervals. Fast ICA algorithm decomposed the five groups HRVinto two independent components. The convergence preci-sion is 0.0001. The two decomposed components were trans-formed by fast Fourier transform (FFT) so that power valueswere obtained. The total HRV (300 RR intervals) was alsocalculated by FFT. The IC1 (=independent component 1) rep-resents the signal that is mainly consisted of low-frequencycomponents and the IC2 (=independent component 2) indi-cates the signal which has more high frequency components.
All the indices are shown in power of total HRVis the total power and it expresses the output power of heart.
The ratio is power of IC1 to total power.
Fig. 2. Power spectrum of ICs in lying. The power spectrum only has the The data between the two body postures was processed significance in comparison with the whitened signal.
by TTEST, and then P value was obtained. The physiologicsignificance was decided by the statistical results. The results The comparing research of the same volunteer in two were regarded as significant when the P value was less than with we can find that IC1 mainly includes low fre- All the data processing have been done in MATLAB R12.
quency component and has the trend to increase from ly-ing to standing, especially in abscissa from 1 to 10; on thecontrary, IC2 has the trend to decrease. In physiology the 3. Results and conclusion
regulation of ANS is: When man relaxes and lies down, hisburden of heart is little and heart rate is low; when changing ICA decomposed HRV into two independent compo- to stand, the heart rate becomes faster and faster, the activ- nents. IC1 was the low frequency component of HRV (see ity of sympathetic nerve increases, however the activity of and then IC2 was the high frequency component parasympathetic nerves will increase or decrease because of of HRV (see IC1 and IC2 were transformed by the antagonistic action between the two nerve fibers. The FFT. The results were shown in In the Figures, fluctuating phenomena of IC1 and IC2 were suitable to the the dot line is the power spectrum of IC2 and the solid line is activity of CSNA and CPNA. That is to say, IC1 could be used to express the modulation of sympathetic nervous sys- Fig. 1. One volunteer’s independent components (ICs) of HRV in lying. The Fig. 3. The same volunteer’s ICs of HRV in standing. The abscissa presents abscissa presents the RR intervals. The signal will be centered and whitened the RR intervals. The signal will be centered and whitened before doing before doing ICA, so that the ordinate has negative number.
ICA, so that the ordinate has negative number.
L. Zhangyong et al. / Colloids and Surfaces B: Biointerfaces 42 (2005) 131–135 Fig. 4. Power spectrum of ICs in standing. The power spectrum only has the Fig. 5. The average power spectrum of IC1 in two postures.
significance in comparison with the whitened signal.
tem and IC2 presents the activity of parasympathetic nervoussystem.
The viewpoint that IC1 and IC2 can express the modula- tion of ANS was showed in 10 volunteers’ data and P values(in Total power of HRV changed significantly fromlying to standing (P < 0.01). Total power expresses the auto-nomic nerve activity and the cardiac output. Different bodypositions have different autonomic nerve regulation in phys-iology. In the power of IC1 increases significantlyfrom lying to standing (P < 0.05); when the volunteer standsup, the tension of adrenergic nerve fiber increases, which isjust correspondence to the feature of IC1, so that the compo-nent can express the modulation of sympathetic nerve. Thesignificance of the ratio could also explain IC1 is the resultof CSNA modulation. We also know that cholinergic nervefibers is a tool of regulating psychic energy backlog and pro-tecting the heart, it has the feature of slowly reacting and Fig. 6. The average power spectrum of IC2 in two postures.
diffusion, then the power of component 2 has no specific sig-nificant variation (P = 0.1815) between two body postures.
of HRV for patients with arrhythmia or heart failure f a man has an FMECPNA and smaller IC2, does it suggest he just confirm the activity of parasympathetic nerve system. In will receive a heart failure? If the changes go to the other di- all, the signals observed by independent component analysis rection, does it indicate his heart function will become better? could be used to investigate autonomic nerve system function.
This will be a potentially valuable research field.
IC1 is a signal dominated by sympathetic nerve; and then IC2is a signal controlled by parasympathetic nerve system.
4. Discussion
phenomena, which the peak of IC2 has the trend to moveforward. We called this as “frequency movement effect of Information coming from the observed signal by ICA CPNA (FMECPNA)”. Thus, we draw the figures of the av- could be used to evaluate the function of autonomic nerve erage power spectrum of both IC1 and IC2 ( system, such as power spectrum and power integral (In and measured the wave peaks of IC2 from two body pos- tures of the same volunteer (data was shown in The respectively and quantificationally appraise the function of peak value of IC2 significantly moved forward after the T test sympathetic nerve and parasympathetic nerve. By compar- (P < 0.05). We can see from the average peak value ing the data of different time, the development of autonomic was around 15 in standing and 18 in lying posture. It has weak nerve system can be obtained. That ICA was used to analyze tension of CPNA and slightly higher frequency component heart rate variability is a successful approach.
L. Zhangyong et al. / Colloids and Surfaces B: Biointerfaces 42 (2005) 131–135 The experiment reached the aim of extracting the sub- [3] J. Andrich, T. Schmitz, C. Saft, et al., Autonomic nervous system signals of HRV, but the number of exponents was a little function in Huntington’s disease, J. Neurol. Neurosurg. Psychiatry bit smaller. A larger exponent will be used and divided into [4] G.X. Wang, S.X. Cai, X.C. Rao, et al., Endothelin-1 and angiotensin some groups with age. The results may be compared with II secretion at different lengths of endothelial cell monolayer in the the result of “Variation of chaotic and spectral features of view of tensile stress accumulation in the upper endothelial cell heart period signal with age” Some patients with ANS membrane, Colloid. Surf. B: Biointerfaces 27 (4) (2003) 345–353.
diseases (such as diabetes and asthma) should be included [5] F. Boury, J. Gautier, Y. Bouligand, et al., Interfacial properties of into the experiment to consummate the conclusion.
amiodarone: the stabilizing effect of phosphate anions, Colloid. Surf.
B: Biointerfaces 20 (3) (2001) 219–227.
The cell excitability, signal transduction and modulation, [6] B.C. Wang, H.C. Zhao, C.R. Duan, et al., Effects of cell wall calcium cell membrane surface tension and cell membrane adhesion on the growth of Chyrsanthemum callus under sound stimulation, energy microcosmic domain, but our investigation Colloid. Surf. B: Biointerfaces 25 (3) (2002) 189–195.
is macroscopic to clarify the relationship between cardiovas- [7] B.C. Wang, H.C. Zhao, Y.Y. Liu, et al., The effects of alternative cular and autonomic nerves system. If studying these two stress on the cell membrane deformability of chrysanthemum calluscells, Colloid. Surf. B: Biointerfaces 20 (4) (2001) 321–325.
domains at the same time, we can obtain their specific rela- [8] X.H. Xu, Z.X. Xie, L.C. Chen, et al., A computerized system ana- tionships. Moreover we can explain cell microcosmic phe- lyzing chaotic characteristics of heart period signal, Chin. J. Biomed.
nomena from electro-physiological approaches.
ICA is a newly developed promising approach for signal [9] Task Force. Heart rate variability, standards of measurement, phys- processing, and we use it in our study to extract sub-signal for iological interpretation and clinic use. Circulation 93 (5) (1996)1043–1065.
cardiovascular system. It is obvious that ICA has the ability to [10] Z.X. Xie, Y.H. Yin, Insight on physics, engineering, physiology separate independent component by the experiment. ICA will and clinics for analyzing HRV, in: Proceedings of the Asia–Pacific become a new effective tool to process bio-medical signal.
Congress on Biomedical Engineering, 2000, pp. 357–359.
[11] R. Vetter, N. Virag, J.M. Vesin, et al., Observer of autonomic cardiac outflow based on blind source separation of ECG parameters, IEEETrans. Biomed. Eng. 47 (5) (2000) 578–582.
Acknowledgment
[12] P. Comon, Independent component analysis: a new concept? Signal This work was supported by foundation of Chongqing [13] Z.X. Xie, Y.H. Yin, L.C. Chen, et al., Analysis for atrial fibrillation waves, in: Proceedings of the 20th Annual International Conference, 200400146), Chongqing Science Committee (No. [1999]- IEEE Eng. Med. Biol. Soc. 20 (1) (1998) 115–118.
[14] T.W. Lee, M. Girolami, T.J. Sejnowski, Independent component anal- 18-58) and Chongqing University of Medical Sciences (No.
ysis using an extended infomax algorithm for mixed sub-Gaussian CX200202). The authors wish to thank for the help of doctor and super-Gaussian sources, Neural. Comput. 11 (2) (1999) 417– Chen Huafu and professor Yao Dezhong from School of Life Science and Technology, University of Electronic Science [15] S. Amari, A. Cichocki, H.H. Yang, A new learning algorithm for blind signal separation, Adv. Neural Inform. Process. Syst. 8 (1996)757–763.
[16] A. Hyv¨arinen, Fast and robust fixed-point algorithms for independent component analysis, IEEE Trans. Neural Network 10 (3) (1999) References
[17] Z.Y. Lee, Z.X. Xie, X.H. Xu, et al., Variation of chaotic and [1] R. Vetter, J.M. Vesin, P. Celka, et al., Observer of human cardiac spectral features of heart period signal with age, in: Proceeding sympathetic nerve activity using noncausal blind source separation, of Asia–Pacific Congress on Biomedical Engineering, 2000, pp.
IEEE Trans. Biomed. Eng. 46 (3) (1999) 322–330.
[2] A. Pitala, T. Banach, K. Szmigiel, et al., The effect of chronic alco- [18] K.D. Tachev, J.K. Angarska, K.D. Danov, et al., Erythrocyte attach- hol use on heart rate variability, Folia-Med-Cracov. 41 (3/4) (2000) ment to substrates: determination of membrane tension and adhesion energy, Colloid. Surf. B: Biointerfaces 19 (1) (2000) 61–80.

Source: http://u.math.biu.ac.il/~louzouy/courses/seminar/ica1.pdf