This action might not be possible to undo. Are you sure you want to continue?

6, September 2010

**Empirical Mode Decomposition Analysis of Heart Rate Variability
**

C.Santhi.M.E., Assistant Professor, Electronics and Communication Engineering, Government College of Technology, Coimbatore-641 013 N.Kumaravel Ph.D

Professor, Head of the Department, Electronics and Communication Engineering, Anna University,Chennai-600 025. Abstract The analysis of heart rate variability (HRV) demands specific capabilities not provided by either parametric or nonparametric spectral estimation methods. Empirical mode decomposition (EMD) has the possibility of dealing with nonstationary and nonlinear embedded phenomena, for a proper assessment of dynamic and transient changes in amplitude and time scales of HRV signal. In this work EMD and a non-linear curve fitting technique are used to study half an hour HRV signal and its intrinsic mode function obtained from 20 healthy young control subjects, 20 healthy old control subjects and 20 subjects with long term ST. The intrinsic oscillations are measured by means of its meanperiod and variance. Significant meanperiod reduction is observed in the intrinsic time scales of healthy old control subjects and subjects with long term ST, which is used to classify the three groups of HRV signal with high sensitivity and specificity. The estimated slope using the non-linear curve fitting technique represents the flexibility of the cardiovascular system. The main advantage of this method is it does not make any prior assumption about the HRV signal being analyzed and no artificial information is introduced into the filtering method. Index Terms- Empirical Mode Decomposition, Heart Rate Variability, Intrinsic Mode Functions, RR intervals, nonlinear curve fitting. 1. Introduction Over the last 20 years there has been widespread interest in the study of variations in the beat-to-beat interval of heart known as heart rate variability (HRV) or RR interval variations. HRV has been used as a measure of mortality primarily with patients who have undergone cardiac surgery. Clinical depression strongly associated with mortality with such patients may be seen through a decrease in HRV [1]. HRV is a non invasive measure of autonomic nervous system balance. Heart rate is influenced by both sympathetic and parasympathetic (vagal) activities of ANS. The influence of both branches of the autonomic nervous system (ANS) is known as sympathovagal balance reflected in the RR interval changes. A low frequency (LF) component provides a measure of sympathetic effects on the heart and generally occurs in a band between 0.04 Hz and 0.15 Hz. A measurement of the influence of the vagus nerve in modulating the sinoatrial node can be made in the high frequency band (HF) loosely defined between 0.15 and 0.4 Hz known as respiratory sinus arrhythmia (RSA), and is a measure of cardiac parasympathetic activity. The ratio of power in the LF and HF bands (LF/HF) provides the measure of cardiac sympathovagal balance. Empirical Mode Decomposition (EMD) retains the intrinsic nonlinear nonstationary property of the signal. Any intrinsic timescale derived from the signal is based on the local characteristics timescale of the signal [2-4]. EMD carries out layer upon layer sifting and obtains ordered array components from smallest scale (highest frequency) to largest scale (lowest frequency) [4]. Empirical mode decomposition has the possibility of dealing with nonstationary and nonlinear embedded phenomena, and owing to its suitability for a proper assessment of the dynamic and transient changes in amplitude and in frequency of the HRV components [2& 3]. Application of EMD to half an hour HRV data yields nine intrinsic mode functions (IMFs). The first scale represents the highest frequency or the shortest period component of the signal. The second scale represents the lower frequency or the longer period component of the signal. Similarly the last IMF represents the lowest time scale present in the HRV signal. The first two scales contain more than 85% of total signal power. The meanperiod and variance of IMFs are computed as time domain measures. The variance of IMF decreases exponentially with respect to increasing timescales (meanperiods). Using nonlinear curve fitting technique the IMFs variations are estimated. The estimated parameter represents the flexibility of the cardiovascular system.The methodology is applied to HRV signal obtained from 20 healthy young control subjects, 20 healthy old control subjects and 20 subjects with long term ST. The intrinsic time scale of IMF 2 classifies the three groups HRV signal with high sensitivity and specificity. 2. Empirical Mode Decomposition (EMD) EMD is a procedure oriented adaptive method for decomposing non-linear non-stationary signals. The components resulting from EMD are called Intrinsic Mode Functions (IMFs) [2]. The IMFs are amplitude frequency modulated intrinsic signals. The IMF’s represents the oscillatory modes imbedded in the signal. It should satisfies the following two conditions. 1. In the whole data set the number of extrema’s and the number of zero crossings must be either equal or differ by at most one. 2. At any point the mean value of the envelope defined by the local minima and the envelope defined by the local maxima is zero.

255

http://sites.google.com/site/ijcsis/ ISSN 1947-5500

**(IJCSIS) International Journal of Computer Science and Information Security, Vol. 8, No. 6, September 2010
**

Heart rate variability 1.8

Intrinsic mode functions 0.2 0 -0.2 0.2 0 -0.2 0.2 0 -0.2 0.2 0 -0.2 0.2 0 -0.2 0.1 0 -0.1 0.1 0 -0.1 0.05 0 -0.05 0.01 0 -0.01 1.4 1.3

0

200

400

600

800

1000

1200

1400

1.7

0

200

400

600

800

1000

1200

1400

1.6

0

200

400

600

800

1000

1200

1400

1.5

0

200

400

600

800

1000

1200

1400

rr intervals

1.4

0

200

400

600

800

1000

1200

1400

1.3

0

200

400

600

800

1000

1200

1400

1.2

0

200

400

600

800

1000

1200

1400

1.1

0

200

400

600

800

1000

1200

1400

1

0

200

400

600

800

1000

1200

1400

0.9

0

200

400

600 800 beat number

1000

1200

1400

0

200

400

600

800

1000

1200

1400

**Fig.1.RR interval signal
**

Reconstructed signal 1.8 1.7

**Fig.2. Intrinsic Mode Functions
**

Detrended HRV signal 0.3 0.2

1.6

0.1

1.5

0

rr intervals

1.4

-0.1

1.3

-0.2

1.2

-0.3

1.1

Step 5: Check h(t) for the conditions of an Intrinsic Mode Functions. [2] If h(t) is an IMF compute residue r(t)=x(t)-h(t) and again the process is repeated to extract the next IMF. If h(t) is not an IMF x(t) is replaced with h(t) and the procedure is repeated from step 1. Fig.6 shows all IMFs of the signal x(t). The process ends when the range of residue is below a predetermined level or the residue has a monotonic trend. In order to guarantee that the IMF components retains enough physical sense in both amplitude and frequency modulations, the sifting process is stopped by limiting the size of standard deviation(SD) which is computed from two consecutive sifting results.

1

-0.4

0.9

-0.5

0 200 400 600 800 beat number 1000 1200 1400

0

200

400

600

800

1000

1200

1400

Fig.3.Reconstructed signal

Fig.4.Detrended signal

SD = ∑ [ h1( k −1) (t ) − h1k (t ) / h 21( k −1) (t )] (1)

2 t =0

T

Figs 1-4 explain the efficiency of EMD for RR interval signal. The ECG data has been collected from the biomedical website [7] http://www.physionet.org. The RR intervals are derived from half an hour ECG signal by identifying the QRS complexes. The signal is manually edited and only noise free ectopic free segments are used for the analysis. A real time RR interval signal and its EMD decomposed IMFs are shown in Fig.1&2. Application of EMD to real time RR interval signal identifies eight to nine IMFs. The IMFs are zero mean amplitude frequency modulated signal. The decomposition is adaptive and lossless. The original RR interval signal is reconstructed using decomposed IMFs (Fig.3). The nonstationary trend is removed when the residue or monotonic trend (last IMF) is omitted while reconstructing the signal (Fig.4). From the RR intervals the HRV signal or ∆RR signal (Ri+1-Ri) is obtained by computing successive difference between consecutive RR intervals. The obtained HRV signal and its IMFs are shown in Fig.5 and Fig.6. Matlab 7.1 tools are used for the analysis. 3. Methodology SIFTING ALGORITHM:

Heart rate variability signal 0.6

where k represents number of siftings. The process of finding an intrinsic mode function requires number of iterations and the process to find all the IMFs requires further more iterations. As a result of this iterative procedure finally yields many IMFs and a residue. By summing up all the IMF functions and the residue, the original signal is reconstructed, given by the mathematical formulae

X (t ) = ∑ hi (t ) + r (n)

i =1

n

(2)

Where each hi represents an intrinsic mode function and r(n) either a mean trend or a constant. For each IMF the meanperiod and variance are computed. The meanperiod is the ratio of distance between the first and last zero-crossings to number of zero-crossings of IMF. The obtained RR interval signal using ECG represents the response of the cardiovascular system to ANS activities not the ANS activities themselves. The characteristics of cardiovascular system determine how the system responds to ANS activity and can alter significantly the characteristics of the HRV signal. The response characteristics are often nonlinear in nature. The IMFs capture the all the variations present in the HRV signal. Plotting the variance of all IMFs against its meanperiods gives a nonlinear function. The variance of IMF decreases with increasing meanperiod and this behavior is approximated using a geometric function (3) Y= aXb where Y represents vector of IMF’s variance,. X represents vector of meanperiods of IMFs, a is constant and b is the exponential decrease of the function. The IMFs meanperiod and variance of healthy young control subjects, healthy old control subjects and long term ST subjects vary significantly. The variations in the IMF are quantified by the parameter b. The parameter b represents the flexibility of cardiovascular system to ANS activities. The parameter b is estimated using nonlinear curve fitting technique explained below.

Intinsic mode functions of HRV signal 0.5 0 -0.5 0.2 0 -0.2 0.1 0 -0.1 0.05 0 -0.05 0.05 0 -0.05 0.02 0 -0.02 0.02 0 -0.02 2 0 -2

0

200

400

600

800

1000

1200

1400

0.5

0

200

400

600

800

1000

1200

1400

0.4

0

200

400

600

800

1000

1200

1400

0.3

0.2

0

200

400

600

800

1000

1200

1400

0.1

0

200

400

600

800

1000

1200

1400

0

0

200

400

600

800

1000

1200

1400

-0.1

0 -3 x 10 0 -3 x 10 0

200

400

600

800

1000

1200

1400

-0.2

200

400

600

800

1000

1200

1400

5 0 -5

-0.3

200

400

600

800

1000

1200

1400

-0.4

0.01 0 -0.01

0 200 400 600 800 1000 1200 1400

0

200

400

600

800

1000

1200

1400

Fig.5.HRV signal

Fig.6. Intrinsic Mode Functions

Step 1: All the minima and maxima of the HRV Signal x (t), are located. Step 2: Spline Interpolate the minima and maxima points to obtain lower and upper envelopes of the signal. Step 3: Compute mean envelope m (t)=(maxima’s+minima’s)/2. Step 4: Subtract local mean from the original Signal to obtain local details h(t)=x(t)- m(t).

256

http://sites.google.com/site/ijcsis/ ISSN 1947-5500

(IJCSIS) International Journal of Computer Science and Information Security, Vol. 8, No. 6, September 2010

**Taking logarithm of equation (3), gives ln Y = ln a + b ln X
**

* * *

**0.15Hz; 3. Very low frequency band from 0.01Hz to 0.04Hz.. (4)
**

0.9 0.8 0.7 0.6 0.5 Relative power 0.4 0.3 0.2 0.1 0 1 3 5 7 9 11 13 15 17 19 Healthy young control records

Relative powers of IMF 1 and IMF 2

putting Y =ln Y , X =ln X , A = ln a then the above nonlinear equation becomes Y* = A* + bX* which is a linear equation in X. The corresponding normal equations are

Relative powers of IMF 1 Relative powers of IMF 2

∑Y

*

= NA* + b∑ X *

(5)

*2

∑X

*

Y =A

*

*

∑X

*

+ b∑ X

Nonlinear response of IMFs

**Fig 8: Relative powers of IMF 1 and IMF 2 (6)
**

IMF1 0.01 4 3 0.005 2 1 x 10

-3

IMF2

0.9

0

0 x 10

-3

0.1

0.2 IMF3

0.3

0.4

0.5

0

0 x 10

-4

0.1

0.2 IMF4

0.3

0.4

0.5

0.8

2

0.7

8 6 4 2

1.5 1

0.6

0.5 0 0 x 10

-4

0.5

0.1

0.2 IMF5

0.3

0.4

0.5

0

0 x 10

-4

0.1

0.2 IMF6

0.3

0.4

0.5

0.4

8 6

8 6 4 2

0.3

4

0.2

2 0 0 0.1 0.2 0.3 0.4 0.5

0

0.1

0

0.1

0.2

0.3

0.4

0.5

0

0

20

40

60

80

100

120

140

160

180

Fig.7. Curve fitting N represents number of IMFs. Solving the normal equations using least mean square method the variables ‘a’ and ‘b’ are estimated [5]. The simulated response function using the estimated parameter is shown in Fig.7. 4. Results and Discussion EMD and curve fitting techniques are applied to half an hour HRV signal of 20 healthy young control subjects, 20 healthy old control subjects and 20 subjects with long term ST. Empirical mode decomposition adaptively decomposes the half an hour HRV signal into number of Intrinsic mode functions (Fig.6). The analysis is done with ∆RR intervals. ∆RR (Ri+1-Ri) represents the difference between successive beat intervals. The IMFs are measured by their absolute variance, relative variance and meanperiods. The meanperiod is the ratio of distance between the first and last zero-crossings to the number of zero-crossings of IMF. First 3 IMFs contains more than 92% of total variance. The IMF1 represents the highest frequency or the shortest period component of the signal. The IMF2 contains the lower frequency or the longer period component of the signal. Since the 1st and 2nd IMF contains more than 85% of total power they are very significant. Relative powers are computed with respect to total power considering all IMFs except the residue with zero meanperiod. In healthy young subjects an increase in relative power of IMF1 decreases the relative power of IMF2 (Fig.8). IMF 1 and IMF 2 are in phase opposition representing different components of the HRV signal. The original signal is interpolated to 2 Hz for a meaningful frequency measure. The Welch periodogram (with window width 1024 and window overlap of 512 samples) of IMFs of a healthy young control subject are shown in Fig.(9). Table1 gives the peak frequency(Hz) and absolute spectralpower (ms2-miliseconds square) of IMFs The figure shows the frequency spectrum of the IMFs falls in the recognized spectral bands of HRV signal: 1.High frequency band from 0.15Hz to 0.5Hz; 2. Low frequency band from 0.04Hz to

Fig 9: Welch periodogram of IMFs IMFs Peak frequency Hz IMF1 IMF2 IMF3 IMF4 IMF5 IMF6 0.2891 0.13 0.068 0.03 0.021 0.01 0.01 0.003 0.002 0.00069 0.0007 0.00062 in Peak power in ms2

Table-1 Spectral values of IMFs The meanperiod of IMF2 of healthy young controls subjects are significantly higher compared to healthy old controls subjects and subjects with long term ST. Considering meanperiod of IMF2 (2.9724 secs) as threshold value, we classified the healthy young control subjects and subjects with long term ST with 95% sensitivity and 90% specificity. The classification is shown in Fig. (10). A threshold value of 2.8 secs classifies the healthy old controls subjects and subjects with long term ST with 90% sensitivity and 70% specificity shown in Fig .(11).

IMF 2 meanperiod of healthy young and subjects with long term st

4 3.5 3 2.5 2 1.5 1 0.5 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 Healthy young Sub. With long term st

Fig.10. Meanperiod comparison of healthy young subjects and subjects with longterm ST.

257

http://sites.google.com/site/ijcsis/ ISSN 1947-5500

**(IJCSIS) International Journal of Computer Science and Information Security, Vol. 8, No. 6, September 2010
**

IMF 2 meanperiod of healthy old and sub.with long term st

4.5 4 3.5 3 2.5 2 1.5 1 0.5 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 Healthy old controls sub.with long term st

Fig.11. Meanperiod comparison of healthy old subjects and subjects with longterm ST. The parameter b of IMFs of the three groups HRV signal are estimated, average plots are shown in Fig. (12). The estimated parameter b of healthy young control subjects, healthy old control subjects and long term ST subjects are 1.49, -1.43 and -1.39 (average values only). The more negative value represents the flexibility of the system. The healthy young control subject’s cardiovascular system is more flexible than healthy old subjects and longterm ST subjects. The flexibility of the system decreases in healthy old control subjects and longterm ST subjects. The absolute powers of healthy young control subjects are significantly higher compared to healthy old subjects and long term ST subjects as shown in Fig.(13) (average values only). The higher values of absolute power represent more fluctuating power in the signal. The results show the HRV of healthy young control subjects contains higher power, longer time scales and more adaptive to ANS activities compared to healthy old control subjects and subjects with long term ST.

Meanperiod vs relative powers of IMFs 0.7

rate variability analysis”, Med.Bio.Eng.Comput., 2001, 39, 471-479. [3]E.P.Souza Neto, M.A.Custaud, J.C.Cejka,P.Abry, J.Frutoso, C.Gharib, P.Flandrin, “Assessment of Cardiovascular Autonomic Control by the Empirical Mode Decomposition”, Methods Inf Med 2004;43:60-5. [4] N.E.Huang, Z.Shen, S.R.Long, M.C.Wu, H.H.Shih, etal.1998. “The empirical mode decomposition and the Hilbert spectrum for nonlinear and nonstationary time series analysis” Proc.R.Soc.A, vol 454, pp.903-995. [5] B.V.Ramana, “Higher Engineering Mathematics”, Tata McGraw-Hill Publishing Company Limited,New Delhi. [6] HRV Analysis Software 1.1, developed by The Biomedical Signal Analysis Group, Department of Applied Physics, University of Kuopio, Finland. http://venda.uku.fi/research/biosignal [7] www.physionet.org. [8] Jan W.Kantelhard, Stephan A, Armin Bunde, 2002, Multifractal Detrended Fluctuation Analysis of Nonstationary Time Series, Physica A 316, 87-114.

Absol ut e powe r s I M F1& I M F2

0.6 * healthy young controls o healthy old controls + longterm st subjects

0.5

0.004 0.003 Absol ut e power s

R elative pow ers

0.4

0.002 0.001 0 1 2 3 Absolute power of IMF1 Absolute power of IMF2

0.3

0.2

0.1

1. Y oung 2. Ol d 3. Long t er m ST

0 0 50 100 150 200 Meanperiod 250 300 350

Fig.12 Correlation graphs Fig.13. Absolute powers 5. Conclusion In order to cope up nonlinearity and nonstationarity issue of HRV signal EMD and nonlinear curve fitting techniques are used in this work. The IMFs of HRV signal are negatively correlated. The frequency spectrum of first two IMFs falls in the recognized HF and LF spectral bands of HRV signal. The meanperiod of IMF2 classifies half an hour HRV signal of healthy young control subjects, healthy old control subjects and subjects with long term ST with high sensitivity and specificity. The nonlinear curve fitting technique estimates the flexibility of cardiovascular system. The method is simple, adaptive and no artificial information is introduced in the analysis. 6. References [1] R. M. Carney, J. A. Blumenthal, P. K. Stein, L. Watkins, D. Catellier, L. F. Berkman, S. M. Czajkowski, C. O'Connor, P. H. Stone, K. E.Freedland, “Depression, Heart Rate Variability, and Acute Myocardial Infarction,” Circulation, vol. 104, no. 17, pp. 2024 – 2028, 2001. [2] J.C.Echeverria, J.A.Crowe, M.S.Woolfson, B.R.HayesGill, “Application of empirical mode decomposition to heart

258

http://sites.google.com/site/ijcsis/ ISSN 1947-5500

- Journal of Computer Science IJCSIS March 2016 Part II
- Journal of Computer Science IJCSIS March 2016 Part I
- Journal of Computer Science IJCSIS April 2016 Part II
- Journal of Computer Science IJCSIS April 2016 Part I
- Journal of Computer Science IJCSIS February 2016
- Journal of Computer Science IJCSIS Special Issue February 2016
- Journal of Computer Science IJCSIS January 2016
- Journal of Computer Science IJCSIS December 2015
- Journal of Computer Science IJCSIS November 2015
- Journal of Computer Science IJCSIS October 2015
- Journal of Computer Science IJCSIS June 2015
- Journal of Computer Science IJCSIS July 2015
- International Journal of Computer Science IJCSIS September 2015
- Journal of Computer Science IJCSIS August 2015
- Journal of Computer Science IJCSIS April 2015
- Journal of Computer Science IJCSIS March 2015
- Fraudulent Electronic Transaction Detection Using Dynamic KDA Model
- Embedded Mobile Agent (EMA) for Distributed Information Retrieval
- A Survey
- Security Architecture with NAC using Crescent University as Case study
- An Analysis of Various Algorithms For Text Spam Classification and Clustering Using RapidMiner and Weka
- Unweighted Class Specific Soft Voting based ensemble of Extreme Learning Machine and its variant
- An Efficient Model to Automatically Find Index in Databases
- Base Station Radiation’s Optimization using Two Phase Shifting Dipoles
- Low Footprint Hybrid Finite Field Multiplier for Embedded Cryptography

Abstract - The analysis of heart rate variability (HRV) demands specific capabilities not provided by either
parametric or nonparametric spectral estimation methods. Empirical mode decomposition (E...

Abstract - The analysis of heart rate variability (HRV) demands specific capabilities not provided by either

parametric or nonparametric spectral estimation methods. Empirical mode decomposition (EMD) has the possibility of dealing with nonstationary and nonlinear embedded phenomena, for a proper assessment of dynamic and transient changes in amplitude and time scales of HRV signal. In this work EMD and a non-linear curve fitting technique are used to study half an hour HRV signal and its intrinsic mode function obtained from 20 healthy young control subjects, 20 healthy old control subjects and 20 subjects with long term ST. The intrinsic oscillations are measured by means of its meanperiod and variance. Significant meanperiod reduction is observed in the intrinsic time scales of healthy old control subjects and subjects with long term ST, which is used to classify the three groups of HRV signal with high sensitivity and specificity. The estimated slope using the non-linear curve fitting technique represents the flexibility of the cardiovascular system. The main advantage of this method is it does not make any prior assumption about the HRV signal being analyzed and no artificial information is introduced into the filtering method.

Index Terms- Empirical Mode Decomposition, Heart Rate Variability, Intrinsic Mode Functions, RR intervals, nonlinear curve fitting.

parametric or nonparametric spectral estimation methods. Empirical mode decomposition (EMD) has the possibility of dealing with nonstationary and nonlinear embedded phenomena, for a proper assessment of dynamic and transient changes in amplitude and time scales of HRV signal. In this work EMD and a non-linear curve fitting technique are used to study half an hour HRV signal and its intrinsic mode function obtained from 20 healthy young control subjects, 20 healthy old control subjects and 20 subjects with long term ST. The intrinsic oscillations are measured by means of its meanperiod and variance. Significant meanperiod reduction is observed in the intrinsic time scales of healthy old control subjects and subjects with long term ST, which is used to classify the three groups of HRV signal with high sensitivity and specificity. The estimated slope using the non-linear curve fitting technique represents the flexibility of the cardiovascular system. The main advantage of this method is it does not make any prior assumption about the HRV signal being analyzed and no artificial information is introduced into the filtering method.

Index Terms- Empirical Mode Decomposition, Heart Rate Variability, Intrinsic Mode Functions, RR intervals, nonlinear curve fitting.

- EMPIRICAL MODE DECOMPOSITION AN INTRODUCTION
- Design Issues in Vital Sign Monitoring Through IR UWB Radar
- EEE223 Lec5 Handouts
- NI Tutorial 3548 En
- 3
- Software biofeedback
- Siemens Videomed Dic Manual
- polar y ecg
- Image Reject Receivers
- Echo Cancellation Algorithms using Adaptive Filters: A Comparative Study
- Noise cancellation
- L14 Procedure
- Signals And Systems Problems
- Adapt 1
- 04766761
- 29 - Final_exam
- 11761_cc1
- note4.2
- Wi Breathe
- Simulation of Adaptive Noise Canceller for an ECG signal Analysis
- MOD4_Discretesignals
- KaslovskyMeyer-EMD-postprint
- Van Bedaf, L.R. - MB2014
- Lab1
- lecture04-05-06-dataandsignal-100412095156-phpapp01
- google
- Fpga Implementation of Noise Cancellation Using Adaptive Algorithms
- Messages of RSPS-2013
- Noise Canceling in Audio Signal With Adaptive Filter
- 92478550-EC2204-NOL
- Empirical Mode Decomposition Analysis of Heart Rate Variability

Are you sure?

This action might not be possible to undo. Are you sure you want to continue?

We've moved you to where you read on your other device.

Get the full title to continue

Get the full title to continue reading from where you left off, or restart the preview.

scribd