Ahmed Bouziane, Benabdellah Yagoubi

WSEAS TRANSACTIONS on BIOLOGY and BIOMEDICINE Luis Vergara, Addisson Salazar

Study of the ANS Sympathovagal Behavior Using the ReliefF and the
KS-Segmentation Algorithms
AHMED BOUZIANE1,2, BENABDELLAH YAGOUBI1, LUIS VERGARA2,
ADDISSON SALAZAR2
1
Department of Electrical engineering
University of Mostaganem
P.O.Box 227 & 118, 27000 Mostaganem
ALGERIA
bouah@doctor.upv.es, yagoubibenabdellah@yahoo.com
2
Institute of Telecommunications and Multimedia Applications
Universitat Politècnica de València
Camino de Vera, s/n 46022 Valencia
SPAIN
lvergara@dcom.upv.es, asalazar@dcom.upv.es

Abstract: - The Autonomous Nervous System (ANS) sympathovagal balance was studied using several features
derived from Heart Rate Variability signals (HRV). The HRV signals are, however naturally, non-stationary
since their statistical properties vary under time transition. A useful approach to quantifying them is, therefore,
to consider them as consisting of some intervals that are themselves stationary. To obtain the latter, we have
applied the so called the KS-segmentation algorithm which is an approach deduced from the Kolmogorov-
Smirnov (KS) statistics. To determine, accurately, these features, we have used the ReliefF algorithm which is
one of the most successful strategies in feature selection; this step allows us to select the most relevant
features from thirty three features at the beginning. As result the ratio between LF and HF band powers of HRV
signal, the Standard Deviation of RR intervals (SDNN), and Detrended Fluctuation Analysis with Short term
slope (DFA α1), are more accurate for each stationary segment, and present the best results comparing with
other features for the classification of the three stages of stress in real world driving tasks (Low, Medium and
High stress).

Key-Words: - HRV, KS-segmentation algorithm, Wavelet packet, ANS sympatho-vagal balance evolution

1 Introduction experimental conditions involving human studies,

It is well established by many studies that have been gathered [6] to validate the following
heartbeat can evaluate the autonomic nervous three main assumptions:
system (ANS) [1, 2, 3]. The latter has become a) The respiratory rhythm of heart rate variability
increasingly popular because information on the (HF) is a marker of vagal modulation.
behavior of the autonomic nervous system b) The rhythm corresponding to vasomotor waves
(ANS) can be noninvasively inferred using and present in heart rhythm and arterial pressure
relatively simple signal processing methods [1]. A variability (LF) is a marker of sympathetic
stress; emotional or mental, may lead to an increase modulation of, respectively, heart rhythm and
in sympathetic activity and a decrease in vasomotion, and c) The ratio LF/HF is a marker of
parasympathetic activity of the ANS [4] which the state of the sympathovagal balance modulating
are, respectively, shown by energy concentration sinus node pacemaker activity [7,8,23].
peaks in low frequency (0.04 to 0.15 Hz) and high To study this sympathovagal balance, Fourier
frequency (0.15 to 0.4 Hz) ranges in the HRV power transform is, usually, used in spectral analysis to
spectral density (PSD) curve [5]. This results in compute the power spectral density (PSD) of the
increased strain on the heart as well as on the HRV [9, 10], but it does not depend on time; hence
immune and hormonal systems which influence the there is a lack of information such as the time when
activity and balance of the autonomic nervous
system (ANS). Many data, obtained in different

E-ISSN: 2224-2902 8 Volume 12, 2015

 Ahmed Bouziane, Benabdellah Yagoubi
WSEAS TRANSACTIONS on BIOLOGY and BIOMEDICINE Luis Vergara, Addisson Salazar

the sympathetic or the parasympathetic activity is datasets, but according to Yong et.al [27], it was
dominant, as well as how long this dominance lasts found that amongst the 17 data sets, only 7
during the test period. This is, particularly, drives datasets (drivers 6, 7, 8, 10, 11, 12, and
important when different events affecting the ANS 15) have clear mark identification and they can
are taking place and we need to link them to the be used in our analysis.
(ANS) activities. The mathematical tools such as the This database contains a collection of
short time Fourier transform (STFT) [16] or the multiparameter recordings from healthy volunteers,
wavelet transform that take into account both taken while they were driving on a prescribed route
frequency and time are, therefore, welcomed. including city streets and highways in and around
In contrast to the STFT which is a time- Boston, Massachusetts. The objective of the study
frequency representation, the continuous wavelet for which these data were collected was to
transform method is a dynamical representation [11- investigate the feasibility of automated recognition
15]. But there is, still, a correlation between the of stress on the basis of the recorded signals, which
scale parameter and its corresponding frequency. include ECG, EMG, GSR measured on the hand and
Despite the useful results provided by these two foot, and respiration[25].
methods [16, 17], difficulties, such as the scale From the duration of different driving segment
range of the adequate continuous wavelet and the obtained from Table.1, it was validated that the rest,
constant width of the STFT window, hampering highway, and city driving periods produce the low,
their applications, are, however, met in each of medium, and high levels of stress, respectively [26].
them. The driver database lacks the information
In this work, and based on our previous work regarding the duration of each Rest, City and
[22, 24], we have used a hybrid method in order to Highway driving task, but the same durations were
classify the three stages of stress in real world mentioned in [31]. The time intervals of the 7
driving tasks. Features were processed using the drivers used in this study (drivers 6, 7, 8, 10, 11, 12,
Relief algorithm [31] in small stationary segments and 15) are given in Table.1.
which were determined and detected, accurately, In this study seven ECG records were used to
using the KS-segmentation algorithm introduced test the algorithm. These recordings were sampled
recently by S. Camargo et al. [18]. at 496 Hz with a 16-bit rate resolution. Lead II from
each record is used here. No episodes have been
excluded from our analysis.
2 Material and Method Band pass filtering is an essential first stage of
any QRS detection algorithm. The purpose of band
pass filtering is to remove the baseline wander and
2.1 Material and data collection high frequencies which do not contribute to
The data analyzed in this work were obtained from QRS complexes detection. In this research, a band
Stress Recognition in Automobile Drivers from pass linear phase FIR digital filter with a Hamming
PHYSIONET website [25], the driver database is window in the frequency range between 3 and 40
originally collected by Healey & Picard from MIT Hz is used [28].
Media Lab [26]. In total there are 17 available

Table.1. Time intervals of driving test.

Drive No Driving period Total rec.
Initiale Rest City1 Highway1 City2 Highway2 City3 Final Rest time (mn)
Driver6 15.05 14.49 7.32 6.53 7.64 12.29 15.05 78.38
Driver7 15.04 16.23 10.96 9.83 7.64 10.15 15.03 84.87
Driver8 15 12.31 7.23 9.51 7.64 13.43 15.07 80.19
Driver10 15.04 15.3 8.66 5.27 7.04 12.06 14.79 78.15
Driver11 15.02 15.81 7.43 7.15 6.96 11.72 14.99 79.08
Driver12 15.01 13.41 7.56 6.5 8.06 11.68 15.01 77.23
Driver15 15 12.54 7.24 5.99 6.82 12.12 15 74.7

E-ISSN: 2224-2902 9 Volume 12, 2015

 Ahmed Bouziane, Benabdellah Yagoubi
WSEAS TRANSACTIONS on BIOLOGY and BIOMEDICINE Luis Vergara, Addisson Salazar

The extraction of R-peaks and correction of 1-The ECG signal is preprocessed using band-pass
artifacts from digitized ECG data were performed filter; later ARTiiFACT tool is used on filtered
using ARTiiFACT [29], which is a software tool ECG signal to enhance the presence of QRS
providing an efficient artifact detection algorithm. complexes and to detect R-Peaks and finally the
All extracted features of HRV signals were HRV signals are calculated.
calculated using the Kubios HRV Analysis 2.0 2-Segment each HRV signals into stationary
software [30]. The program calculates all the patches by applying the KS-segmentation
commonly used time and frequency domain algorithm.
parameters and the nonlinear Poincare plot. 3-Compute 33 features in each stationary segment
Advanced spectrum estimation methods and of the temporal signals using Kubios HRV Analysis
detrendingoptions are included as well. All records Software. Table.2 presents a detailed description of
all features.
were processed using MATLAB (R2012a
4- The next step is to calculate and extract the most
7.14.0.739).
relevant features; we have used the ReliefF
algorithm, this algorithm try to find features that
help separate data of different classes. If a feature
2.2 Hybrid method for studying the ANS has no effect on class-based separation, it can be
activity removed.
The suggested hybrid method to classify stress in These main steps of this suggested hybrid
order to study the ANS balance can be described by method are depicted in the following diagram in
the following steps: Fig.1.
Table.2. Description of the 33 extracted features.

Features Units Description

Mean RR [ms] The mean of RR intervals
STD RR (SDNN) [ms] Standard deviation of RR intervals
Mean HR [1/min] The mean heart rate
Time Domain

STD HR [1/min] Standard deviation of intantaneous HR values

RMSSD [ms] Square root of the mean squared differences
between successive RR intervals
NN50 Number of successive RR interval pairs
that differ more than 50 ms
pNN50 [%] NN50 divided by the total number of RR intervals
HRV triangular The integral of the RR interval histogram
index Divided by the height of the histogram
TINN [ms] Baseline width of the RR interval histogram
Peak frequency [Hz] VLF, LF, and HF band peak frequencies
Frequency

[ms2 ]
Domain

Absolute power Absolute powers of VLF, LF, and HF bands

Relative power [%] Relative powers of VLF, LF, and HF bands
Normalized power [n.u.] Powers of LF and HF bands in normalized units
LF/HF Ratio between LF and HF band powers
SD1, SD2 [ms] The standard deviation of the Poincare plot perpendicular to
(SD1) and along (SD2) th line-of-identity
ApEn Approximate entropy
SampEn Sample entropy
D2 Correlation dimension
Nonlinear

DFA α1 Detrended fluctuation analysis with Short term slope

DFA α2 Detrended fluctuation analysis with Long term slope
RPA Lmean [beats] Recurrence plot analysis of Mean line length
RPA Lmax [beats] Recurrence plot analysis of Maximum line length
RPA REC [%] Recurrence plot analysis of Recurrence rate
RPA DET [%] Recurrence plot analysis of Determinism
RPA ShanEn Recurrence plot analysis of Shannon entropy

E-ISSN: 2224-2902 10 Volume 12, 2015

 Ahmed Bouziane, Benabdellah Yagoubi
WSEAS TRANSACTIONS on BIOLOGY and BIOMEDICINE Luis Vergara, Addisson Salazar

Input ECG
Records Fs=496 Hz

FIR Filter
3-40 Hz

R-Peak Detection
Correct artifacts
Fig.2 Cardiac interbeat (RR) interval time series.

RR signal The RR interval signal is usually interpolated to

interpolated with recover an evenly sampled signal. Cubic
Fs=4Hz interpolation is adopted and the resampling
frequency of the interpolated signal is Fs=4Hz [21].
As mentioned above in the introduction, the
Segmentation of sympathetic and the parasympathetic activities of
Stationary fragments the ANS are empirically situated in low frequency
With KS-Algorithm (LF) and high frequency (HF) ranges, respectively
(ρ=0.05, l 0  300 ) in the Fourier space [22]. When these activities are
taking place, the HRV power spectral density curve
shows large concentration of energy in these
Compute 33 features for all stationary frequency ranges, and the sympathovagal balance
segments is, usually, estimated using the PSD in the LF to
that in the HF ratio (LF/HF). So, in addition of this
PSD ratio, we have computed 32 other features to
Select the most relevant 3 track the evolution of the ANS behavior as well as
features
the localization in time of its activities.
Fig.1 The flow chart of the proposed hybrid In order to compute, accurately, this features,
method. first, we have to segment these HRV signals into
stationary segments, since any signal is thought
of as being composed of small stationary
segments. To identify accurately the latter, many
3 Results and Discussion
algorithms, such as those based on Student’s t
The aim of a heart rhythm representation is to
provide a signal that should accurately reflect statistic [29], have been proposed, but they present
variations in the heart rhythm behavior, and which some difficulties in determining, accurately, the
should lend itself to much kind of processing. An stationary segments. S. Camargo et al [18] have,
example of such HRV variation in time is depicted recently, introduced an algorithm named K-S
in Fig.2. The heart rhythm is, usually, represented segmentation algorithm to determine more
in terms of rate which is given by the inverse of the accurately the stationary segments for a given
RR inter beat intervals. The variation of this inverse signal using an approach based on
in time, represents, thus, the HRV signal. Kolmogorov-Smirnov (KS) statistic. It should be
The goal of our method, to study the ANS, is to noted that the KS-segmentation algorithm which is
use features from HRV signals to detect stress of described in (Fig.3) is different from the KS
automobile driver’s, HRV analysis is commonly statistic. More details about this segmentation
used as a quantitative marker depicting the activity algorithm can be found in S Camargo et al work
of ANS that may be related to mental stress [32]. [18].

E-ISSN: 2224-2902 11 Volume 12, 2015

 Ahmed Bouziane, Benabdellah Yagoubi
WSEAS TRANSACTIONS on BIOLOGY and BIOMEDICINE Luis Vergara, Addisson Salazar

Start

. Initial value
𝑥𝑖 (i=1…N): time series
𝛼0 : Significance level
𝑙0 : Minimal size

𝑆1 =1 the two sides index

𝑆2 =N of the initial segment

Fig.4 Part from temporal HRV signal for

driver06, the stationary segments obtained using
For p=S1 S2-1 compute: the KS-segmentation algorithm are delimited by the
*D (p): KS statistic between the two
fragments 𝑥𝐿 [𝑥𝑠1,.. 𝑥𝑝 ] and
vertical lines.
𝑥𝑅 [𝑥𝑝+1,.. 𝑥𝑠2 ]
*𝛼 (p): statistical significance -A minimal stationary segment length or
number of points 𝐿0 = 720 (3mn), to distinguish
between the three main spectral components [1],
Find the index n of the ( as a result, the lowest frequency that can be
Maximum distance 𝐷𝑀𝑎𝑥 (n) resolved is 1/180≈0.0055Hz, just above the lower
limit of the VLF region).
-A statistical significance level 𝑃0 = 0.99
Check these conditions:
considered as an acceptable and standard level
1) 𝛼(n)>= 𝛼0
for this kind of distribution test [18], is illustrated
2) n-𝑆1 +1, 𝑆2 -n>=𝑙0
in (Fig.4) for the Driver 06 as an example. The
vertical discontinuous lines shown in this figure
delimit these stationary segments obtained by the
KS-segmentation algorithm.
NO YES Once these stationary patches have been
obtained, all features will be computed using
* Acceptance of the cut Kubios HRV Analysis Software, for each stationary
* The segment which is segment i of the HRV signals.
Check
S2=N on the left of the pointer will For thre feature selection ReliefF algorithm is
be analyzed context sensitive, robust, and can deal with datasets
* S2 will be changed with highly interdependent features, with
incomplete and noisy data and can be used for
evaluating the feature quality in multi-class
problems. Instead of n nearest hits and misses,
NO ReliefF searches for n nearest instances from each
class. The contributions of different classes are
*The next segment not analyzed YES weighted with their prior probabilities. For a more
on the right of S2 will be thorough overview of feature quality measures, see
analyzed [31]. The description of the Algorithm is provided
*S2 & S1 will be changed END in the Fig.5. Initial values are:
-M learning instances xk (N features and C
Fig.3 The KS-Segmentation algorithm
classes);
-Probabilities of classes py; Sampling parameter
To obtain a more accurate segmentation of the
m;
HRV temporal signals into stationary patches,
-Number n of nearest instances from each class;
we have chosen to work with:

E-ISSN: 2224-2902 12 Volume 12, 2015

 Ahmed Bouziane, Benabdellah Yagoubi
WSEAS TRANSACTIONS on BIOLOGY and BIOMEDICINE Luis Vergara, Addisson Salazar

periods according to the Table.1, with respect to

different stress class values drivers, each period has
at least one stationary segments.
These 33 features were used to create a single
vector representing each of the segments used in
the feature analysis. A total of 83 segments were
extracted from the 7 drivers: 28 from rest periods,
27 from highway driving, and 28 from city driving.
The resulting 83 feature vectors were then used
with target values taken from Table.1, for the
Fig.8 ReliefF Algorithm. selection of the best relevant features.
We have calculated the weights of 33 features
For output value, each feature Fi a quality weight for all 7 drivers, and to get a good idea we have
− 1 ≤ W [i] ≤ 1; calculated the sum of the weights for all features.
Quality estimations W can also be negative, To better observe the values we have multiplied all
however, W [Fi] ≤ 0 means that feature Fi is the values by 100, so this is a percentage
irrelevant. (-100 ≤ W [i] ≤ 100). All results are shown in
We have divided the whole driving time into 7 Table.3, Table.4, and Table.5.

Table.3 The percentage of feature weights in Time Domain

Features Driver6 Driver7 Driver8 Driver10 Driver11 Driver12 Driver15 Total
Mean RR
5,4265141 0,94608789 2,15877276 -0,3673764 7,08424454 2,06492077 3,82870738 21,141871
STD RR (SDNN)
-1,12031473 5,30112026 3,64183896 12,4138418 1,41023324 1,36558239 6,36862085 29,3809227
Mean HR
5,04576079 1,48723939 2,18555675 -0,1654806 7,43518712 5,80068335 4,02173856 25,8106854
STD HR
0,64459831 4,03596799 3,00599826 5,48175498 0,99390922 2,48964014 7,08112023 23,7329891
RMSSD
4,43330983 5,27276105 4,63776719 3,60629526 2,4912859 1,3390948 -0,6543828 21,1261312
NN50
4,93499129 3,75716757 5,89490642 4,09302749 2,42414003 -1,1514342 -1,1191262 18,8336723
pNN50 - -
4,1131347 5,2578275 5,70706436 2,49139871 7,06734887 2,09602672 1,00740149 18,5653515
HRV triangular
index
3,98374543 8,66282 1,17147445 3,79857956 2,9558076 2,54509804 4,03950875 27,1570338
TINN
-0,50005994 3,94967263 2,98269825 7,31407287 3,10674207 3,03159291 5,41442861 25,2991474

Table.4 The percentage of feature weights in Frequency Domain

Features Driver6 Driver7 Driver8 Driver10 Driver11 Driver12 Driver15 Total
Peak VLF -1,870995 -0,240218 -0,501896 -4,6579505 -1,4490866 5,25749220 -1,97545442 -5,438110603
Peak LF -1,455172 3,2278905 -0,101692 -6,6279138 -0,8270396 6,11780911 0,33046205 0,664342529
Peak HF 9,7373692 2,8587799 1,250429 27,686010 2,0393704 1,55724029 -2,43878988 42,69041027
Absolute power VLF -0,043268 1,8839376 3,3143432 -2,4180459 2,6976153 -1,4494387 7,43173916 11,41688249
Absolute power LF 0,6045780 4,7026317 2,6913606 12,1786356 1,6473066 -2,2639601 0,15869727 19,71924984
Absolute power HF 5,9716729 2,9236121 6,7313311 5,95757061 6,68760858 -1,2680803 1,55811979 28,56183483
Relative power VLF 4,0004530 3,4704776 5,1832699 -5,7655098 3,8176044 -2,3362531 3,41283561 11,78287771
Relative power LF 1,3066848 4,5676953 0,8556610 3,5804381 -0,0989818 -0,3357334 0,92373378 10,79949792
Relative power HF 4,3671313 -3,404726 6,4272542 7,72285027 5,6769006 5,29946549 6,28937362 32,37824898
Normalized power LF 1,6750151 2,7055632 3,7379327 9,11338152 4,34139307 14,0069736 3,75061345 39,33087282
Normalized power HF 2,1862527 1,8313037 5,2671011 11,3037182 5,53590517 7,75041689 7,03267480 40,90737281
LF/HF 2,1862527 1,8313037 5,2745638 11,3037182 5,53590517 7,75041689 7,03267480 40,91483546

E-ISSN: 2224-2902 13 Volume 12, 2015

 Ahmed Bouziane, Benabdellah Yagoubi
WSEAS TRANSACTIONS on BIOLOGY and BIOMEDICINE Luis Vergara, Addisson Salazar

Table.5 The percentage of feature weights in Nonlinear measures

Features Driver6 Driver7 Driver8 Driver10 Driver11 Driver12 Driver15 Total
SD1 4,481071723 5,265676992 4,625379042 3,590119996 5,908918707 1,270657762 -0,63200023 24,50982399
SD2 -0,89913093 5,155896285 3,949342831 12,30070408 1,299309636 1,740719033 6,916938265 30,4637792
RPA 1,130658558 1,487489835 1,559821106 -5,17918950 1,18132893 5,403955316 0,010057975 5,594122218
Lmean
RPA 1,897946925 1,75672714 4,240080664 -2,51557777 2,330251115 7,531546255 2,820652947 18,06162727
Lmax
RPA REC 5,285340028 2,39446695 2,990843723 -4,00252028 2,33608176 3,493944552 4,860862169 17,3590189
RPA DET 8,336087485 2,357117419 1,297534119 2,076421438 3,749714864 2,896763256 6,060281398 26,77391998
RPA 4,45967594 1,406204097 0,885551438 -2,86795753 2,450600875 4,312108289 2,26727989 12,913463
ShanEn
ApEn 2,65257873 4,343730032 4,2239E-16 -7,35371131 1,078834562 8,666119087 1,96788185 11,35543294
SampEn 3,302008024 4,215304831 0,489753063 4,950714233 2,078023777 4,666590115 -1,21892079 18,48347325
DFA α1 2,596206618 1,329679378 4,468054587 10,33841569 6,569025917 0,610299735 7,625394253 33,53707618
DFA α2 4,827956397 6,17605819 3,694075334 -7,99819723 2,255482084 -2,11124932 5,513654621 12,35778007
D2 6,301947561 -0,91726545 0,283827833 -6,39944059 -1,81097229 6,043045703 0,313220789 3,814363548

Comparing between all the results In these three Acknowledgements: The authors are very
tables we can see that there are some features are grateful to the European Commission presented by
relevant and we can easily see the difference, in the the EMMAG program, for the support of this work.
time domain features Standard deviation of RR
intervals present the most relevant feature for the References:
distinction between the three stress stages, [1] Task Force of the European Society of
following by The integral of the RR interval Cardiology and the North American Society of
histogram divided by the height of the histogram Pacing and Electrophysiology, Heart Rate
(HRV triangular index). Variability: Standards of Measurement,
In the Frequency domain there are 4 relevant Physiological Interpretation, and Clinical Use,
features can be used in stress detection, they are: Circulation, Vol.93, No.5, 1996, pp.1043–
1065.
Ratio between LF and HF band powers, Powers of
[2] Kohara K, Nishida W, Maguchi M, Hiwada K,
LF and HF bands in normalized units and HF band
Autonomic Nervous Function in Non-Dipper
peak frequencies.
Essential Hypertensive Subjects: Evaluation by
In Nonlinear measures it is obvious that Power Spectral Analysis of Heart Rate
Detrended fluctuation analysis with Short term Variability, Hypertension, Vol.26, No.5, 1995,
slope present the best choice to classify stress in pp.808–814.
similar situations. [3] Winchell RJ, Hoyt DB, Spectral Analysis of
Heart Rate Variability in the Icu: A Measure of
Autonomic Function, Journal of Surgical
4 Conclusion Research, Vol.63, No.1, 1996, pp.11–16.
The suggested hybrid method based on the KS- [4] Raquel Bailón, Luca Mainardi, Michele Orini,
segmentation algorithm, to determine the stationary Leif Sörnmo, Pablo Laguna, Analysis of Heart
segments, as well as the application of the ReliefF Rate Variability During Exercise Stress Testing
algorithm for the feature selection to study the Using Respiratory Information, Biomedical
HRV evolution in time, seems to be simple, Signal Processing and Control, Vol.5, No.4,
more accurate and easy to implement. 2010, pp.299–310.
According to our results using the proposed hybrid [5] M. Malik and A. J. Camm, Heart Rate
method, it is possible to confine the correlation Variability, Futura Publishing, 1995
between the LF/HF ratio and sympathovagal [6] Malliani, A., M. Pagani, F. Lombardi, and S.
balance during stress period. Also we can propose Cerutti, Cardiovascular Neural Regulation
in feature work study in more details relation Explored in the Frequency Domain,
between stress and some nonlinear measures as Circulation, Vol.84, No.2, 1991, pp.482–492.
for example Detrended fluctuation with Short term [7] N Montano, T G Ruscone, A Porta, F
slope for the good results obtained. This study may Lombardi, M Pagani and A Malliani, Power
help in linking the ANS behavior to the Spectrum Analysis of Heart Rate Variability to
corresponding stress situation. Assess the Changes in Sympathovagal Balance
During Graded Orthostatic Tilt, Circulation,
Vol.90, No.4, 1994, pp.1826–1831.

E-ISSN: 2224-2902 14 Volume 12, 2015

 Ahmed Bouziane, Benabdellah Yagoubi
WSEAS TRANSACTIONS on BIOLOGY and BIOMEDICINE Luis Vergara, Addisson Salazar

[8] M. Bootsma, C. A. Swenne, H. H. Van [22] A.Bouziane, B.Yagoubi, L.Vergara,

Bolhuis, P. C. Chang, V. M. Cats, and A. V. A.Salazar, The ANS Sympathovagal Balance
Bruschke, Heart Rate and Heart Rate Using a Hybrid Method Based on the Wavelet
Variability as Indexes of Sympathovagal Packet and the KS-Segmentation Algorithm,
Balance, Am J Physiol Heart Circ Physiol, Advances in Circuits, Systems, Signal
Vol.266, No.4, 1994, pp. H1565– H1571. Processing and Telecommunications, 2015,
[9] Kay SM, Marple SL, Spectrum Analysis: A pp.75-83.
Modern Perspective, Proc IEEE, Vol.69, [23] S. SEYEDTABAII, Analysis of Heart Rate
No.11, 1981, pp.1380–1419. Variation Filtering Using LMS Based Adaptive
[10] Juan Sztajzel, Heart Rate Variability: A Systems, WSEAS TRANSACTIONS on SIGNAL
Noninvasive Electrocardiographic Method to PROCESSING, Vol.04, No.04, 2008, pp.241–
Measure the Autonomic Nervous System, 250.
Swiss Medical Weekly, Vol.134, No.35, 2004, [24] A.Bouziane, B.YagoubiAnd,
pp.514–522. S.Benkraouda, Method for Estimating the
[11] U. Wklund, M. Akay and U. Niklasson, Duration of the ANS Sympathetic and
Short-Term Analysis of Heart-Rate Variability Parasympathetic Activities, Advances in
by Adapted Wavelet Transforms, IEEE Eng. Electrical and Computer Engineering, 2015,
Med. Biol, Vol.16, No.5, 1997, pp.113–138. pp.178-182.
[12] Y. Meyer, Les Ondelettes, Algorithmes, et [25] http://www.physionet.org/physiobank/data
Applications, Armand Colin, 1992. base/drivedb/
[13] I. Daubechies, Ten Lectures on Wavelets, [26] Healey JA, Picard RW, Detecting stress
SIAM, 1992. during real-world driving tasks using
[14] G. Strang and T. Nguyen, Wavelets and physiological sensors, IEEE Transactions
Filter Banks, Wellesley-Cambridge Press, in Intelligent Transportation Systems,
1996. Vol.06, No.2, 2005, pp.156–166.
[15] R. Kronland-Martinet, J. Morlet and A. [27] Deng Y, Wu Z, Chu CH, Zhang Q, Hsu
Grossmann, Analysis of Sound Patterns
DF, Sensor feature selection and
Through Wavlet Transforms, Int.J Pat Recogn.
Artif. Intel, Vol.1, No.2, 1987, pp.97–126. combination for stress identification using
[16] S. Elsenbruch, Z. Wang, W. C. Orr, and D. combinatorial fusion, Int J Adv Rob Syst,
Z. Chen, Time-Frequency Analysis of Heart Vol.10, No.3, 2013, pp.1–10.
Rate Variability Using Short-Time Fourier [28] Ahambi, J. S., Tandon, S. N., Bhatt, R. K.
Transform, Physiol Meas, Vol.21, No.2, 2000, P, Using wavelet transforms for ECG
pp.229–240. characterization. An on-line digital signal
[17] M. Akay, Introduction:Wavelet processing system, Engineering in Medicine
Transforms in Biomedical Engineering, Annals and Biology Magazine, IEEE, Vol.16, No.1,
of Biomedical Engineering, Vol.23, No.5, 1997, pp.77–83.
1995, pp.529–530. [29] Kaufmann, T., Sütterlin, S., Schulz, S.M.,
[18] S. Camargo, S. M. Duarte Quieros and C. Vögele, ARTiiFACT: a tool for heart rate
Anteneodo, Nonparametric Segmentation of artifact processing and heart rate variability
Nonstationary Time Series, Physical Review E, analysis, Behavior Research Methods, Vol.43,
Vol.84, 2011, 046702. No.4, 2011, pp.1161–1170.
[19] Bernaola-Galvan, P. Ch. Ivanov, L. A. N. [30] Niskanen, J.-P., Tarvainen, M.P., Ranta-
Amaral, and H. E. Stanley, Scale Invariance in Aho, P.O., Karjalainen, P.A., Software for
the Nonstationarity of Human Heart Rate, advanced HRV analysis, Computer Methods
Phys. Rev. Lett, Vol.87, 2001, 168105. and Programs in Biomedicine, Vol.76, No.1,
[20] Alberto Malliani, The Pattern of 2004, pp.73–81.
Sympathovagal Balance Explored in the [31] Huan Liu and Hiroshi Motoda,
Frequency Domain, Physiology, Vol.14, No.3, Computational Methods of Feature Selection,
1999, pp.111–117. Chapman and Hall/CRC Press, 2008.
[21] Bernaola-Galvan, P. Ch. Ivanov, L. A. N. [32] L. Salahuddin and D. Kim, Detection of
Amaral, and H. E. Stanley, Scale Invariance in Acute Stress by Heart Rate Variability Using a
the Non-stationarity of Human Heart Rate, Prototype Mobile ECG Sensor, International
Phys. Rev. Lett, Vol.87, 2001, 168105. Conference on Hybrid Information Technology,
Vol.2, 2006, pp.453–459.

E-ISSN: 2224-2902 15 Volume 12, 2015