Application of the Discriminant Analysis for Diagnostics of the
Arterial Hypertension
Analysis of Short-Term Heart Rate Variability Signals
Vladimir Kublanov, Anton Dolganov and Vasilii Borisov
Research and Development Medical and Biological Engineering Center of High Technologies, Ural Federal University,
Mira 19, 620002, Yekaterinburg, Russian Federation
Keywords: Heart Rate Variability, Discriminant Analysis, Classification, Multifractal Analysis, Arterial Hypertension.
Abstract: The investigation of the diagnostic possibilities for the arterial hypertension is presented. The 41 features of
the statistical, geometric, spectral and nonlinear methods during functional loads were considered for two
groups: healthy volunteers and patients suffering from the arterial hypertension of the II-III degree.
Application of the linear and quadratic discriminant analysis showed particular features that have high
classification efficiency.
1 INTRODUCTION
Nowadays, vascular disorders of the brain are in the
spotlight due to alarming epidemiological state of the
insult morbidity in the Russian Federation, disastrous
consequences of different cerebrovascular
pathologies for physical and mental health of the
nation.
Mortality from the cerebrovascular diseases in the
Russian Federation is among the highest in the world
and has an increasing trend. The number of the insults
in the Russian Federation is higher than 400000. Only
10% of the insult cases appear to be relatively mild
and can be cured during the first weeks of the
diseases. In other cases, sick people who survive,
retain, at some degree, pronounced neurological
defect that often leads to sustained disability and loss
of the ability to work. Up to 15 % of the sick people
after the insult are chained to bed to the end of their
lives.
Moreover, in the Russian Federation there are no
less than 1,5 mln people suffering from the chronic
cerebrovascular diseases with vascular dementia as
the end result. Among the variety of the
cerebrovascular diseases, there is significant share of
the chronic forms of the vascular disorders of the
brain, like hypertonic and atherosclerotic
encephalopathy.
A number of factors contribute for development
of the chronic forms of the brain blood flow disorders:
high prevalence of the arterial hypertension, improper
treatment of it for people with diagnosed disease,
significant prevalence of the cerebrovascular risk
factors (smoking, stress, excessive consumption of
alcohol, dyslipidemia), high frequency of the acute
forms of the brain flow disorders (Suslina and
Varakin, 2015).
Therefore, problem of the cerebrovascular disease
in the Russian Federation can be without a doubt
labeled as extreme. Specialists of the different
disciplines should unite their efforts to solve it.
Aftermath of the arterial hypertension mainly
manifest in the cardiovascular, cerebrovascular and
renovascular systems. Complications after the arterial
hypertension include common cerebrovascular
disorders: insult, transient ischemic attacks,
dementia, hypertonic encephalopathy (Wilkinson and
Waring, 2005).
Study of the heart rate variability (HRV) signals
(or R-R intervals) is one of the popular methods for
functional diagnostics of the human. Accumulated
experience of the HRV signals analysis by means of
the most common methods was generalized in
methodological recommendations of the Europeen
craniological and North-American electrophysical
societies(Malik, 1996), and in works of the Russian
experts (Baevskiy, 2001). These recommendations
are meant for the short-term records of the heart rate.
The results of these recommendations can be applied
to improve diagnostic efficiency. Such data can be
Kublanov, V., Dolganov, A. and Borisov, V.
Application of the Discriminant Analysis for Diagnostics of the Arterial Hypertension - Analysis of Short-Term Heart Rate Variability Signals.
DOI: 10.5220/0006044000450052
In Proceedings of the 4th International Congress on Neurotechnology, Electronics and Informatics (NEUROTECHNIX 2016), pages 45-52
ISBN: 978-989-758-204-2
Copyright
c
2016 by SCITEPRESS – Science and Technology Publications, Lda. All rights reserved
45
recorded not only in stationary medical institutions,
but also in case of outpatient and remote monitoring.
Various factors, like neurohumoral mechanisms
of the higher autonomic centers, influence on the
HRV signals. These factors cause nonlinear nature of
the heart rate changes. Methods of the non-linear
dynamics are promising tools to describe internal
structure of the R-R intervals time series.
In general, R-R intervals time series manifest
features of the determined chaos. This makes
methods applied for theory of chaotic systems
suitable for HRV analysis (Lin and Sharif, 2012).
Among the variety of non-linear dynamics
approaches commonly applied for the biomedical
signal processing the most perspective are methods
for estimation of the scale invariance, like fractal and
multifractal analysis (Lewis et al., 2012).
Thus, application of the multifractal analysis for
evaluation of the HRV signals allows one to obtain
new knowledge about patient state and effectiveness
of the treatment process. For the diagnostic means
multifractal approach can be considered both,
separately and in combination with more common
(traditional) methods of the data analysis.
In works of many authors the Discriminant
analysis is used as the classifying method: in tasks of
automatic sleep staging (Ebrahimi et al., 2013),
mental load estimation (Cinaz et al., 2013),
arrhythmia detection (Sivanantham and Shenbaga
Devi, 2014), real-life stress detection (Melillo et al.,
2011) and for automatic assessment of heart failure
severity (Melillo et al., 2014).
To our knowledge, our study is the first to attempt
to apply Discriminant analysis for evaluation of the
various HRV features descriptiveness for diagnostics
of arterial hypertension. The goal of this works is to
analyze possibilities and informativeness of the
different classes of methods for evaluation of the
short-term heart rate variability signals in task of the
arterial hypertension diagnostic during functional
loads.
2 MATERIALS AND METHODS
2.1 Recorded Data
The study was conducted on two groups: 21 relatively
healthy volunteers and 60 patients suffering from the
arterial hypertension of the II–III degree before
treatment. The clinical records were performed in
Sverdlovsk Clinical Hospital of Mental Diseases for
Military Veterans (Yekaterinburg, Russian
Federation). For the R-R interval signals registration
corresponding channel of the electroencephalograph-
analyzer “Encephalan-131-03” was used. The records
of the signals were obtained in two functional states:
functional peace (to be referred as state F) and passive
orthostatic load (to be referred as state O). Both states
were recorded for approximately 300 seconds. The
rotating table Lojer performed the spatial position
change of the patient during state O.
2.2 Heart Rate Variability Features
In this work, we investigated diagnostic possibilities
of the arterial hypertension by the different methods
of the short-term HRV signals analysis. Prior to the
processing the original time series were cleaned from
the artifacts. By the artifacts in this study, we
considered values of the R-R intervals that differed
from the mean by more than three values of standard
deviation. NN is the abbreviation for the “normal to
normal” time series, i.e. without artifacts.
For spectral and multifractal analyses NN time
series were interpolated using cubic spline
interpolation with the 10 Hz sampling frequency. All
calculations were computed using the in-house
software in the Matlab.
2.2.1 Statistical Features
Statistical methods are used for the direct quantitative
evaluation of the HRV time series. Main quantitative
features are:
M, the mean value of the R-R intervals:
,
(1)
where N is the number of elements in the NN, NN
i
is
the i-th element in time series;
SDNN, the standard deviation of the R-R
intervals
;
(2)
RMSSD is the square root of the mean of the
squares of the differences between successive
elements in NN:
;
(3)
NN5O,the number of pairs of successive
elements in NN that differ by more than 50 ms;
CV, the coefficient of variation, defined as ratio
of standard deviation SDNN to the mean M,
expressed in percent:
.
(4)
NEUROTECHNIX 2016 - 4th International Congress on Neurotechnology, Electronics and Informatics
46
2.2.2 Geometric Features
Geometric methods analyze distribution of the R-R
intervals as a random numbers. The common features
of these methods are:
М
0
, the mode, the most frequent value in the R-
R interval. In case of the normal distribution is
close to the mean M;
АМ
0
, the amplitude of the mode, is a number of
the R-R intervals that correspond to the mode
value. AM
0
shows the stabilizing effect of the
heart rate management, mainly caused by the
sympathetic activity;
VR, the variation range, is the difference
between the lowest R-R interval and the highest
R-R interval in the time series. VR shows
variability of the R-R interval values and
reflects activity of the parasympathetic
department of the autonomic nervous system
(ANS).
The following indexes are derived from common
geometric features:
SI, the Stress Index that reflects centralization
degree of the heart rate and mostly characterize
the activity of the sympathetic department of
the ANS
;
(5)
IAB, the Index of the Autonomic Balance,
depends on the relation between activities of
the sympathetic and parasympathetic
department of the ANS:
;
(6)
ARI, the Autonomic Rhythm Index, which
shows parasympathetic shifts of the autonomic
balance: smaller values of the ARI correspond
to the shift of the autonomic balance to the
parasympathetic activity:
;
(7)
IARP, the Index of Adequate Regulation
Processes, that reflects accordance of the
autonomic function changes of the sinus node
as a reaction of the sympathetic regulatory
effects on the heart
.
(8)
2.2.3 Spectral Features
Spectral analysis is used to quantify periodic
processes in the heart rate by the means of the Fourier
transform (Fr). The main spectral components of the
HRV signal are High Frequency – HF (0.4 – 0.15 Hz),
Low Frequency – LF (0.15 – 0.04 Hz), Very Low
Frequency – VLF (0.04 – 0.003 Hz), and Ultra Low
Frequency – ULF (lower than 0.003 Hz) (Malik,
1996, Baevskiy, 2001). For 300 seconds short-term
time series ULF spectral component is not analyzed.
HF spectral component characterize activity of the
parasympathetic department of the ANS and activity
of the autonomic regulation loop. LF spectral
component mainly characterize activity of the
sympathetic vascular tone regulation center. VLF
spectral component is defined by the suprasegmental
regulation of the heart rate, as the amplitude of the
VLF waves is related to the psycho-emotional strain
and functional state of the cortex(Baevskiy, 2001).
The quantitative features of spectral analyzes are
Spectral power of the HF, LF, VLF
components
Total power of the spectrum – TP;
Normalized values of the spectral components
by the total power - HF
n
, LF
n
and VLF
n
;
The LF/HF ratio, also known as the autonomic
balance exponent;
IC, the Index of centralization
.
(9)
2.2.4 Wavelet Transform Features
For nonstationary time series one can also use the
wavelet transform (wt), that can simultaneously study
time-frequency patterns. It is possible to acquire same
spectral features by means of the wavelet transform:
Spectral power of the HF, LF, VLF
components
Normalized values of the spectral components
by the total power - HF
n
, LF
n
and VLF
n
;
The LF/HF ratio.
Moreover, one can study informational
characteristics of the wavelet transform by the
analyze of the
function, where
and
are time series of the LF and HF spectral
components acquired by means of the wavelet
transform.
As the features of
is possible to use
number of the dysfunctions N
d
, maximal value of the
dysfunction (LF/HF)
max
, and intensity of the
dysfunction (LF/HF)
int
. By the dysfunction, we
consider values of
that suppress decision
threshold . According to our previous studies =10
(Egorova et al., 2014). For wavelet transform
Application of the Discriminant Analysis for Diagnostics of the Arterial Hypertension - Analysis of Short-Term Heart Rate Variability
Signals
47
computation in this work, we used wavelet Coiflet of
the fifth order.
2.2.5 Nonlinear Features
As the nonlinear method we adopted the multifractal
detrended fluctuation analysis (MFDFA) (Stanley et
al., 1999). Algorithm and application features of the
MFDFA method to estimation of short-term TS are
described in details in (Ihlen, 2012).
The main steps of the method include:
the detrending procedure with second degree
polynomial on non-overlapping segments,
length of the segments corresponds to the
studied time scale boundaries;
determination of the fluctuation functions for q
in range q=[-5,5]
,
(10)
where NN
is the local trend in the segment , N
s
is
the number of segments, s is the scale;
estimation of the slope exponent H
x
in log-log
plot of the fluctuation function against scale s
for each q
;
(11)
calculation of the scaling exponent (q)
;
(12)
the Legandre transform application for the
probability distribution of the spectrum
estimation
D
(13)
Fig. 1 represents the main features of the
multifractal spectrum estimated by the MFDFA
method. Here, H
0
is the height of the spectrum,
represents the most probable fluctuations in the
investigated time scale boundary of the signal; H
2
is
the generalized Hurst exponent (also known as
correlation degree);
min
represents behavior of the
smallest fluctuations in the spectrum;
max
represents
behavior of the greatest fluctuations in the spectrum;
W =
max
-
min
, is the width of multifractal spectrum
that shows the variability of fluctuations in the
spectrum. Multifractal characteristics are quantitative
measures of the self-similarity and may characterize
functional changes in the regulatory processes of the
organism.
In this study, we investigated time scale
boundaries that correspond to the LF and VLF
Figure 1: The characteristic features of the multifractal
analysis.
frequency bands: (6-25) sec and (25-300) sec
respectively. Previously it was shown that
multifractal analysis of the HF component is not
informative because of the noising, that was noted in
our earlier works and by other authors (Makowiec et
al., 2012).
2.3 Classification
As the classification method, we adopted linear and
quadratic discriminant analysis (DA) (Krzanowski,
2000). Linear DA aims to find such linear
combination of the features that can be used for
adequate separation between two classes. In turn,
quadratic DA aims to find quadratic combination of
the features for separation. In case of the current
study, two classes are healthy volunteers and patients
with the arterial hypertension.
Evaluation of the classifiers efficiency was
computed with typical measures for binary
classification performance. Let us make following
abbreviations:
P, the number of patients with arterial
hypertension;
N, the number of healthy volunteers;
TP – True Positive, the number of correctly
classified patients with arterial hypertension;
TN – True Negative, the number of correctly
labelled healthy volunteers;
FP – False Positive, the number of people
incorrectly classified as patients with arterial
hypertension;
FN – False Negative, the number of people
incorrectly classified as healthy volunteers.
Then, in accordance with the abbreviations,
binary classification measures are:
Total classification accuracy (ACC)
;
(14)
NEUROTECHNIX 2016 - 4th International Congress on Neurotechnology, Electronics and Informatics
48
Sensitivity (SEN)
;
(15)
Specificity (SPE)
;
(16)
Positive Predictive Value (PPV)
;
(17)
Negative Predictive Value (NPV)
.
(18)
For the performance measures evaluation
estimation we adopted 3-fold cross-validation
scheme (Jain et al., 2000). This technique imply
developing 3 classifiers according to following steps:
division of the original dataset randomly into 3
subsamples (i.e. 20 patients for a group with
arterial hypertension and 7 volunteers for
healthy group);
successive exception of one subsample (testing
subset);
development of a classifier with the remaining
2 subsamples (training subset);
testing of classifier with the excluded
subsample;
computation of the binary classification
measures;
averaging of the performance measures over 3
classifiers.
Division of the original dataset into 3 subsamples
allowed to obtain person-independent testing.
3 RESULTS AND DISCUSSIONS
The classifier efficiency was tested for 41 features in
two-dimensional space “state F – state O”. Tables 1
and 2 presents mean values and standard deviations
of the tested features, for group of healthy volunteers
and for group of patients with the arterial
hypertension respectively.
Data in tables 1 and 2 shows that such statistical
and geometric features as M, SDNN, CV, RMSSSD,
M
0,
VR, has no significant difference between state F
and state O for group of healthy volunteers. On the
other hand, same features changes significantly for
group of patients with arterial hypertension.
Moreover, mean values of the autonomic balance
features (N
d
, LF/HF(wt), (LF/HF)
int
, LF/HF(Fr),
(LF/HF)
max
) are much greater compared to those of
Table 1: Mean values and standard deviation of tested
features for state F and state O for group of healthy
volunteers.
Feature
State F
State O
М
955±88
940±89
SDNN
88±40
82±28
CV
9,2±4,1
8,8±2,8
RMSSD
55±20
56±21
NN50
42±5
48±9
M
0
949±100
935±111
VR
208±61
214±64
AM
0
37±7
41±9
SI
127±71
129±65
IAB
240±132
235±108
ARI
6,2±2,3
5,9±2,0
IARP
40±9
45±12
HF(Fr)
576±322
625±408
LF(Fr)
766±685
779±658
VLF(Fr)
480±259
491±225
TP(Fr)
1821±1090
1895±1099
HFn(Fr)
35±13
35±11
LFn(Fr)
31±11
32±12
VLFn(Fr)
34±16
33±13
LF/HF(Fr)
1,28±0,88
1,33±0,90
IC
3,44±2,25
3,21±2,04
HF(wt)
585±310
616±432
LF(wt)
835±707
731±636
VLF(wt)
542±294
451±191
HFn(wt)
38±13
38±11
LFn(wt)
28±10
29±10
VLFn(wt)
34±16
34±14
LF/HF(wt)
1,03±0,66
1,00±0,53
(LF/HF)
max
60±38
68±49
(LF/HF)
int
5553±6018
4859±5093
N
d
215±197
189±158
max
LF
0,65±0,29
0,65±0,24
min
LF
0,09±0,12
0,05±0,13
W LF
0,56±0,24
0,61±0,25
H
2
LF
0,19±0,10
0,17±0,11
H
0
LF
0,33±0,11
0,30±0,10
max
VLF
0,24±0,13
0,24±0,13
min
VLF
-0,04±0,11
-0,06±0,10
W VLF
0,28±0,19
0,30±0,19
H
2
VLF
0,03±0,08
0,01±0,06
H
0
VLF
0,12±0,07
0,13±0,09
healthy volunteers, especially in state O. This can be
interpreted as the shift of the autonomic balance to the
sympathetic department of the ANS in case of arterial
hypertension (Baevskiy, 2001).
Finally, one can note increase of the the
Multifractal exponents H
0
and H
2
for group of people
Application of the Discriminant Analysis for Diagnostics of the Arterial Hypertension - Analysis of Short-Term Heart Rate Variability
Signals
49
Table 2: Mean values and standard deviation of tested
features for state F and state O for group of patients with
arterial hypertension.
Feature
State F
State O
М
892±111
766±111
SDNN
49±27
54±24
CV
5,3±2,6
7,0±2,8
RMSSD
39±28
19±12
NN50
42±6
49±20
M
0
883±106
756±113
VR
163±80
160±61
AM
0
48±12
55±14
SI
285±160
415±318
IAB
461±244
545±358
ARI
10,3±4,5
12,3±6,9
IARP
57±17
79±30
HF(Fr)
375±467
143±171
LF(Fr)
474±528
455±438
VLF(Fr)
482±428
586±356
TP(Fr)
1331±1384
1184±899
HFn(Fr)
22±12
9±6
LFn(Fr)
33±8
31±10
VLFn(Fr)
44±14
60±11
LF/HF(Fr)
2,38±1,53
5,55±3,47
IC
1,67±1,02
0,79±0,41
HF(wt)
206±858
47±183
LF(wt)
238±1037
103±419
VLF(wt)
348±718
161±543
HFn(wt)
23±12
9±7
LFn(wt)
31±8
28±10
VLFn(wt)
46±15
63±12
LF/HF(wt)
2,19±1,54
5,66±3,67
(LF/HF)
max
132±73
251±134
(LF/HF)
int
14675±12107
26929±19043
N
d
469±321
628±383
max
LF
0,66±0,20
0,85±0,21
min
LF
0,21±0,15
0,37±0,37
W LF
0,47±0,22
0,56±0,26
H
2
LF
0,30±0,15
0,52±0,26
H
0
LF
0,40±0,16
0,64±0,23
max
VLF
0,25±0,08
0,60±0,27
min
VLF
0,04±0,05
0,08±0,09
W VLF
0,21±0,09
0,51±0,24
H
2
VLF
0,09±0,05
0,19±0,09
H
0
VLF
0,14±0,05
0,36±0,18
with arterial hypertension. It is worthy to mention that
for patients in state O for the time scale boundary that
correspond to the LF frequency band there is
qualitative change of the multifractal behavior: on
average estimates become persistent (Makowiec et
al., 2012).
Tables 3 and 4 presents binary classification
measures of the features, that has ACC higher than 75,
for linear and quadratic DA respectively.
Table 3: Efficiency of the classification for the linear
DA,%.
Feature
SEN
SPE
ACC
PPV
NPV
М
92
85
90
95
78
HF
n
(wt)
93
79
90
94
79
M
0
92
80
89
93
77
HF
n
(Fr)
93
74
89
92
78
VLF
n
(Fr)
92
74
87
92
75
HF(Fr)
97
55
86
87
87
VLF
n
(wt)
90
74
86
92
70
H
0
LF
88
75
85
92
70
H
2
VLF
90
69
85
90
77
RMSSD
95
55
85
87
81
H
2
LF
87
65
81
88
63
SDNN
95
39
81
83
75
H
0
VLF
92
45
80
84
62
LF/HF(Fr)
93
38
80
84
40
LF/HF(wt)
93
38
80
85
55
AM
0
90
50
80
84
66
max
VLF
93
34
79
81
75
IARP
88
50
79
85
69
IC
100
14
79
78
33
min
VLF
97
24
79
80
44
W VLF
100
14
79
78
67
HF(wt)
95
30
79
80
80
VR
93
31
77
80
67
(LF/HF)
int
92
33
77
84
19
N
d
87
50
77
85
53
CV
95
25
77
79
75
(LF/HF)
max
93
24
76
80
19
TP (Fr)
95
20
76
78
75
SI
98
5
75
76
33
ARI
95
14
75
77
20
VLF(wt)
97
10
75
76
44
LF(wt)
95
15
75
77
75
Table 3 shows that highest 3-fold cross-validation
estimate of the total classification accuracy was
achieved by М, HF
n
(wt), M
0
, HF
n
(Fr), VLF
n
(Fr), HF
n
(Fr), VLF
n
(wt), H
0
LF, H
2
VLF and RMSSD.
However, only features М, HF
n
(wt), M
0
has high level
of Specificity and Sensitivity at the same time.
Overall, the highest classification efficiency is
achieved by the feature M (SEN = 92,% SPE = 85,%
ACC=90%).
NEUROTECHNIX 2016 - 4th International Congress on Neurotechnology, Electronics and Informatics
50
Table 4: Efficiency of the classification for the quadratic
DA, %.
Feature
SEN
SPE
ACC
PPV
NPV
М
92
90
91
96
79
HF
n
(wt)
93
79
90
94
79
RMSSD
93
69
87
90
76
M0
90
80
87
93
75
VLF
n
(wt)
92
74
87
92
73
N
d
87
85
86
95
69
HF
n
(Fr)
90
74
86
92
70
VLF
n
(Fr)
90
74
86
92
70
IC
93
60
85
88
72
H
2
LF
93
55
84
86
79
AM
0
87
75
84
92
65
LF/HF(wt)
82
90
84
96
62
(LF/HF)
int
82
90
84
96
62
H
0
LF
83
85
84
94
63
max
LF
85
79
84
93
66
H
2
LF
83
84
84
94
63
HF(wt)
95
45
82
84
83
HF(Fr)
90
55
81
86
64
max
VLF
95
35
80
81
75
LF/HF(Fr)
77
90
80
96
56
H
0
VLF
82
74
80
91
59
W VLF
97
19
77
79
27
min
VLF
78
74
77
91
55
SDNN
87
51
77
84
67
TP (Fr)
92
35
77
81
67
IARP
77
75
76
90
55
LF(wt)
97
15
76
77
44
TP
93
25
76
79
63
(LF/HF)
max
70
90
75
96
51
NN50
97
10
75
76
17
LF,%
100
0
75
75
0
VR
90
31
75
80
59
Data in table 4 shows that application of the
quadratic DA, in general, improves classification
efficiency. At the same time classification efficiency
of the features М, HF
n
(wt), M
0
does not change much
compared to the results of linear DA.
One can notice that for both linear and quadratic
DA, most features has low level of the Specificity. At
the same time application of the quadratic DA allows
to improve this index for the following features: N
d
,
LF/HF(wt), (LF/HF)
int
, LF/HF(Fr), (LF/HF)
max
, H
0
LF, H
2
LF. In this case total accuracy and sensitivity
do not change much.
Wavelet and Fourier transform spectral features
has almost comparable results, with wavelet
transform features having slightly higher
classification efficiency for most of features.
Figure 2 presents classification rule for the M
feature, which has the highest level of total accuracy.
There and below circles are healthy people, squares
are people with arterial hypertension; solid line is the
classification rule.
Figure 2: Classification rule of the linear DA for the feature
M.
Figure 3 presents classification rule for the feature
LF/HF(wt), which has of the highest level of
specificity.
Figure 3: Classification rule of the quadratic DA for the
feature LF/HF(wt).
4 CONCLUSIONS
In this article, the diagnostic possibilities of the
features of the statistical, geometric, spectral and
nonlinear methods were investigated as the indicators
of the arterial hypertension during functional studies.
The results of current study suggests particular
features that could be effective for diagnostics of the
arterial hypertension.
The highest estimates of the classification
efficiency was obtained for the following features:
Application of the Discriminant Analysis for Diagnostics of the Arterial Hypertension - Analysis of Short-Term Heart Rate Variability
Signals
51
mean value of the R-R intervals, mode of the R-R
intervals and normalized spectral power in HF
frequency band, for both linear and quadratic
discriminant analysis. However, most features have
low specificity rate. In addition, it was noted, that for
quadratic discriminant analysis features of wavelet
transform LF to HF ratio and multifractal exponents
in LF frequency band has highest rate of specificity,
while having relatively high rates of sensitivity and
accuracy.
Nevertheless, it is of great interest for further
research on a larger sample size to increase specificity
of the classification. One of the subject for our future
investigation, which is currently underway, is to
evaluate robustness of the classifier based on either
linear or quadratic combination of features set.
ACKNOWLEDGEMENTS
The work was supported by Act 211 Government of
the Russian Federation, contract № 02.A03.21.0006.
REFERENCES
Baevskiy, R. M., 2001. Аnaliz variabelnosti serdechnogo
ritma pri ispolzovanii razlichnykh ehlektrokardio-
graficheskikh sistem (metodicheskie rekomendatsii)
[Analysis of heart rate variability using different
electrocardiographic systems (guidelines)]. Vestnik
aritmologii [Herald Arhythmology], (24), 65-87.
Cinaz, B., Arnrich, B., La Marca, R., Tröster, G., 2013.
Monitoring of mental workload levels during an
everyday life office-work scenario. Personal and
Ubiquitous Computing, 17 (2), 229-239.
Ebrahimi, F., Setarehdan, S.-K., Ayala-Moyeda, J.,
Nazeran, H., 2013. Automatic sleep staging using
empirical mode decomposition, discrete wavelet
transform, time-domain, and nonlinear dynamics
features of heart rate variability signals. Computer
Methods and Programs in Biomedicine, 112 (1), 47-57.
Egorova, D. D., Kazakov, Y. E., Kublanov, V. S., 2014.
Principal Components Method for Heart Rate
Variability Analysis. Biomedical Engineering, 48 (1),
37-41.
Ihlen, E. A. F., 2012. Introduction to multifractal detrended
fluctuation analysis in Matlab. Frontiers in Physiology,
3 JUN.
Jain, A. K., Duin, R. P. W., Mao, J., 2000. Statistical pattern
recognition: A review. IEEE Transactions on Pattern
Analysis and Machine Intelligence, 22 (1), 4-37.
Krzanowski, W. J., 2000. Principles of multivariate
analysis: a user’s perspective. Rev edition. Oxford
Oxfordshire. New York: Oxford University Press.
Lewis, M. J., Short, A. L., Suckling, J., 2012. Multifractal
characterisation of electrocardiographic RR and QT
time-series before and after progressive exercise.
Computer Methods and Programs in Biomedicine, 108
(1), 176-185.
Lin, D. C., Sharif, A., 2012. Integrated central-autonomic
multifractal complexity in the heart rate variability of
healthy humans. Frontiers in Physiology, 2 FEB.
Makowiec, D., Rynkiewicz, A., Wdowczyk-Szulc, J.,
Zarczynska-Buchowiecka, M., 2012. On reading
multifractal spectra. multifractal age for healthy aging
humans by analysis of cardiac interbeat time intervals.
Acta Physica Polonica B, Proceedings Supplement, 5
(1), 159-170.
Malik, M., 1996. Heart rate variability: Standards of
measurement, physiological interpretation, and clinical
use. Circulation, 93 (5), 1043-1065.
Melillo, P., Bracale, M., Pecchia, L., 2011. Nonlinear Heart
Rate Variability features for real-life stress detection.
Case study: students under stress due to university
examination. BioMedical Engineering OnLine, 10 (96),
1-13.
Melillo, P., Pacifici, E., Orrico, A., Iadanza, E., and
Pecchia, L., 2014. Heart rate variability for automatic
assessment of congestive heart failure severity. IFMBE
Proceedings, 41, 1342-1345.
Sivanantham, A., Shenbaga Devi, S., 2014. Cardiac
arrhythmia detection using linear and non-linear
features of HRV signal. In: 2014 International
Conference on Advanced Communication Control and
Computing Technologies (ICACCCT). 795-799.
Stanley, H. E., Amaral, L. A. N., Goldberger, A. L., Havlin,
S., Ivanov, P. C., Peng, C.-K., 1999. Statistical physics
and physiology: Monofractal and multifractal
approaches. Physica A: Statistical Mechanics and its
Applications, 270 (1-2), 309-324.
Suslina, A., Varakin, Y. Y., 2015. Clinical guidelines for
the early diagnosis, treatment and prevention of
vascular diseases of the brain. [in Russian]. Moscow:
MEDpress-inform.
Wilkinson, J. B., Waring, S., 2005. Cockroft DR Arterial
hypertension. Answers to your questions. London:
Elsevier Syns.
NEUROTECHNIX 2016 - 4th International Congress on Neurotechnology, Electronics and Informatics
52
