Heart rate variability as a biomarker for epilepsy seizure prediction

Abstract

Objective:
Epilepsy is a neurological disorder that causes seizures of many different types. Recent research has shown that epileptic seizures can be predicted by using the electrocardiogrami instead of the electroencephalogram. In this study, we used the heart rate variability that is generated by the fluctuating balance of sympathetic and parasympathetic nervous systems to predict epileptic seizures.

Methods:
We studied 11 epilepsy patients to predict the seizure interval. With regar tos the fact that HRV signals are nonstationary, our analysis focused on linear features in the time and frequency domain of HRV signal such as RR Interval (RRI), mean heart rate (HR), high-frequency (HF) (0.15-0.40 Hz) and low-frequency (LF) (0.04-0.15 Hz), as well as LF/HF. Also, quantitative analyses of Poincaré plot features (SD1, SD2, and SD1/SD2 ratio) were performed. HRV signal was divided into intervals of 5 minutes. In each segment linear and nonlinear features were extracted and then the amount of each segment compared to the previous segment using a threshold. Finally, we evaluated the performance of our method using specificity and sensitivity.

Results:
During seizures, mean HR, LF/HF, and SD2/SD1 ratio significantly increased while RRI significantly decreased. Significant differences between two groups were identified for several HRV features. Therefore, these parameters can be used as a useful feature to discriminate a seizure from a non-seizure The seizure prediction algorithm proposed based on HRV achieved 88.3% sensitivity and 86.2 % specificity.

Conclusion:
These results indicate that the HRV signal contains valuable information and can be a predictor for epilepsy seizure. Although our results in comparison with EEG ares a little bit weaker, the recording of ECG is much easier and faster than EEG. Also, our finding showed the results of this study are considerably better than recent research based on ECG (Tab. 1, Fig. 10, Ref. 17).

Full text

Indexed and abstracted in Science Citation Index Expanded and in Journal Citation Reports/Science Edition
Bratisl Med J 2017; 118 (1)
3–8
DOI: 10.4149/BLL_2017_001
CLINICAL STUDY
Heart rate variability as a biomarker for epilepsy seizure prediction
Moridani MK, Farhadi H
Department of Biomedical Engineering, Tehran Medical Branch, Islamic Azad University,
Tehran, Iran. m.karimi@gmail.com
ABSTRACT
OBJECTIVE: Epilepsy is a neurological disorder that causes seizures of many different types. Recent research
has shown that epileptic seizures can be predicted by using the electrocardiogrami instead of the electroen-
cephalogram. In this study, we used the heart rate variability that is generated by the fl uctuating balance of
sympathetic and parasympathetic nervous systems to predict epileptic seizures.
METHODS: We studied 11 epilepsy patients to predict the seizure interval. With regar tos the fact that HRV
signals are nonstationary, our analysis focused on linear features in the time and frequency domain of HRV
signal such as RR Interval (RRI), mean heart rate (HR), high-frequency (HF) (0.15–0.40 Hz) and low-frequency
(LF) (0.04–0.15 Hz), as well as LF/HF. Also, quantitative analyses of Poincaré plot features (SD1, SD2, and
SD1/SD2 ratio) were performed. HRV signal was divided into intervals of 5 minutes. In each segment linear
and nonlinear features were extracted and then the amount of each segment compared to the previous seg-
ment using a threshold. Finally, we evaluated the performance of our method using specifi city and sensitivity.
RESULTS: During seizures, mean HR, LF/HF, and SD2/SD1 ratio signifi cantly increased while RRI signifi cantly
decreased. Signifi cant differences between two groups were identifi ed for several HRV features. Therefore, these
parameters can be used as a useful feature to discriminate a seizure from a non–seizure The seizure prediction
algorithm proposed based on HRV achieved 88.3% sensitivity and 86.2 % specifi city.
CONCLUSION: These results indicate that the HRV signal contains valuable information and can be a predictor
for epilepsy seizure. Although our results in comparison with EEG ares a little bit weaker, the recording of ECG
is much easier and faster than EEG. Also, our fi nding showed the results of this study are considerably better
than recent research based on ECG (Tab. 1, Fig. 10, Ref. 17). Text in PDF www.elis.sk.
KEY WORDS: epileptic seizure, heart rate variability, linear and non–linear analysis, prediction.
Department of Biomedical Engineering, Tehran Medical Branch, Islamic
Azad University, Tehran, Iran
Address for correspondence: M.K. Moridani, No. 29, Floor 4, Farjam
St., Tehran–Pars, Tehran, Iran.
Postal Code: 1653989618
Phone: +982177874289, Fax: +982188675452
Introduction
Epilepsy refers to chronic disorders and brain dysfunction in
which normal communication between nerve cells in the brain is
disturbed. Seizures are a sudden surge of electrical activity in the
brain. Seizures occur alternately at intervals in the brain. Indeed,
seizures depend on the location of epileptic discharges in the brain
as well as symptoms and type of epileptic seizures. There are dif-
ferent types of seizures, which are associated with various risk
factors for the patients depending on type and severity of epilepsy.
Partial epilepsy is caused by a damage to one side of the brain,
which is divided into two categories: simple partial epilepsy and
simple complex epilepsy. Pure epilepsy lasts less than a minute.
The symptoms depend on the location of epileptic discharges.
Complex epilepsy lasts one to two minutes. Generalized epilepsy
affects both sides of the brain and the affected individual suddenly
becomes unconscious. Sympathetic nervous system dysfunction
occur in epilepsy with febrile seizure. A part of the body shakes
uncontrollably in focal epilepsy, which may spread to other areas of
the body and the whole body may shake uncontrollably. However,
the affected individual does not lose consciousness completely.
An epileptic patient may be affected by two different kinds
of seizures such as unilateral and bilateral seizures depending on
the type of epilepsy. In single seizures, electrical discharge affects
some areas of the brain (cortex), but the affected individual does
not lose consciousness immediately. This type of seizure is called
focal sensory epilepsy. In the bilateral assault, both hemispheres
are simultaneously affected, and the affected individual loses con-
sciousness immediately.
An algorithm that can predict the occurrence of epileptic sei-
zures extremely ensures the patients and their caregivers of prog-
nosis of epileptic seizures to either prevent irreparable damage
to the patients or take necessary measures to reduce the severity
of damages.
Electroencephalogram (EEG) signals are a convenient and
well-known method for examining epileptic seizures. EEG refers
to either superfi cial (noninvasive) or invasive records of brain
activities. Needle electrodes are implanted in the scalp in the in-
vasive method. Given that the brain is the origin of epileptic sei-
zures recording EEG signal in epileptic patients is problematic and
not easily accessible. Epileptic seizures also affect the autonomic

Bratisl Med J 2017; 118 (1)
3–8
4
nervous system and consequently activities of both sympathetic
and parasympathetic nerves. Thus, the heart is one major organ
affected by epileptic seizures. In this paper, ECG signals were used
to predict epileptic seizures, which can be done more easily and
more quickly. It is expected that the results of ECG recordings do
not differ from the results of EEG signals. Hopefully, ECG signals
can be used as a complementary approach to prediction of epilep-
tic seizures. Finally, a combination of two EEG and ECG signals
can help to improve the results and effi ciency of the algorithm.
Many studies were performed to predict epileptic seizures.
However, EEG signals were used in most of these studies. Mar-
tinerie et al covered 19 seizures in 11 patients in 1998. Density
correlation method was utilized in the recent survey in which sen-
sitivity was reported as 89 % (1).
Le Van Quyen et al used EEG signals to predict 11 epileptic
seizures in 9 patients in 2000. In the above study, similarity index
algorithm was used with 94 % sensitivity (2). Navaro et al used
similarity index method to predict 41 epileptic seizures in 11 pa-
tients in 2002. They obtained sensitivity of 83 % (3). Netoff T et
al used power spectral density method to predict 45 seizures in 9
patients in 2009. They obtained sensitivity of 77.8 % (4).
Yun Park et al used power spectral density method for predic-
tion of 80 seizures in 18 patients in 2011. They obtained sensitivity
of 92.5 % (5). The results of the recent study showed that epileptic
patients are more prone to abnormal heart rhythms not only during
the seizure but also before the seizure (6).
In general, epileptic seizures are associated with an apparent in-
crease in heart rate in most cases. However, arrhythmia occurs, and
heart rate decreases in some cases such as partial epilepsy and gen-
eralized epileptic seizures (7–8). The article is organized as follows.
In the second section, used database and method to predict
seizures with ECG signal processing metrics and evaluation meth-
ods are discussed. In the third section, the results of these meth-
ods are discussed. Discussion and conclusion are presented in the
fourth section. In this section, a summary and comparison of the
proposed approach with previous studied are shown. Work limi-
tations are also discussed. Some recommendations are also given
for future work.
Materials and methods
Heart rate fl uctuates under the infl uence of sympathetic and
parasympathetic nervous systems. Short-term and long-term chang-
es in heart rate refl ect the autonomic nervous system function (9).
A typical ECG tracing of the cardiac cycle (heartbeat) consists
of a P wave (atrial depolarization), a QRS complex (ventricular
depolarization), and a T wave (ventricular repolarization and a U
wave. Each wave and the distances between them are related to
different parts of the heart and can be used to assess cardiac health,
but R wave is more signifi cant than other waves, which indicates
ventricular contraction or a heartbeat. R–R interval is also called
beat-to-beat or normal-to-normal (NN) interval, which represents
time interval between heart beats (Fig. 1).
A change in cardiac signal during two consecutive beats is
called heart rate variability (10). HRV signal shows different modes
of cardiovascular diseases. Therefore, analysis of heart rate vari-
ability can be used as a tool to monitor changes in the function
of the autonomic nervous system. However, less variable heart
beats show relatively low health. Naturally, heart rate variability
is directly related to individual health and healing. Increased heart
rate variability increases individual health (11).
Studies have shown that correct information on autonomic
nervous system function can be obtained using noninvasive HRV
signal analysis in all brain disorders resulting from sympathetic and
parasympathetic imbalance. HRV signal is a multivariate variable
of cardiovascular systems, which represents dynamic characteris-
tics, short-term and long-term correlations and complexities of the
cardiac signal and autonomous nervous system. Nowadays, there
are different methods for HRV signal processing, which can be
divided into linear and nonlinear methods.
Used database
The data on partial epileptic patients available in Physionet
Database is used in this study (12). Seven patients were examined
in this study. In total, 11 seizures were observed during hospital-
ization of the patients. ECG signals with 200-kHz frequency sam-
pling, 12-bits per sample and 5-mV resolution were digitalized and
recorded in epileptic patients. Seizure interval was specifi ed in all
patients and labeled in the database.
Extracting linear and nonlinear features of HRV signal
Changes are calculated with linear statistical methods, which
are divided into time-domain and frequency-domain methods. A
simple calculation is one main advantage of these features. How-
ever, statistical properties depend on the quality of recorded data.
This quality may be affected by environmental noises. In time-
domain method, R–R intervals show, analyze and examine high
frequency changes or short-term changes in which various features
can be extracted as follows: R–R mean interval, standard deviation
of NN intervals (SDNN), Root Mean Square of the Successive Dif-
ferences (RMSSD), the number of differences in consecutive R–R
intervals greater than 50ms (NN50), the ratio obtained by dividing
total number of NN50 intervals to R–R (pNN50). Analysis of heart
rate variability in adults showed that range of HRV signal consists
of three frequencies as follows: low frequency (LF) (0.04–0.1
Hz), high frequency (HF) (0.15–0.4 Hz) and very low frequency
(VLF) (0.0001– 0.04 Hz) (13). Fluctuations in these two compo-
nents represent sympathetic and parasympathetic activity. In the
Fig. 1. ECG signal and R–R intervals.

Moridani MK, Farhadi H. Heart rate variability as a biomarker for epilepsy seizure prediction
xx
5
frequency-domain method, the ratio of these two components is
used as a measure of the balance of sympathetic and parasympa-
thetic function. In the frequency-domain method, VLF, LF, HF,
LF / HF parameters can be extracted from HRV signal. In fact,
calculated energy is different in various frequency bands (14).
It should be noted that properties of complex systems are
ignored or deleted in linear methods, which cause an additional
error. Physiological systems are fundamentally and inherently
nonlinear. Since standard parameters in HRV only describe lin-
ear and periodic behaviors analysis, more complex and nonlinear
relationship cannot be detected. Recent advances in the theory of
nonlinear dynamics have facilitated signal analysis in nonlinear
living organisms.
Nowadays, nonlinear techniques can describe processes in
living biological organisms.
Poincaré return map is a relatively new technique for analyzing
nonlinear dynamics such as HRV signal. Each point in the graph
is specifi ed as n=1, 2, 3,…, k (RRn,RRn+1) where k represents sig-
nal strength (15). Statistically, this mapping graphically displays
the correlation between consecutive R–R intervals. The mapping
gives useful information on short-term and long-term fl uctuations.
The mapping is shown in Figure 2. SD1 and SD2 parameters are
shown in this diagram. SD1 shows beat-to-beat rapid changes,
which is more related to respiratory sinus arrhythmia. However,
SD2 describes long-term beat-to-beat changes. SD2
SD1 is calculated
to describe the relationship between these components (16). SD1
and SD2 values in Poincare mapping directly depend on statistical
values of standard deviation of heart rate signal and two consecu-
tive intervals of R peaks, which are calculated using equation (1)
where RRn+1 refers to the (n+1)th beat-to-beat interval, RRn repre-
sents to nth beat-to-beat interval and SD denotes standard deviation.
(1)
Prediction algorithm
After recording ECG signals in epileptic patients, R peaks were
detected in these signals using the method proposed by Pan and
Tompkins (17). Then, R–R intervals were calculated, and HRV
signals were organized by identifying the location of the peaks.
Figure 3 shows ECG signal in an epileptic patient recorded in
about 83 minutes and 20 seconds. The patient experienced epileptic
seizures from the 14th minute and 36th second to the 16th minute
and 12th second. In other words, seizures lasted from the 972nd
second to 876th second. Figures 4 and 5, respectively, show ECG
signal in an epileptic patient in a seizure interval and HRV signal
in periods before, during and after the seizure.
Since location and duration of the seizure were specifi ed in
available data, HRV signal was divided into time intervals using
Fig. 2. Poincaré return map of RR intervals for a healthy individual
(n = 100).
Fig. 3. ECG signal in an epileptic patient.
Fig. 4. Seizure interval in an epileptic patient.

Bratisl Med J 2017; 118 (1)
3–8
6
window function. Then, linear and nonlinear properties were ex-
tracted in each interval in the signal. Paired t-test was used to de-
tect those features that were signifi cantly different in this interval
compared to the previous interval. The detected intervals were
used to predict epileptic seizures. Figure 1 shows a diagram of
the proposed algorithm.
Selecting an optimal threshold
Threshold values with the best optimum sensitivity and the
least prediction error were chosen. In cases where specifi city
(predicting non–seizure interval) was acceptable for different
thresholds, the threshold with the lowest prediction error would
be selected. An alarm is given when the desired feature reaches
the threshold. Given that the threshold is different for each patient,
a single threshold is selected for each patient. For example, mean
and variance values of nth window is half the average values of
(n–1)th window. First, specifi city and initial sensitivity values were
calculated. Then, sensitivity values and the rate of incorrect pre-
dictions have been computed by changing the threshold. Finally,
the best threshold with the highest specifi city and maximum sen-
sitivity and the least prediction error were selected.
Introducing prediction criteria
In this paper, several features were used to predict epileptic
seizures in patients. The goal of all these features lied in achiev-
ing optimum results to obtain complete information on the future
status of patients for nurses and practitioners. Therefore, it is es-
sential to defi ne some criteria in this area.
An important criterion is the false positive rate or false-alarm
rate, which shows how much the right time of the epileptic seizure
was predicted in the proposed system or algorithm. In other words,
lower criteria represent the higher effi ciency of the algorithm in the
prediction. The second criteria are prediction of the horizon, which
shows the time of alarm to nurses and practitioners and specifi es
how long before the seizure the alarm is given. The greater the
prediction horizon, the greater the effi ciency of the system, so
Fig. 5. HRV signal in epileptic patients before, during and after the
seizure.
Fig. 6. An algorithm to determine a threshold to the prediction of the
future condition of the patient.
Fig. 7. Comparison of R–R Interval and mean heart rate in Poincare
plot in the seizure and non-seizure interval.
Fig. 8. Comparison between low-frequency and high-frequency in
Poincare plot in the seizure and non-seizure interval.

Moridani MK, Farhadi H. Heart rate variability as a biomarker for epilepsy seizure prediction
xx
7
that nurses and practitioners would have enough time to provide
more facilities and care measures for the patients (Figs 6 – 10).
Statistical test
In general, statistical tests aim to determine criteria for the
feature or features extracted from the signals recorded from the
patients. This paper aimed to examine the features extracted from
HRV signal at different time intervals. Thereby, paired sample t-
test was used to determine changes and distinction in extracted
features at various time intervals. Paired t-test also called before
and after test aimed to compare means in the two groups. The as-
sumption of normality of variables in the two groups should be
observed before the test. There are two observations (observa-
tions close to seizure and intervals far away from the seizure) for
each in this test.
In the output of the test, mean and standard deviation of the
variables are shown to describe the data. Then, the results of cor-
relation between values close to seizure and far away from the
seizure (this part does not affect the interpretation of mean com-
Fig. 9. Comparison between SD1 and SD2 in Poincare plot in the sei-
zure and non-seizure interval.
parison results) are given. Then, the results of mean equality in
the two groups by the mean difference in the two intervals with
zero value are presented. If signifi cance values were less than α,
the assumption of equality of means in the two groups would be
rejected at α error level. In this study, the signifi cance level was
considered less 0.5.
Evaluation of the performance of the proposed algorithm
To evaluate the performance of the proposed algorithm, we
introduced two criteria, namely, sensitivity and specifi city. The
sensitivity (Sn) of seizure detection is the probability that the de-
tection is positive when the HRV segments are with the seizure.
The specifi city (Sp) is defi ned as the probability that the sei-
zure detection result says a non–seizure segment, when in fact,
they are seizure free. The sensitivity and specifi city measures are
given in equation (2). True positive (TP): Sick people correctly
identifi ed as sick, False positive (FP): Healthy people incorrectly
identifi ed as sick, True negative (TN): Healthy people correctly
identifi ed as healthy and False negative (FN): Sick people incor-
rectly identifi ed as healthy.
(2)
Results
In order to observe c hanges in cardiac signal of epileptic pa-
tients, linear and nonlinear values mentioned in the previous sec-
tion at following intervals were calculated: ten to fi ve minutes
before the seizure , from fi fteen to ten minutes prior to seizure
, from twenty to fi fteen minutes before the seizure and a distant
interval from the seizure range (two hours before the seizure ).
Table 1 shows a qualitative (statistical) comparison of linear
and nonlinear features using paired t-test and calculated p-value
at two-time intervals ranging from ten to fi ve minutes before the
seizure and two hours before the seizure. It can be observed that
RRI dropped in moments close to the seizure, but Mean HR,
and SD2
SD1 and signifi cantly increased in most intervals compared
to the two hours before the seizure. Since p-values for RRI, Mean
HR, and SD2
SD1 features were less than 0.01, it can be concluded
that these features have acceptable prediction values and can be
regarded as appropriate criteria for predicting the future condi-
tion of the patients.
Also, examining behavioral features extracted at different time
intervals refl ects changes in the interval of about 30 minutes before
the seizure but most changes occurred in the interval of 15 min-
utes before the seizure. Therefore, the intervals of 10–15 minutes
and 5–10 minutes before the seizure were more emphasized than
other intervals and HRV signal features were more studied in the
previous intervals than the last intervals (Tab. 1).
Two ratios of SD2
SD1 and SD2
SD1 showed identical behaviors, which
confi rmed a signifi cant correlation between these two features since
these two indicate parasympathetic and sympathetic activities.
Any increase or decrease in these two features is due to different
Fig. 10. Comparison between low-frequency to high-frequency ratio
and SD2 to SD1 ratio in Poincare plot in the seizure and non-seizure
interval.

Bratisl Med J 2017; 118 (1)
3–8
8
extraction method. Thus, it can be concluded that HRV signal be-
havior before 20–5 minutes interval prior to seizure is suitable for
extraction of the features with regard to comparable intervals in
which maximum changes were observed in the extracted features.
To determine the threshold for the seizure predicting measure-
ment system, the best feature to investigate HRV signal behavior
at various time intervals with signifi cant distinction were selected.
These features have been chosen using paired t-test with a p-value
less than 0.01.
The threshold values with the best sensitivity and the least
prediction error have been selected. In cases where acceptable
specifi city for different thresholds (predicting noninvasive inter-
val) were obtained, the threshold with the least prediction error has
been determined. An alarm is given when the threshold reaches the
desired feature. Given that the threshold is different for each pa-
tient, a single threshold should be considered for each patient. For
example, mean and variance values of nth window were half the
average values of (n–1)th window. First, initial values of specifi c-
ity and sensitivity were calculated. Then, sensitivity value and the
rate of incorrect predictions were determined for other thresholds
by changing the threshold. Finally, the best threshold values with
the maximum specifi city and high sensitivity and the least false
predictio rate were selected.
The results generated based on thresholding method show that
it is able to distinguish high and low risk episodes and the seizure
prediction algorithm proposed based on HRV achieved 88.3 %
sensitivity and 86.2 % specifi city.
Conclusion
This study analyzed 8 HRV features to predict epileptic sei-
zures from clinical data collected from 7 patients. The analysis re-
sults showed that HRV features, such as mean HR, SD2
SD1 and
changed 5–10 minutes before seizure onset in all seizure episodes.
The possibility of realizing a HRV-based seizure prediction system
was shown through these analyses.
Our data showed that epileptic seizures are associated with
increased heart rate, indicating an increased sympathetic tone
and reduced vagal tone. HRV parameter changes are more dif-
fi cult to interpret and need further investigation. We credit that
these features are useful biomarkers for clinical use. Now, we are
in the process of working with physicians to fi nd out more fea-
tures for detection and prediction of epileptic seizures with HRV
signal information.
References
1. Martinerie J, Adam C, Le Van Quyen M, Baulac M, Clemenceau S,
Renault B, Varela F. Epileptic seizures can be anticipated by non–linear
analysis. Nat Med 1998; 4: 1173–1176.
2. Le van Quyen M, Adam C, Martinerie J, Baulac M, Clemenceau
S, Varela F. Spatio-temporal characterizations of nonlinear changes in in-
tracranial activities prior to human temporal lobe seizures. Eur J Neurosci
2000; 12: 2124–2134.
3. Navaro V, Martinerie J, Le Van Quyen M, Clemenceau S, Adam
C, Baulac M, Varela F. Seizure anticipation in human neocortical partial
epilepsy. Brain 2002; 125: 640–655.
4. Netoff T, Park Y, Parh K. Seizure prediction using cost-sensitive support
vector machine. In: Proceedings of 31st Annual International Conference
of the IEEE EMBS Minneapolis, Minnesota, USA, pp. 3322–3325, 2009.
5. Park Y, Luo L, Parhi KK, Netoff T. Seizure prediction with the spec-
tral power of EEG using cost-sensitive support vector machine. Epilepsia
2011; 5 (10): 1761–1770.
6. Malarvili MB, Mesbah M, Boashash B. Time-Frequency Analysis of
Heart Rate Variability for Neonatal Seizure Detection. EURASIP Journal
on Advances in Signal Processing, pp. 50396– 50406, 2007.
7. Harnod T, Yang CC, Hsin YL, Wang PJ, Shieh KR, Kuo TB. Heart
rate variability in patients with frontal lobe epilepsy. Seizure 2009; 18
(1): 21–25.
8. Britton JW, Ghearing GR, Benarroch EE, Cascino GD. The ictal
bradycar dia syndrome: Localization and lateralization. Epilepsia 2006;
47: 737–744.
9. Malik M. Heart Rate Variability:Standards of Measurement, Physi-
ological Interpretation, and Clinical Use. Ann Noninvas Electrocardiol
1996; 1: 151–181.
10. Gang Y, Malik M. Heart Rate Variability Analysis in General Medi-
cine. Ind Pacing Electrophysiol J 2003; 3 (1): 34–40.
11. Khazan IZ. Heart rate variability. The clinical handbook of biofeed-
back: A step by step guide for training and practice with mindfulness. John
Wiley & Sons publication, 2013, 97–117.
12. http://physionet.org/physiobank/database/szdb
13. Berntson GG, Bigger JT, Eckberg DL,Grossman P, Kaufmann PG,
Malik M. Heart rate variability:origins, methods, and interpretive caveats.
Psychophysiology 1997; 34: 623–648
14. Malliani A, Pagani M, Lombardi F, Cerutti S. Cardiovascular, neu-
ral regulation explored in the frequency domain. Circulation 1991; 84:
482–492.
15. Lerma C, Infante O, Grovas H, Jose MV. Poincare´ plot indexes
of heart rate variability capture dynamic adaptations after haemodialysis
in chronic renal failure patients. Clin Physiol Func Im 2003; 23: 72–80.
16. Toledo E, Gurevitz O, Hod H, Eldar M, Akselrod S. Wavelet analysis
of instantaneous heart rate: a study of autonomic control during throm-
bolysis. Am J Physiol Regul Integr Comp Physiol 2003; 284: 1079–1091.
17. Pan J, Tompkins WJ. A real-time QRS detection algorithm. IEEE
Transact Biomed Engineer 1985; 32: 230–236.
Received August 8, 2016.
Accepted October 27, 2016.
Feature type 2 hours before
seizure
10–5 min before
seizure p
RRI (ms)* 885.46±57.64 614.21±43.81 0.008
Mean HR* 73.43±8.58 106.32±13.23 0.006
LF 163.14±19.6 190.09±26.21 0.34
HF 256.28±44.57 177.63±31.72 0.028
LF/HF* 0.64±0.23 1.07±0.56 0.009
SD1 35.56±4.82 39.12±3.08 0.36
SD2 50.44±7.67 65.12±4.19 0.0025
SD2/SD1* 1.33±0.13 1.80±0.22 0.008
Tab. 1. Comparison of linear and non-linear features between 10–5
min and 2 hours before seizure.