Epileptic Seizure Prediction in Scalp EEG using One Dimensional

Local Binary Pattern based Features

Thasneem Fathima

, Paul Joseph K.

and M. Bedeeuzzaman

Dept. of Electrical Engineering, National Institute of Technology, Calicut, Kerala, 673601, India

Dept. of Applied Electronics and Instrumentation, MES College of Engineering, Kuttippuram, Kerala, 679573, India

Keywords: Epilepsy, Seizure Prediction, Electroencephalogram, Local Binary Pattern, Classifier.

Abstract: Seizure prediction will deeply improve the quality of life of epileptic patients. In this paper, a new method

of automatic seizure prediction is presented using one dimensional local binary pattern (1D-LBP) based

features in scalp electroencephalogram (EEG). In the feature extraction stage, the preictal and interictal

EEG signals were transformed to the 1D-LBP domain and histogram features were extracted. These features

were submitted to two different types of classifiers: linear discriminant analysis (LDA) and support vector

machine (SVM). In order to reduce the false prediction rate (FPR), a simple post processing stage was also

incorporated. The classification using SVM showed improvement over LDA in terms of sensitivity,

prediction time and FPR. The proposed method was evaluated using the scalp EEG recording from 13

patients with a total number of 47 seizures. It could achieve a sensitivity of 96.15%, an average prediction

time of 51.25 minutes with an FPR of 0.463.

1 INTRODUCTION

Epilepsy is a common neurological ailment that is

characterized by a sudden and recurrent brain

discharges termed “seizure.” These seizures reflect

the clinical signs of an excessive and hyper

synchronous activity of neurons in the brain (Fisher

et al., 2005). The disturbance of consciousness and

sudden loss of motor control often occur without any

warning. Experiences of staring, walking aimlessly

or loss of awareness may be harmless if they occur

at home. However, it can be life threatening if they

occur while the patient is driving, crossing a busy

street or swimming. Epileptic patients may have

some physiological changes prior to seizure onset.

These changes include changes in heart rate,

increase in cerebral oxygenation and blood oxygen

levels (Kerem and Gena 2005; Adelson et al., 1999;

Federico et al. 2005).

Recent studies show some changes in

Electroencephalogram (EEG) indicative of an

upcoming seizure and thereby give credence to the

idea of predicting seizures. The ability to herald

epileptic seizures far enough in advance would

reduce patients anxiety, alleviate the constraints in

everyday life and will improve the quality of life

and safety of epileptic patients (Winterhalder et al.,

2003). Knowing in advance that a seizure will occur

will be helpful in developing new treatment

strategies. It may lead to the design of more

effective drugs for the disruption of the brain’s

preparedness for an oncoming seizure. The

prediction will help many individuals whose

epilepsy cannot be controlled by medications, or

who are not in a position to undergo epilepsy

surgery. Also, long-term treatment with antiepileptic

drugs, which may cause cognitive or other

neurological side effects, could be reduced to a

targeted and short-acting intervention. The

medications could be replaced by electrical

stimulation or drug infusion activated only during

the pre-seizure period. The state just before the

occurrence of the seizure is termed as the ‘preictal

state’ and the normal state of a patient as the

‘interictal state’. Identifying the preictal states based

on EEG has been the goal of many research studies

on epileptic seizure prediction.

Based on the placement of electrodes the EEG

can be classified into two: scalp and intracranial. In

the scalp EEG, the electrodes are placed over the

scalp whereas in intracranial EEG, the electrodes are

placed inside the scalp. In this case, neurosurgeons

typically implant strip, grid or penetrating depth

electrodes under the dura mater. The signals

Fathima, T., K., P. and Bedeeuzzaman, M.

Epileptic Seizure Prediction in Scalp EEG using One Dimensional Local Binary Pattern based Features.

DOI: 10.5220/0005623000250033

In Proceedings of the 9th International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC 2016) - Volume 4: BIOSIGNALS, pages 25-33

ISBN: 978-989-758-170-0

recorded from intracranial EEG are on a different

scale of activity than the brain activity recorded

from scalp EEG. Scalp electrodes provide the global

information, whereas the intracranial one provides

the local information from the brain structure

(Cosandier-Rimélé et al., 2007). Low voltage or

high frequency components that cannot be seen

easily in scalp EEG can be seen clearly in

intracranial EEG. Also the scalp EEG is susceptible

to different types of artifacts and noise compared to

intracranial.

Several linear and nonlinear measures have been

reported in the literature to predict seizures from

intracranial EEG time series. Wavelet-based

nonlinear similarity index (Ouyang et al., 2007),

cross correlation and lyapunov exponents (Mirowski

et al., 2009), autoregressive coefficients (Chisci et

al., 2010), time, frequency and wavelet domain

features (Soleimani-B et al., 2012), mean absolute

deviation and wavelet entropy (Bedeeuzzaman et al.,

2012), spike rate (Li et al. 2013), statistical

dispersion measures (Bedeeuzzaman et al., 2014),

dominant amplitude and frequency components

(Wang and Lyu, 2014) are among the features

implemented in intracranial EEG. To make the

prediction techniques more clinically applicable,

methods based on scalp EEG have also been a

subject of research. Different features used for the

seizure prediction using scalp EEG include non

linear similarity (Quen et al., 2001), phase space

similarity measures (Hively and Protopopescu,

2003), average spiking rate (Schad et al., 2008),

phase synchrony measures (James and Gupta, 2009),

wavelet coherence values (Chiang et al., 2011),

variational Gaussian mixture model (Zandi et al.,

2011; Zandi et al., 2013), statistical moments and

spectral information (Direito et al., 2011) and

spectral power (Bandarabadi et al., 2012).

Local binary pattern (LBP) has been extensively

used for texture analysis of 2D images because of its

discriminative power. One dimensional local binary

pattern (1D-LBP), which is derived from LBP, has

been successfully applied to voice activity in speech

signals that are non-stationary in nature (Chatlani &

Soraghan 2010). So, it can be considered as an

effective approach of feature extraction of EEG

signals, which are inherently non-stationary (Kaya et

al. 2014).

The aim of this study is to develop a new

algorithm for prediction of epileptic seizures with

maximum possible sensitivity and prediction time.

In this paper, an algorithm based on 1D-LBP is

proposed to extract features from interictal and

preictal scalp EEG signals. These features are used

for the prediction of epileptic seizures. The proposed

scheme consists of two stages: extraction of features

from EEG signals and classification using the

extracted features. In the first stage, features are

extracted from interictal and preictal EEG signals. In

the classification stage, these features are applied to

two different classifiers: linear discriminant analysis

(LDA) and support vector machine (SVM). The

proposed method is tested on the scalp EEG dataset,

which is obtained from Massachusetts Institute of

Technology. Data from 13 epileptic patients with a

total number of 47 seizures are used in the present

study. The seizure prediction performance is

assessed in terms of sensitivity, prediction time and

false prediction rate (FPR).

Section 2 provides the details of the dataset used

and the proposed feature extraction method using

1D-LBP. In section 3, the performance of the

proposed system is evaluated by means of the results

obtained with the scalp EEG dataset. This section

includes the performance comparison with the

reported works using the same database. The paper

ends in Section 4 with some concluding remarks.

2 MATERIALS AND METHODS

2.1 Data used

The scalp EEG database used to evaluate the

prediction algorithm was recorded from patients

undergoing medication withdrawal for epilepsy

surgery evaluation in Children’s Hospital Boston

(Shoeb, 2009; Goldberger et al., 2000; CHB-MIT

Scalp EEG Database, 2015). Signals were recorded

with a sampling frequency of 256 Hz and 16-bit

resolution using the international 10-20 system of

electrode placement scheme. The seizures

experienced by the patients were judged by experts

and indicated the start and end of each seizure in the

EEG. The EEG data of each patient were segmented

into records of typically one hour duration. The

records containing one or more seizures are called

seizure records and those without seizures are

labeled as non-seizure records. Most of the EEG

files contain 23 channels whereas a few contain 24

or 26 channels.

2.2 Local Binary Pattern

The local binary pattern(LBP) was introduced by

Ojala et al., (1996) for the texture analysis, defined

as gray scale invariant texture measure, derived from

comparison with the local neighborhood. LBP has

BIOSIGNALS 2016 - 9th International Conference on Bio-inspired Systems and Signal Processing

also been used for face recognition, dynamic texture

recognition and shape localization (Guo et al.,

2010).

An LBP code for each pixel in a two

dimensional image is produced by thresholding the

neighboring values with the value of the center

pixel. The definition of LBP is extended to include

all circular neighborhoods with any number of

pixels. In general, LBP is denoted as LBP

P,R

where

P is the number of neighbors involved and R is the

radius of the model (Fig. 1).

The basic version of LBP considers only eight

neighbors. As shown in Fig. 1, the LBP operator

labels each pixel, using the value of the center pixel

as a threshold value. Each pixel is assigned a value

1 if it is greater than or equal to the threshold value,

otherwise it takes 0. Thus the binary code is

produced using these values that gives the local

structural information around the given pixel. Each

pixel value is replaced with the decimal value

corresponding to this binary code (Chatlani and

Soraghan, 2010).

Figure 1: Calculation of LBP codes for a 3x3 sample

block. P=8, R=1.

Given a pixel in the image, an LBP is computed

comparing it with the local neighborhood.

() ()

zGv Gv=−

(1)

() ()2

LBP v S z



(2)

where the sign function S(.) is given by,

1, z 0

S(z)

0, z 0

≥







is the location of the center pixel,

is the

location of the

neighboring pixel,

(.)G

is the

pixel intensity value.

2.3 One Dimensional Local Binary

Pattern (1D-LBP)

1D-LBP was first introduced in (Chatlani and

Soraghan, 2010) for applying in speech signals

which are non stationary in nature. It is adapted from

the implementation steps in 2D LBP. The LBP code

for a neighborhood of sampled data is produced by

thresholding the neighboring samples against the

centre sample of a processing window. This

procedure is iteratively done across the entire signal

and a segment of the 1-D signal is alternatively

described by a sparser occurrence histogram of LBP

codes.

The 1-D LBP operating on a sample value

y[i] is

defined as

()

[]

[][][]

[]







≥





















−+++

−−+



−

Where

][

, for y

iyriyS

riyS

iyLBP

(3)

Where P is the number of neighboring samples

thresholded around the centre sample from the

signal

][iy

of length N for

]2/:2/[ PNPi −=

. The sign function

[.]S

makes a P- bit binary code from these

differences. The decimal value of this binary code

gives a unique LBP code. The 1D- LBP operator is

described step by step in Fig. 2 using a sample

segment of an EEG signal where P is set to 8. The

four neighboring samples taken before (N0, N1, N2,

N3) and after (N4, N5, N6, N7) are threshold against

the centre sample (NC). If the neighboring value is

greater than or equal to the center value, the

assigned value is 1, otherwise 0. Thus a binary code

of 11110000 is produced and the corresponding

decimal gives the LBP code 240. The LBP codes

represent the local structure information around the

given sample using the difference between the

sample and its neighbors. These differences cluster

near zero for constant or slowly varying signals

Epileptic Seizure Prediction in Scalp EEG using One Dimensional Local Binary Pattern based Features

whereas at peaks and troughs the differences will be

relatively large. At edges, the differences in some

directions will be larger than those from other

directions.

Figure 2: Computation of 1D local binary pattern.

LBP signal is formed by applying the above

procedure to all samples, which has values ranging

from 0 to 255. A segment of the interictal EEG

signal of 1 second duration of patient 1 is depicted in

Fig. 3 (a) and the corresponding LBP applied signal

is given in Fig. 3(b). Fig. 3(c) and Fig. 3 (d) shows

the values corresponding to one second of the

preictal EEG of first seizure of patient1.

The distribution of LBP can describe the local

patterns formed from

][iy

[]

()()



−≤≤

liyLBPH

(4)

Where

nl ...2,1=

and

is the number of

histogram bins and

()

ji,δ

is the Kronecker delta

function. The occurrences of each LBP code are

plotted as a histogram. The numbers of occurrences

corresponding to 8 histogram bins are selected as the

features for classification between interictal and

preictal signals.

Figure 3: (a) A 1-second segment of interictal EEG signals

of patient 1. (b) Interictal signal transformed to LBP

domain, which has values ranging from 0 to 255. (c) A 1-

second segment of preictal EEG signal from the first

seizure of patient 1. (d) Preictal signal transformed to LBP

domain.

3 RESULTS

The purpose of this study is to extract the

representative features from EEG by utilizing the

potential of 1D-LBP for the prediction of epileptic

seizures. The EEG classification system using the

proposed 1D-LBP based feature extraction is

depicted in Fig. 4. The raw EEG signals were given

as the input to the classification system and the

output was the classified EEG pattern. The features

were extracted over the non overlapping frames of 1

minute length. Firstly the EEG signal in the time

domain is transformed into the LBP domain through

the process described in section 2.3.

The histogram of LBP codes is produced as an

alternative representation of the signal. The numbers

of occurrences of the LBP values in 8 histogram

bins are selected as the discriminating features. For

23 channels, this feature vector is of dimension 23x8

for each frame. In order to reduce the feature

dimension, averaging is done across the channels,

reducing the dimension to 1x8 for one frame.

Sample histogram features, averaged across 23

channels for patient1 for 1 minute data is given in

BIOSIGNALS 2016 - 9th International Conference on Bio-inspired Systems and Signal Processing

Figure 4: Schematic of the EEG classification system.

Table1. These values show a significant difference

in magnitude and were given as input to the

classifier.

In the classification stage, the 1D-LBP based

features are applied to a classifier, for the

classification between interictal and preictal EEG

signals. Here the performances of the proposed

method are evaluated using two different classifiers:

Linear Discriminant Analysis (LDA) and Support

Vector Machine (SVM).

Only those seizures with at least one hour

preictal data are considered for the study. In the case

of seizures occurring without much time gap, only

the first seizure is considered for the prediction.

Hence, the evaluation of the algorithm is done on

13 patients and 47 seizures that satisfy this

condition. The number of seizures used for training

and testing is given in Table 2. For example in the

case of patient number 5 (P5), 2 seizures are used

for training and 2 for testing. The one hour interval

preceding each seizure onset has been used to

produce training and test samples.

During classification, the labels ‘0’ and ‘1’ are

assigned for interictal and preictal data respectively.

Whenever a change from ‘0’ to ‘1’ occurs, the

prediction system will raise an alarm. If the alarm is

in the preictal period, it is considered as prediction

and if it is in the interictal period, taken as false

prediction. The one hour preictal period prior to

each seizure under test is considered to evaluate the

prediction. Prediction time is taken as the time gap

between the first alarm and the seizure onset. The

interictal data is tested to check whether it gives any

misclassification. A simple post processing stage has

also been incorporated to reduce the FPR. In the post

processing phase, consecutive ‘1’ labels are

searched for giving a ‘1’ in the output (Fig. 5).

Figure 5: Post processing scheme for 2- minute window:

Two consecutive ‘1’ labels give a ‘1’ in the output.

The performance of the prediction system is

analyzed in terms of sensitivity, prediction time and

FPR. Sensitivity and prediction time for each patient

using LDA classifier is given in Table 2. For patient

5, 2 seizures were used for testing and 2 for training.

The algorithm with a 1-minute window correctly

predicted the two tested seizures, thus exhibited

100% sensitivity. The average prediction time

(APT) of the 2 tested seizures for patient 5 is found

to be 59 minutes. As the algorithm using 1-minute

window predicted all the tested seizures of all the

patients, an average sensitivity of 100% achieved.

Also, the average APT with a 1-minute window

width was 57.08 minutes. FPRs of each patient for 1,

2, 3, 4 and 5 minute window widths are shown in

Fig. 6(a). Although the results are good in terms of

sensitivity and prediction time, the average FPR for

1 minute window width was 3.69, which is a bit high

Figure 6: (a) FPR for each patient for different window

widths using LDA classifier (b) FPR for each patient for

different window widths using SVM classifier.

Epileptic Seizure Prediction in Scalp EEG using One Dimensional Local Binary Pattern based Features

Table 1: Sample features extracted for one minute EEG of patient 1.

Class Bin1 Bin2 Bin3 Bin4 Bin5 Bin6 Bin7 Bin8

Preictal 5027 851 228 752 545 350 855 23

Interictal 4126 1038 259 961 759 392 1016 25

Table 2: Sensitivity and prediction time using 1D-LBP based features and LDA classifier for different window widths. SE-

Sensitivity, APT- Average prediction time, NS-Number of seizures.

Patient No.

NS for training

NS for testing

Window width in minutes

1 2 3 4 5

(%)

APT

(min)

(%)

APT

(min)

SE (%)

APT

(min)

(%)

APT

(min)

(%)

APT

(min)

P1 2 2 100 59 100 58 100 57 100 56 100 55

P2 1 1 100 58 100 56 100 54 100 52 100 50

P3 3 1 100 59 100 58 100 57 100 56 100 55

P4 1 1 100 59 100 58 100 57 100 56 100 55

P5 2 2 100 59 100 58 100 57 100 56 100 55

P6 5 4 100 59 100 46.5 100 42 100 41 50 53.3

P7 2 1 100 38 0 - 0 - 0 - 0 -

P8 2 2 100 59 100 58 100 57 100 56 100 37.5

P9 2 2 100 59 50 58 50 57 50 56 50 30

P10 2 1 100 58 100 56 100 54 100 48 100 45

P11 2 1 100 59 100 58 100 57 100 56 100 50

P12 2 1 100 59 100 58 100 57 100 56 100 55

P13 1 1 100 57 100 56 100 45 100 44 100 45

Average 100 57.08 88.46 56.41 88.46 54.00 88.46 52.45 84.61 50.25

Table 3: Sensitivity and average prediction time using 1D-LBP based features and SVM classifier for different window

widths. NS- Number of seizures, SE- Sensitivity, APT- Average prediction time.

Patient

No.

Window width in minutes

1 2 3 4 5

(%)

APT

(min)

(%)

APT

(min)

(%)

APT

(min)

(%)

APT

(min)

(%)

APT

(min)

P1 100 59 100 58 100 52.5 100 52 100 50

P2 100 58 100 56 100 54 100 48 100 45

P3 100 59 100 58 100 57 100 56 100 55

P4 100 59 100 58 100 57 100 56 100 55

P5 100 59 100 58 100 57 100 56 100 55

P6 100 56.25 100 46.5 100 44.5 100 43.25 100 40

P7 100 59 100 58 100 57 100 56 100 55

P8 100 59 100 58 100 57 100 56 100 55

P9 100 46 50 58 50 57 50 44 50 20

P10 100 59 100 58 100 57 100 56 100 55

P11 100 59 100 58 100 57 100 56 100 50

P12 100 59 100 58 100 57 100 56 100 55

P13 100 57 100 56 100 45 100 44 100 45

Average 100 57.55 96.15 56.70 96.15 54.33 96.15 52.94 96.15 51.25

for seizure prediction. The FPR for each patient,

given in Fig. 6 (a), shows a decrease in FPR with

respect to the increase in window width. Sensitivity

and APT using the SVM classifier, given in Table 3,

show a slight increase compared to the method using

LDA classifier. The FPR using the SVM classifier

BIOSIGNALS 2016 - 9th International Conference on Bio-inspired Systems and Signal Processing

for each patient is given in Fig. 6(b) and shows a

small decrease compared to the other one.

Comparison of average sensitivity, average APT and

average FPR using LDA and SVM classifier is

depicted in Fig. 7. The SVM classifier shows

improvement in all the three parameters of

evaluation, compared to the LDA classifier.

4 DISCUSSION

A seizure prediction method based on 1D-LBP in

scalp EEG has been presented in this paper. The

comparison of prediction time and sensitivity

obtained using 5 minute window width and SVM

classifier with other methods using the same dataset

is given in Table 4. Wavelet coherence values were

used as features and tested on 7 patients from the

same dataset in (Chiang et al., 2011). They could

predict seizures of 4 patients out of 7, giving a

sensitivity of 57.14%, but it didn’t report the

prediction time and FPR. Another algorithm using

the variational Bayesian Gaussuan mixture model

for prediction was tested on 3 patients of the dataset

in (Zandi et al., 2011). Compared to this, the

proposed work which is experimented with 13

patients shows an improvement in sensitivity from

83.8% to 96.15%. Also, APT is increased from 19.8

to 51.25 minutes. But the FPR reported in (Zandi et

al., 2011) is 0.165 whereas in this work it is 0.463.

An increased window width may help to decrease

the FPR. But it will affect the sensitivity and

prediction time. Though the increased FPR obtained

in the proposed work can be ascribed to the

increased number of patients in contrast to the 3

patients used in (Zandi et al., 2013), this has to be

reduced to make the method useful for real life

situations.

Table 4: Comparison of results with other methods using

same dataset.

Method Feature

No. of

patients

analysed

APT

(minutes)

Sensitivity

(%)

(Chiang et

al. 2011)

Wavelet

coherence

values

7 - 57.14

(Zandi et al.

2013)

Gaussian

Mixture Model

3 19.8 83.8

Proposed

method

1D-LBP

Based

13 51.25 96.15

To the best of authors’ knowledge, the only work

reported in EEG signal analysis that used 1D-LBP is

(Kaya et al., 2014). This work was for the epileptic

seizure detection, a retrospective analysis of EEG

signals to find out the seizure that has already

happened. Whereas, in the proposed work 1D-LBP

is used for epileptic seizure prediction which

involves the analysis of EEG signals for an

oncoming seizure well before its occurrence.

Even though the proposed method is developed

for noninvasive scalp EEG, it may also be used for

the intracranial EEG recordings. As the artifacts and

noise will be less in the intracranial EEG, a better

performance can be anticipated when applied to

depth recording.

As the method presented in this paper used the

data for training and testing for the same patient, the

algorithm can predict only the seizures of a patient

whose prior database is already available. Therefore,

more studies should be performed by incorporating

new features to make a patient independent method.

Also the data used for this study was acquired from

patients under medical care in hospitals. So to

broaden the utility of the proposed method, it has to

be applied to the continuous data recorded during

routine daily activities.

Figure 7: Performance comparison of classifiers: LDA and SVM. (a) Average Sensitivity (b) Average APT (c) Average

FPR.

Epileptic Seizure Prediction in Scalp EEG using One Dimensional Local Binary Pattern based Features

5 CONCLUSIONS

A new patient specific seizure prediction algorithm

based on 1D-LBP in scalp EEG has been proposed

in this study. The idea is to classify between preictal

and interictal EEG using appropriate features. For

this purpose, histogram features are extracted from

the 1D-LBP applied signal. These features are

submitted to two different classifiers: LDA and

SVM. In order to reduce the false alarms, a simple

post processing is also incorporated. The

classification using SVM shows improvement over

LDA in terms of sensitivity, prediction time and

FPR. When this algorithm is applied to scalp EEG

recordings from 13 patients with a total number of

47 seizures, it could achieve a sensitivity of 96.15%,

an APT of 51.25 minutes with an FPR of 0.463.

Comparison with the previous works using the same

database shows improvement in terms of APT and

sensitivity.

REFERENCES

Adelson, P. D., Nemoto, E., Sheuer, M., Painter, M.,

Morqan, J., Yonas, H., 1999, ‘Noninvasive continuous

monitoring of cerebral oxygenation preictally using

near-infrared spectroscopy: A preliminary report’,

Epilepsia,vol. 40, pp. 1484-1489.

Bandarabadi, M., Teixeira, C. A., Direito, B., Dourado,

A., 2012, ‘Epileptic Seizure Prediction based on a

bivariate spectral power methodology’, Conf. Proc.

IEEE Eng. Med. Biol. Soc. 2012, 5943-5946.

Bedeeuzzaman, M., Fathima, T., Khan, Y. U., Farooq, O.,

2012. Mean absolute deviation and wavelet entropy

for seizure prediction. J. Med. Imag. Health. Inform. 2,

238-243.

Bedeeuzzaman, M., Fathima, T., Khan, Y. U., Farooq, O.,

2014. Seizure prediction using statistical dispersion

measures of intracranial EEG. Biomed. Signal.

Process. Control. 10, 338-341.

Chatlani, N., Soraghan, J. J., 2010. Local binary patterns

for 1-D signal processing. 18

Europian Signal

Process. Conf. Denmark, 95-99.

CHB-MIT Scalp EEG Database. Available from: < http://

physionet.org/physiobank/database/chbmit>. (10

February 2015).

Chiang, C-Y., Chang, N. F., Chen, T. C., Chen, H. H.,

Chen, L. G., 2011. Seizure Prediction Based on

Classification of EEG Synchronisation Patterns with

On-line Retraining and Post-Processing Scheme. Conf.

Proc. IEEE Eng. Med. Biol. Soc. 2011, 7564-7569.

Chisci, L., Mavino, A., Perferi, G., Sciandrone, M., Anile,

C., Colicchio, G., Fuqqetta, F., 2010. Real time

epileptic seizure prediction using AR models and

Support Vector Machine. IEEE Trans. on Biomedical

Eng. 57, 1124-1132.

Cosandier-Rimélé, D., Badier, J.M., Chauvel, P.,

Wendling, F., 2007. Modeling and interpretation of

scalp-EEG and depth-EEG signals during interictal

activity. Proc. 29th Annual Int. Conf. IEEE EMBS.

4277-4280.

Direito, B., Ventura, F., Teixeira, C., Dourado, A., 2011.

Optimized Feature Subsets for Epileptic Seizure

Prediction Studies. Conf. Proc. IEEE Eng. Med. Biol.

Soc. 2011 1636-1639.

Federico, P., Abbott, D. F., Briellmann, R. S., Harvey, A.

S., Jackson, G. D., 2005. Functional MRI of the pre-

ictal state. Brain.128, 1811-1817.

Fisher, R. S., Boas, W. E. , Blume, W., Elger, C., Genton,

P., Lee, P., Engel, J. Jr., 2005. Epileptic Seizures and

Epilepsy: Definitions Proposed by the International

League Against Epilepsy (ILAE) and the International

Bureau for Epilepsy (IBE). Epilepsia. 46, 470-472.

Goldberger, A. L., Amaral, L. A. N., Glass, L.,

Hausdorff, J. M., Ivanov, P. C., Mark, R. G., Mietus,

J. F., Moody, G. B., Peng, C.-K., Stanley, H. E. 2000.

Physiobank, physiotoolkit, and physionet:

Components of a new research resource for complex

physiologic signals. Circ. 101, e215-e220.

Guo, Z., Zhang, L., Zhang, D., 2010. A completed

modeling of local binary pattern operator for texture

classification. IEEE Trans. image process. 19, 1657-

1663.

Hively, L., Protopopescu, V., 2003. Channel-consistent

forewarning of epileptic events from scalp EEG. IEEE

Trans. Biomed. Eng. 50, 584-593.

James, C. J., Gupta, D., 2009. Seizure prediction for

epilepsy using a multi-stage phase synchrony based

system. Conf. Proc. IEEE Eng. Med. Biol. Soc. 2009,

25-28.

Kaya, Y., Uyar, M., Tekin, R., Yildirim, S., 2014. 1D-

local binary pattern based feature extraction for

classification of epileptic EEG signals. Appl.

Mathematics. Computation. 243, 209-219.

Kerem, D. H., Geva, A. B., 2005. Forecasting epilepsy

from the heart rate signal, Med. Biological Eng.

Computing. 43, 230-239.

Li, S., Zhou, W., Yuan, Q., Liu, Y., 2013. Seizure

Prediction Using Spike Rate of Intracranial EEG.

IEEE Trans. Neural Systems. Rehabilitation. Eng. 21,

880-886.

Mirowski P. W., Madhavan, D., LeCun, Y., Kuzniecky,

R., 2009. Classification of patterns of EEG

synchronization for seizure prediction. Clinical.

Neurophysiol. 120, 1927-1940.

Ojala, T., Pietikäinen, M., Harwood, D., 1996. A

comparative study of texture measures with

classification based on feature distributions. Pattern.

Recognit. 29, 51-59.

Ouyang, G., Li, X., Li, Y., Guan, X., 2007. Application of

wavelet-based similarity analysis to epileptic seizures

prediction. Computers. Biol. Med. 37, 430 – 437.

Quen, M. L. V, Martinerie, J., Navaro, V., Boon, P.,

D’Havé, M., Adam, C., Renault, B., Varela, F.

Baulac, M., 2001. Anticipation of epileptic seizures

from standard EEG recordings. Lancet, 357, 183-188.

BIOSIGNALS 2016 - 9th International Conference on Bio-inspired Systems and Signal Processing

Schad, A., Schindler, K., Schelter, B., Maiwald, T.,

Brandt, A., Timmer, J., Bonhage, A. S., 2008.

Application of a multivariate seizure detection and

prediction method to no-invasive and intracranial

long-term EEG recordings. Clin. Neurophys. 119, 197-

211.

Shoeb, A. Application of machine learning to epileptic

seizure onset detection and treatment. Ph. D.

dissertation, Massachusetts Inst. Techno. Cambridge,

Sep. 2009.

Soleimani-B., H., Lucas, C., Araabi, B. N., Schwabe, L.,

2012. Adaptive prediction of epileptic seizures from

intracranial recordings. Biomed. Signal Process.

Control, 7, 456 464.

Wang, N., Lyu, M. R., 2014. Extracting and Selecting

Distinctive EEG Features for Efficient Epileptic

Seizure Prediction. IEEE J. Biomed. Health. Inform,

doi: 10.1109/JBHI.2014.2358640.

Winterhalder M., Maiwal, T., Voss, H. U.,

Aschenbrenner-Scheibe, R., Timmer, J., Schulze-

Bonhage, A., 2003. The seizure prediction

characteristic: A general frame work to assess and

compare seizure prediction methods. Epilepsy.

Behave.4, 318-325.

Zandi, A. S., Tafreshi, R., Javidan, M., Dumont, G. A.,

2013. Predicting Epileptic Seizures in Scalp EEG

Based on a Variational Bayesian Gaussian Mixture

Model of Zero-Crossing Intervals. IEEE Trans.

Biomed. Eng., 60, 1401-1413.

Zandi, A. S., Dumont, G. A., Javidan, M., Tafreshi, R.,

2011 Epileptic Seizure Prediction using Variational

Mixture of Gaussians. Conf. Proc. IEEE Eng. Med.

Biol. Soc. 2011, 7549-7569.

Epileptic Seizure Prediction in Scalp EEG using One Dimensional Local Binary Pattern based Features