Advancements in Computer Aided Methods for EEG-based Epileptic

Detection

Malik Anas Ahmad

1,2

, Waqas Majeed

and Nadeem Ahmad Khan

Signal Image & Video Processing Lab, SBASSE, LUMS, Lahore, Pakistan

Robotics and Intelligent Systems Engineering Lab, SMME, NUST, Islamabad, Pakistan

Keywords: Epilepsy, Electroencephalography (EEG), Machine Learning, Biomedical Signal Processing.

Abstract: During the diagnosis of epilepsy, computer aided methods can significantly supplement a neurologist by

automatically identifying the epileptic patterns in an EEG. In the last decade immense amount of work has

been done in the field of EEG based computer aided diagnosis of epilepsy. Even after so much work these

tools are not getting used up to their full potential. In this paper we have very briefly discussed some of the

previously used signal processing and machine learning techniques which are proposed for epileptic pattern

detection. We have concluded this paper by suggesting some additions in the previous method which can

make these systems more helpful, detailed and precise for the neurologist.

1 INTRODUCTION

Epilepsy is a recurring neurological disorder, which

is characterized by excessive neural activity yield in

the brain. Almost 1% of the human population

suffers from epilepsy (WHO | Epilepsy 2012)

(Adelia, Zhoub & Dadmehrc 2003). Detection and

localization of abnormal, epilepsy-related brain

activity is very important for diagnosing and curing

of an epileptic disorder. Electroencephalogram

(EEG) is a method for recording of electrical activity

along the scalp. EEG signal represents fluctuations

in the voltage caused by the flow of ionic current in

the neurons. Epileptic seizures are accompanied by

unique patterns in EEG, and therefore EEG is widely

used to detect and locate the epileptic seizure and

zone.

Duration of a typical diagnostic EEG recording

varies from 40 minutes to a few hours. However,

prolonged EEG is opted if a seizure is not detected

in shorter recordings. A prolonged EEG can last as

long as 72 hours. Diagnostic procedures like this

generate a huge amount of data to be manually

inspected by the neurologist. Manual inspection of

all of the data for multiple patients could prove to be

a daunting task for a neurologist.

Computer assisted analysis of an EEG

supplements a neurologist in efficiently analysing

the EEG data. It highlights the epileptic patterns in

the EEG up to a significant level, thus reducing the

data to be analysed and lessening up the fatigue.

These analysis software tools apply different signal

processing and machine learning techniques on the

EEG data to detect the epochs with epileptic

patterns. Currently available commercial computer

assisted diagnosis tools for epilepsy are not a lot

user-friendly and lack adaptability/intelligence

(NeuroExplorer Home n.d.) (Neuralynx ~ Spike Sort

3D Software n.d.). These software tools require the

clinician to have an understanding of signal

processing algorithms to exploit the full potential of

the software (Tucker-Davis Technologies n.d.)

(Brain Products GmbH / Products & Applications /

Analyzer 2 n.d.). For this they hire technicians and

rely on them which make this analysing procedure

prone to misinterpretation and over-interpretation as

the manual marking get dependent on the expertise

of the technicians (Benbadis and Tatum 2003) rather

than the clinician himself.

In the next section we will briefly describe about

the existing work in the field of computer assisted

analysis of EEG for Epilepsy. Then in Section 3 we

will discuss the key factors involved in a computer

assisted analysis of EEG. After wards in Section 4

we will discuss the future research direction which

according to our analysis can be followed to

improve the existing work.

289

Anas Ahmad M., Majeed W. and Ahmad Khan N..

Advancements in Computer Aided Methods for EEG-based Epileptic Detection.

DOI: 10.5220/0004912802890294

In Proceedings of the International Conference on Bio-inspired Systems and Signal Processing (BIOSIGNALS-2014), pages 289-294

ISBN: 978-989-758-011-6

 2014 SCITEPRESS (Science and Technology Publications, Lda.)

2 EXISTING WORK

EEG signals are non-stationary. Methods for

analysing non-stationary signals, such as Discrete

Wavelet Transform (DWT), Empirical Mode

Decomposition (EMD) and time-frequency analysis,

have been frequently used for automated seizure

detection using EEG signals. Epileptic seizures give

rise to changes in certain frequency bands which are

δ (0.4 – 4 Hz), θ (4 – 8 Hz), α (8 - 12 Hz) and β (12

– 30 Hz) (Adelia, Zhoub and Dadmehrc 2003).

That’s why usually the spectral content of the EEG

is used for diagnosis.

Usually the approach toward the detection of

epileptic patterns start with dividing the EEG data in

multiple small epochs, then multiple signal

processing steps are applied on these epochs to

extract out the features which are then used to

classify that epoch as epileptic or non-epileptic.

Majority of the work done in the line of epileptic

pattern detection usually do not involve fusion of

information obtained from multiple channels.

Instead, all the channels are processed in

series/sequentially, as if the EEG signal source is

one long signal instead of multiple parallel signals.

However Chang et al. (Chang et al., 2010)

appreciated the effects of multiple channels’ being

processed in parallel. They grouped 0.3 sec epochs

of multiple channels simultaneously in five different

clusters to avoid noise. Then they used FastICA to

discriminate between features, noise and background

of the signal. They then applied DWT with

Daubechies-4 (db4) as mother wavelet on the two

most independent parts of the signal. Then they

applied customized threshold to classify them as

epileptic or not. In this work Chang et al. showed

that consideration of multiple channels in group

improve the accuracy of your system.

Xanthopoulos et al. (Xanthopoulos et al., 2010)

used sliding variance on Continuous Wavelet

Transformed (CWT) epochs to detect the clinically

important epileptic patterns up to 98.625% accuracy.

Luo et al.’s (Luo and Luo 2010) work advocates

the importance of feature reduction techniques like

Principal Component Analysis (PCA). He evaluated

the effectiveness of six features. The PCA showed

that almost three of the six features has contribution

ratio of 79 %. So discarding of other three features

could improve the processing time without a

significant damage to the classification accuracy.

His stance was verified by the Artificial Neural

Network (ANN) classifier whose accuracy only

dropped by 2.5% with the exclusion of three

features. The six features were Hurst Index,

Standard Deviation and Periodicity, Shannon

Entropy, Approximate Entropy and periodicity of

smoothed EEG signal, where the first three are the

most contributing features.

Petersen et al.’s (Petersen et al., 2011) work

shows that to detect the generalised seizure using

only one channel, usage of the energy of the detail

coefficients of the wavelet transformed one second

epoch of a F7-FP1 in an SVM classifier can result

with as good as 99.1% sensitivity.

In Abdullah et al.’s (Abdullah, Abdullah and

Abdullah 2012) work Hidden Markov Model

(HMM) was applied on vector quantized Stationary

Wavelet Transform coefficients of intracranial EEG

signal. Their work resulted with 96.38% and 96.82%

average sensitivity and specificity respectively.

Sousa et al. (Sousa, Mendes and Ribeiro, 2012)

studied how rhythms analysis identifies the various

events recorded in the EEG. Their work resulted

with 95.5% accuracy.

Abdullah et al. (Abdullah, Saufiah and Ibrahim

2012) simultaneously used features extracted from

DWT and Fourier transform in an ANN classifier.

Their work resulted with 98.889% accuracy.

Khan et al. (Khan, Rafiuddin and Farooq 2012)

used energy and normalized coefficients of variance

of multi-level DWT coefficients. These features

were used by a Linear Discriminant Analysis (LDA)

to classify the EEG epochs with an accuracy of

91.8%.

Due to the heavy computational burden of

marching pursuit (MP) algorithm (Guo et al., 2012)

proposed a reduce complexity of sparse

representation to adopt harmony search method in

searching the best atoms. Their efforts resulted with

huge amount of improvement in the latency. Wang

et al. (Wang et al., 2012) used these features with

Adaptive Neuro-Fuzzy Inference System (ANFIS)

as a classifier. Here they integrated the artificial

neural networks and fuzzy logic together. Their

effort resulted in 97.4% accuracy.

Choi et al. (Choi, Zeng and Qin, 2012) selected

the optimal frequency band features by using the

Sequential Floating Forward Selection (SFFS)

algorithm. These features were fed to three types of

classifier. These classifiers were linear, quadratic

and cubic discriminant function. They found QDF

with best accuracy which was 97.2%.Sezer et al.

(Sezer, Işik and Saracoğlu, 2012) tested multiple

types of ANN and found Elman method to be most

accurate along with DWT as feature extraction

method.

Alam et al. (Alam and Bhuiyan, 2013) used the

higher order statistical parameters like variance,

BIOSIGNALS2014-InternationalConferenceonBio-inspiredSystemsandSignalProcessing

290

skewness and kurtosis of empirical mode

decomposed EEG signal with ANN.

Seng et al.’s (Seng et al., 2013) used simple

features like mean, variance, dominant frequency,

mean of power spectrum and the signal data itself of

the EEG epochs in linear SVM. They tried multiple

epoch sizes which were 23.6 sec, 11.5 sec, 5.8 sec,

and 1 sec. The result showed that smaller epoch size

results in better accuracy whereas bigger epoch size

results in better latency.

Ocbagabir et al. (Ocbagabir, Aboalayon and

Faezipour 2013) used Butterworth band pass filter to

decompose the EEG signal into 5 sub-bands and

then used Energy, Entropy, and Standard Deviation

as features for a SVM classifier. This classification

approach resulted in 95% accuracy.

Kaleem et al. (Kaleem, Guergachi and Krishnan

2013) applied a novel variation of the EMD called

Empirical Mode Decomposition-Modified Peak

Selection (EMD-MPS). They used Energy, sum of

the amplitude spectrum, sparsity of the amplitude

spectrum and the sum of derivative of the amplitude

spectrum as the input features to a simple 1-NN

classifier which resulted with 98.2% accuracy.

Murugavel et al. (Muthanantha Murugavel et al.,

2013) used a novel feature named as Combined

Seizure Index as a feature which they extracted from

wavelet packet coefficients. These features in a

multi scale SVM resulted with 97.3% accuracy.

3 DISCUSSION

After providing a short literature survey we proceed

to discuss the state-of-the art research from different

important perspectives.

3.1 EEG Databases

There are two major databases for scalp EEG which

are used for validating the performance of the

automatic epileptic pattern detection algorithms.

First of the two data sets are from Klinik fur

Epileptologie at the Universitat Bonn, German

(Klinik fur Epileptologie, Universitat Bonn n.d.).

This dataset has two sets of 100-channel EEG data

consisting of normal and epileptic subjects with

segment duration of 23.6sec, 4096 sampling point

and 173.61 Hz sampling frequency. The first EEG

data set is a scalp EEG of 5 normal subjects. The

other EEG data set is a intracranial EEG of 5

epileptic patients, recorded during the occurrence of

the epileptic seizures.

The second most used EEG database is the

CHBMIT scalp EEG database (Goldberger et al.

2000). It been provided by Children Hospital Boston

and is available at physionet website (Shoeb, 2000).

This database consists of 916 hours of continuous

scalp EEG recordings collected from 24 subjects

suffering from intractable seizures. Out of 664 EEG

recordings files, 129 files consisted of one or more

seizures. The 23 channel EEG signal has a sampling

frequency of 256 Hz with 16 bit resolution.

In the following lines we will discuss some of

our observations on these databases.

3.1.1 Size of Database & over-fitting

The first database is less versatile in comparison to

the second one and it also has lesser number of

examples. This makes the classifiers trained on the

first dataset more prone to over-fitting. Probably this

is the reason behind algorithms not performing with

the cited accuracy on the real life data.

3.1.2 Labelling

Another issue with both of these databases is that the

labelling does not specify the epileptic pattern type.

Apparently they are labelled for generalized 3Hz

spike & wave which is a symptom for absence

seizure which is one of many types of epilepsy.

3.2 Epileptic Patterns

EEG recorded from epileptic patient exhibits

distinctive signal patterns. Patterns like Spikes,

Sharp wave, Benign epileptiform discharges of

childhood, Spike-wave complexes, Slow spike-wave

complexes, 3-Hz spike-wave complexes, Polyspikes,

Hypsarrhythmia, Seizure pattern, Status pattern are

considered as epileptiform (Lüders and Noachar,

2000; Noachtar et al., 1999).

Mostly 3Hz spike & wave detection has

remained the focus of the past work. Whereas other

epileptic patterns detection are usually ignored.

Sousa et al. (Sousa, Mendes and Ribeiro, 2012) are

among those few people who have addressed the

detection of multi-type epileptic patterns.

3.3 Feature Extraction & Reduction

3.3.1 Epoch Size

The choice of epoch size depends a lot on the

sampling frequency and feature extraction

techniques. Epoch size as low as 0.3 sec (Chang et

al., 2010) and as high as 23.6 sec (Ocbagabir,

Aboalayon and Faezipour, 2013) has been cited, but

AdvancementsinComputerAidedMethodsforEEG-basedEpilepticDetection

291

the most commonly used epoch size is of 1 sec. A

comparison among 23.6 sec, 11.5 sec, 5.8 sec, and 1

sec in Seng et al.’s work resulted with 1 sec epoch

size to be working best in terms of accuracy.

3.3.2 Feature Types

The most commonly used feature extraction method

is WT with db4 or Morelet as a mother wavelet. This

also seems to be a promising method yielding the

best performance figures. Majority of the algorithms

use detailed coefficients’ energy, variance, standard

deviation, entropy, mean, maxima and minima or

any combination or slight modification of these as

features whereas few algorithms use the detailed

coefficients without any modification.

Other than Wavelets, Fourier transform, EMD,

MP, SFFS or their modified versions has also been

cited as the feature extraction techniques.

3.3.3 Feature Reduction

There are few examples where some authors like

Luo et al. (Luo and Luo, 2010) have used feature

reduction techniques like PCA before classifying.

The motivation behind applying reduction is to

avoid noise and redundancy.

3.4 Classifiers

SVM, Linear/ Quadratic/ Cubic Discriminant

Analysis, ANN, Genetic Algorithms (GA), HMM,

Fuzzy Based classifier and the adaptive thresholding

techniques have been cited to be used as a

classification method.

Support Vector Machine (SVM) is the most

widely used classifier. Multiple comparative studies

like (Yuan, 2010), (Mohamed Bedeeuzzaman,

Farooq and Khan, 2010), and (Harikumar,

Vijaykumar and Palanisamy, 2011) shows that SVM

yielded one of the best with least amount of latency

alongside different and versatile feature extraction

method. Yuan showed that it works 0.44% more

accurate than Neural Networks.

In some particular conditions number of these

classifier has been cited to perform with 100%

accuracy. This accuracy is probably a result of over-

fitting as same techniques when applied on real life

data do not result with such high accuracy. So there

should be method which should keep these

algorithms improving their detection with increment

in the available examples. There should be a method

introduced where a neurologist can suggest

corrections as per his desire while observing wrong

marking by the computer aided system and the

system should learn from that correction.

3.5 Latency

To make these computer aided system useable in

real life condition, latency is important. There are

multiple reasons behind an increment of latency. It is

the training process which usually takes a lot of

time. Other then the computation involved during an

algorithm, latency also largely depends up on the

processing power, speed and size of the memory

devices in a machine which is running the algorithm.

Selection of appropriate Epoch size is not only

important for batter accuracy but it also affects the

latency. According to Seng et al. (Seng et al., 2013)

shortening the epoch size improves the accuracy on

the cost of latency.

Different algorithms result in different ways

when training examples are increased. Nasehi et al.

(Nasehi and Pourghassem 2011) reported that their

algorithms’ latency and detection delay got

decreased with the increment in training examples

with seizures. Geetha et al. (Geetha and

Geethalakshmi, 2011) and Abdullah et al.

(Abdullah, Saufiah and Ibrahim, 2012) describe that

after a certain amount of training examples,

increment in the training examples worsen the

accuracy and latency.

4 FUTURE RESEARCH

DIRECTIONS

In this section we will highlight the shortcomings of

the existing approaches and suggest some possible

future research directions.

Firstly, most of the existing work is about

detecting the generalized 3Hz spike & wave pattern

which is a symptom for absence seizure. To detect

localized epileptic activities, each channel should be

classified separately. This addition in the current

method will help in diagnosing focal epilepsy.

Secondly, many types of epilepsy are diagnosed

by EEG. These epilepsies are identified by unique

patterns or combination of few epileptic patterns.

3Hz spike & wave is just one of the many. Noachtar

et al. described ten types of epileptic patterns. To

counter this issue there should be exclusive and

independent trainers for each channel and each

channel should have exclusive and independent

trainer for each epileptic patterns. In this way we

BIOSIGNALS2014-InternationalConferenceonBio-inspiredSystemsandSignalProcessing

292

could detect, classify and label all types of epileptic

patterns.

Commercially available EEG analyzing software

tools are too user dependent. They need the user to

have a prior knowledge of signal processing to fully

exploit the full potential of this software. In this case

neurologist seeks the help of neurotechnicians,

which in case of inexperience or naive

neurotechnicians may lead to misinterpretation or

over-interpretation.

Seizure detection methods should be able to

improve their detection capability after initial

training. Addition of some post development

training mechanism can help the system improve its

performance over time. One way to handle this issue

is that after initial training and classification of the

EEG, a user interface showing the labelled epileptic

wave should allow the neurologist to mark an

epileptic chunk a.k.a “an epoch” as wrongly

classified. These markings will be saved in the

background as training example and they will be

later used to retrain the classifier alongside previous

examples. This addition will make the classifier

improve its detection with passage of time.

There are few papers that have cited that

increasing training data above a certain point can

drop down the accuracy, for this we suggest that

before classifier’s training feature reduction

techniques like PCA should be applied so that we

can remove the noise and redundancy without

compromising the important data.

In long term, this may make the system adapt

itself according to a neurologist’s corrective

marking, thus the system may start mimicking

neurologist detection by time regardless of any

merit. This is in accordance to our motivation i.e.

facilitating the neurologist. Another issue is that of

personalizing this software tool according to a

neurologist’s liking. Neurologists some time have a

disagreement with each other’s diagnostic markings.

So instead of forcing those to follow our brand of

detected patterns our suggested solution will resolve

both issues at a time. It will adapt its classification

ability as per every neurologist own desire.

(Noachtar & Rémi 2009)

5 CONCLUSIONS

Epilepsy is an important neurological disorder.

Computer assisted analysis of EEG for diagnosing

Epilepsy significantly helps a neurologist. To avoid

misinterpretation and over-interpretation a computer

assisted system should be user friendly, accurate,

robust and above informative. Lots of work has been

done in this regard. With the addition of our

suggested steps in the existing work robustness and

the classification accuracy can be improved. Other

than the approach we will like to suggest that there

should be a post development training system

attached to our-all detection algorithm so that it may

improve itself from the correction marked by the

neurologist. Though with the passage of time,

training system will try to adapt itself according to

the neurologist choice and its classification will get

biased as per his choice, but the whole point of this

effort was to supplement the neurologist.

REFERENCES

Abdullah, M. H., Abdullah, J. M. & Abdullah, M. Z. 2012,

'Seizure Detection by Means of Hidden Markov Model

and Stationary Wavelet Transform of

Electroencephalograph Signals', IEEE-EMBS

International Conference on Biomedical and Health

Informatics, IEEE, Hong Kong.

Abdullah, A. A., Saufiah, A. R. & Ibrahim, A 2012,

'Development of EEG-based Epileptic Detection using

Artificial Neural Network', International Conference

on Biomedical Engineering (ICoBE), IEEE, Penang.

Adelia, H, Zhoub, Z & Dadmehrc, N 2003, 'Analysis of

EEG records in an epileptic patient using wavelet

transform', Journal of Neuroscience Methods, vol 123,

no. 1, pp. 69–87.

Alam, S. M. S. & Bhuiyan, M. I. H. 2013, 'Detection of

Seizure and Epilepsy Using Higher Order Statistics in

the EMD Domain', IEEE Journal of Biomedical and

Health Informatics, vol 17, no. 2, pp. 312-318.

Benbadis, S.R. & Tatum, WO 2003, 'Overinterpretation of

EEGs and misdiagnosis of epilepsy', Neurophysiol, vol

20, pp. 42-44.

Brain Products GmbH / Products & Applications /

Analyzer 2, viewed 17 May 2013, <http://

www.brainproducts.com/productdetails.php?id=17>.

Chang, H. Y., Yang, S. C., Lan, S. H. & Chung, P. C.

2010, 'Epileptic seizure detection in grouped multi-

channel EEG signal using ICA and wavelet transform',

International Symposium on Circuits and Systems

(ISCAS), IEEE, Paris.

Choi, K. S., Zeng, Y & Qin, J 2012, 'Using sequential

floating forward selection algorithm to detect epileptic

seizure in EEG signals', International Conference on

Signal Processing (ICSP), IEEE, Beijing.

Geetha, G & Geethalakshmi, S. N. 2011, 'Detecting

Epileptic Seizures Using Electroencephalogram: A

New and Optimized Method for Seizure Classification

Using Hybrid Extreme Learning Machine',

International Conference on Process Automation,

Control and Computing (PACC), IEEE, Coimbatore.

Goldberger, A. L, Amaral, L, Glass, L, Hausdorff, J. M.,

Ivanov, P. C., Mark, R. G, Mietus, J. E, Moody, G. B,

AdvancementsinComputerAidedMethodsforEEG-basedEpilepticDetection

293

Peng, C. K & Stanley, H. E 2000, 'PhysioBank,

PhysioToolkit, and PhysioNet: Components of a New

Research Resource for Complex Physiologic Signals',

Circulation , vol 23, no. 101, pp. 215-220, viewed 11

November 2013, <http://physionet.org/pn6/chbmit/>.

Guo, P, Wang, J, Gao, XZ & Tanskanen, JMA 2012,

'Epileptic EEG Signal Classification with Marching

Pursuit based on Harmony Search Method',

International Conference on Systems, Man, and

Cybernetics, IEEE, Seoul.

Harikumar, R, Vijaykumar, T & Palanisamy, C 2011,

'Performance analysis of fuzzy techniques hierarchical

aggregation functions decision trees and Support

Vector Machine (SVM)for the classification of

epilepsy risk levels from EEG signals', Recent

Advances in Intelligent Computational Systems

(RAICS), IEEE, Trivandrum.

Kaleem, M, Guergachi, A & Krishnan, S 2013, 'EEG

Seizure Detection and Epilepsy Diagnosis using a

Novel Variation of Empirical Mode Decomposition',

International Conference of the EMBS, IEEE, Osaka.

Khan, Y. U., Rafiuddin, N & Farooq, O 2012, 'Automated

seizure detection in scalp EEG using multiple wavelet

scales', IEEE International Conference on Signal

Processing, Computing and Control (ISPCC), IEEE,

Waknaghat Solan.

Klinik fur Epileptologie, Universitat Bonn, viewed 11

November 2013, <http://epileptologie-bonn.de/cms/

front_content.php?idcat=193&lang=3&changelang=3>.

Lüders, H. O. & Noachar, S 2000, Atlas and classification

of electroencephalography, Elsevier Science Health

Science Division.

Luo, K & Luo, D 2010, 'An EEG Feature-based Diagnosis

Model for Epilepsy', International Conforence on

Computer Application and System Modeling

(ICCASM), IEEE, Taiyuan.

Mohamed Bedeeuzzaman, V, Farooq, O & Khan, YU

2010, 'Automatic Seizure Detection Using Higher

Order Moments', International Conference on Recent

Trends in Information, Telecommunication and

Computing, IEEE, Kochi, Kerala.

Muthanantha Murugavel, AS, Ramakrishnan, S,

Maheswad, U & Sabetha, BS 2013, 'Combined

Seizure Index with Adaptive Multi-Class SVM for

epileptic EEG classification', International Conference

on merging Trends in VLSI, Embedded System, Nano

Electronics and Telecommunication System

(ICEVENT), IEEE, Tiruvannamalai.

Nasehi, S & Pourghassem, H 2011, 'Epileptic Seizure

Onset Detection Algorithm Using Dynamic Cascade

Feed-Forward Neural Networks', International

Conference on Intelligent Computation and Bio-

Medical Instrumentation (ICBMI), IEEE, Wuhan,

Hubei.

Neuralynx ~ Spike Sort 3D Software, viewed 20 May

2013, <http://neuralynx.com/

research_software/spike_sort_3d>.

NeuroExplorer Home, viewed 5 May 2013,

<http://www.neuroexplorer.com/>.

Noachtar, S, Binnie, C, Ebersole, J, Mauguière, F,

Sakamoto, A & Westmoreland, B 1999, 'The

International Federation of Clinical Neurophysiology:

a glossary of terms most commonly used by clinical

electroencephalographers and proposal for the report

form for the EEG findings', Clinical Neurophysiology,

vol 52, pp. 21-41.

Noachtar, S & Rémi, J 2009, 'The role of EEG in epilepsy:

A critical review', Epilepsy & Behavior

, pp. 22-33.

Ocbagabir, H. T., Aboalayon, K. A. I & Faezipour, M

2013, 'Efficient EEG analysis for seizure monitoring

in epileptic patients ', Long Island Systems,

Applications and Technology Conference (LISAT),

IEEE, Farmingdale, New York.

Petersen, E. B., Henriksen, J. D., Mazzarettoa, A, Kjær,

TW, Thomsen, C. E. & Sorensen, H. B. D. 2011,

'Generic Single-Channel Detection of Absence

Seizures', Annual International Conference of the

EMBS, IEEE, Boston.

Seng, C. H, Demirli, R, Khuon, L & Bolger, D 2013,

'Seizure Detection in EEG Signals Using Support

Vector Machines', Northeast Bioengineering

Conference (NEBEC), IEEE, Philadelphia.

Sezer, E, Işik, H & Saracoğlu, E 2012, 'Employment and

Comparison of Different Artificial Neural Networks

for Epilepsy Diagnosis from EEG Signals', Journal of

Medical Systems, vol 36, no. 1, pp. 347-362.

Shoeb, A 2000, CHB-MIT Scalp EEG Database, viewed

11 November 2013, <http://physionet.org/pn6/chbmit/>.

Sousa, T, Mendes, P & Ribeiro, J 2012, 'The impact of

rhythms analysis technique on electrographic seizure

detection (EEG)', 2nd Portuguese Meeting in

Bioengineering (ENBENG), IEEE, Coimbra.

Tucker-Davis Technologies, viewed 20 May 2013,

<http://www.tdt.com/index.htm>.

Wang, J, Gao, XZ, Tanskanen, JMA & Guo, P 2012,

'Epileptic EEG Signal Classification with ANFIS

based on Harmony Search Method', International

Conference on Computational Intelligence and

Security (CIS), IEEE, Guangzhou.

WHO | Epilepsy 2012, <http://www.who.int/mediacentre/

factsheets/fs999/en/>.

Xanthopoulos, P, Rebennack, S, Liu, C. C, Zhang, J,

Holmes, G. L., Uthman, B. M. & Pardalos, P. M.

2010, 'A Novel Wavelet Based Algorithm for Spike

and Wave Detection in Absence Epilepsy',

International Conference on BioInformatics and

BioEngineering (BIBE), IEEE, Philadelphia.

Yuan, Y 2010, 'Detection of Epileptic Seizure Based on

EEG Signals', International Congress on Image and

Signal Processing (CISP2010), IEEE, Yantai.

BIOSIGNALS2014-InternationalConferenceonBio-inspiredSystemsandSignalProcessing

294