Classify and Compare Using S-SVM and LS-SVM for EMD Based ... · 946 M.Dhivya and Ms.I.Manju I....

International Journal of Electronics Engineering Research.

ISSN 0975-6450 Volume 9, Number 7 (2017) pp. 945-956

© Research India Publications

http://www.ripublication.com

Classify and Compare Using S-SVM and LS-SVM for

EMD Based Feature Extraction of EEG Signal

M.Dhivya Ms. I.Manju

M.E, Applied Electronics Senior Grade Assistant Professor

Velammal Engineering College Velammal Engineering College

Abstract

Disease identification is major task in the field of biomedical. This research

paper presents a feature extraction from Electroencephalogram (EEG) signals

using empirical mode decomposition (EMD). It discriminate the EEG signals

corresponding to healthy persons and epileptic patients during seizure - free

intervals and seizure attacks. It gives an effective time-frequency analysis of

non-stationary signals. The intrinsic mode functions (IMF) obtained by the

result of EMD give the decomposition of a signal according to its frequency

components. This project presents the usage of temporal statistics, and

spectral features including spectral centroid , coefficient of variation and the

spectral skew of the IMFs for feature extraction from EEG signals. These

features extraction is relevant to find out the normal and pathological EEG

signal. The normal EEG signals have different temporal and spectral centroids

, dispersions and symmetries when compared with the pathological EEG

signals. The structured support vector machine (S-SVM) and least square

support vector machine (LS- SVM) are used for the classification purposes. In

this paper both this two classifier used to classify the EEG signals , Finally

compare the performance and accuracy of both this two classifier, determine

whether the S-SVM is more suitable for the EEG signals classification .

Index terms: EEG signal, Epilepsy ,Empirical mode decomposition, feature

extraction, classification.

946 M.Dhivya and Ms.I.Manju

I. INTRODUCTION

Electroencephalography (EEG) is a method to record electrical activity of the brain.

The EEG signals can be effectively used for various applications such as emotion

recognition , brain computer interfaces (BCI) etc. One of the most important

applications of the analysis of EEG signals is its use in neuroscience to diagnose

diseases and brain disorders. It is typically noninvasive, with the electrodes placed in

the scalp, although invasive electrodes are sometimes used in specific applications.

EEG measures voltage fluctuations. Epilepsy is a world common neurological

disorder in human beings. It is also known as epileptic fit which create sign or

symptoms due to abnormal excessive or synchronous neuronal activity in the brain

causes the loss of consciousness or a whole body convulsion. The outward effect can

vary from uncontrolled jerking movement(tonic-clonic seizure) to as subtle as a

momentary loss of awareness(absence seizure).Diseases of the brain characterized by

an enduring predisposition to generate epileptic seizures are collectively called

epilepsy. Seizures can also occur in people who do not have epilepsy for various

reasons including brain trauma, drug use, elevated body temperature, low blood sugar

and low levels of oxygen. EEG represents a signal containing information about the

condition of the brain. Patients are unaware of seizure due to the random nature of

them which may increase the risk of physical injury. The algorithm required for

automated seizure detection and prediction employs feature computation and

subsequent classification. Electroencephalogram is an important tools for diagnosis

and analysis of epilepsy. It represents the electrical activity produced by firing of

neuron within the brain. Epileptic seizures can make different changes in perception

and behavior temporarily. In the human EEG, they are throw back by ictal sequence,

where epileptic seizure becomes real as quality, rhythmic signals frequently

coinciding with preceding before the usual observable changes in behavior .since the

EEG data include the transient signals in noise and non stationary signals, where

nonlinear time-series analysis should be carried out with caution. Its detection is

typically done by the physicians using a visual scanning of the EEG signals which is a

time consuming process and may be inaccurate. These inaccuracies are particularly

significant for long time duration EEG signals.

The parameters extracted from the EEG signals using various signal processing

methods are very useful for diagnostics. The spectral parameters based on the Fourier

transform are useful for analyzing the EEG signals and have shown good results on

their classification. However, it is important to note that the Fourier domain does not

exhibit any time-domain characteristics in the signal giving the features which are

sub-optimal for feature extraction from some signal processing scenarios . Several

other methods based on time-frequency domain have been developed for the detection

of epileptic seizures from EEG signals. These methods include the use of short time

Fourier transform (STFT) . Although good results are obtained using these methods,

Classify and Compare Using S-SVM and LS-SVM for EMD Based Feature… 947

the STFT does not yield a multiresolution analysis of the signals. This is because of

the fact that the STFT uses the filters of the same bandwidth for signal decomposition

at all frequencies. This limitation is typically resolved using the wavelet analysis in

which a multiresolution time-frequency analysis is facilitated by forming band pass

filters with varying bandwidths . Researchers have found the wavelet analysis to be a

very useful tool for various signal processing applications

In this paper the wavelet transform is used to solve multiresolution problem. In this

analysis multiresolution time-frequency analysis is facilitated by forming band pass

filters with varying bandwidths. The artifact in the EEG signals are removed using a

wavelet ICA(Independent Component Analysis) based method giving good results on

suppression of artifacts in EEG signals. EEG signals , which shown that their

frequency components change over a period of time making them non-stationary.

Hence, the signal processing methods which are more suitable for such signals are

desired.

Recently, new techniques for the analysis of non-stationary and non-linear signals

have been proposed which are mainly based on empirical mode decomposition

(EMD).The EMD is a time-frequency based method to decompose a signal into a

number of intrinsic mode functions(IMFs)which are the oscillatory components. The

EMD is effective for the time-frequency analysis of the non-stationary signals.

In this paper, the classification of EEG signals involving three stages. In the first

stage, the EMD is used to decompose the signal into the number of IMFs. the IMFs is

nothing but the oscillatory component. The second stage involves in the first three

number of IMFs are used for the feature extraction .in the feature extraction method

both the third order temporal statistics and spectral statistics are involved. In the third

stage, the calculated features are finally applied to both structured support vector

machine (S-SVM) and least square support vector machine(LS-SVM),which is used

to classify the normal and pathological EEG signal. After the comparition of both two

classifiers and determine that the S-SVM is more suitable for the classification of eeg

signals.

Fig 1. Block diagram of proposed system


II. DATASET

In this study, we have used an EEG dataset that is publicly available online The

dataset consists of three subsets(denoted as sets A -C) each containing 50 single

channel EEG signals, each one having a duration of 23.6 seconds. signals have been

selected from continuous multichannel EEG recording after visual inspection of

artifacts. The Sets A and B consist of surface EEG segments collected from five

healthy volunteers in awaken and relaxed state with their eyes opened and closed

respectively.Set C contains signals corresponding to seizure attacks (i.e., ictal EEG),

recorded using all the electrodes. The signals are recorded in a digital format at a

sampling rate of 173.61 Hz. Thus, the sample length of each segment is

173.61×23.6≈4097.

Figure 2: Sample EEG signals from three different sets from rows 1to 3(A, B and C

respectively)

II. METHODS

A. Empirical Mode Decomposition – EMD

The EMD is a method of decomposing a signal without leaving a time domain. It can

be compared to other analysis methods like fourier transforms and wavelet

decomposition. This process is useful for analyzing natural signals,which are most

often non-linear and non-stationary . EMD filters out functions which form a

complete and nearly orthogonal basis for the original signal. completeness is based

on the method of the EMD, the way it is decomposed implies completeness. This

functions, known as Intrinsic Mode Functions(IMFs),are therefore sufficient to

describe the signal, even though they are not necessarily orthogonal. consider x(t) is

a given EEG signal ,the calculation of its IMFs involves the following steps.

1) Identifying all the extrema (maxima and minima) in x(t).

2) Interpolate between minima and maxima generating the envelopes el (t) and em

(t).


3) Determine the local mean as

𝑎(𝑡) =𝑒𝑚 (𝑡) + 𝑒1 (𝑡)

2.

4) Extract the detail i.e h1(t) –x(t) a(t).

5) Decide whether h1(t) is an IMF or not based on two basic conditions for IMFs

mentioned above.

6) Repeat step 1 to 4 until An IMF is obtained.

After the first IMFs is obtained, then we have to define the c1(t) – h1(t),which is

the smallest temporal scale in x(t).A residual signal obtained as r1(t) – x1(t) c1

(t).

At the end of the decomposition, the original signal can be represented as follows

𝑥(𝑡) − ∑ 𝑐𝑚 (𝑡) + 𝑟𝑀 (𝑡)

𝑀

𝑚0 1

Where M is the number of IMFs, cm(t) is the mth IMFs and rM(t) is the final residue.

B. Analytic representation of IMFs:

After the extraction of IMFs is done from the EEG signals, their analytic

representation is obtained. This representation the DC offset from the spectral

component of the signals, Which is the aspect to compensate for the non-stationary of

the signals.

Then the IMFs cm(t) is obtained and the analytic representation is

𝑦(𝑡) = 𝑐𝑚 (𝑡) + 𝑖𝐻{𝑐𝑚(𝑡)}

where H{cm(t)} is the Hilbert transform of cm(t), which is the mth IMF extracted

from the signal x(t). After find the EMD of the signal, the IMFs are used for feature

extraction purposes.

C. Temporal statistics of analytic IMFs:

The statistical features of the IMFs are useful for discriminating between normal and

pathological EEG signals. The distribution of samples in the data are characterized by

their asymmetry, dispersion and concentration around the mean. In the IMFs the

visual analysis is done from healthy and epilepsy patients during interictal and ictal

periods after Hilbert transform. By using HT, the difference between appropriately

captured using the statistics of the IMFs. For an IMF, these statistics can be obtained

by


𝜇 𝑡 =1

𝑁 ∑ 𝑦𝑖

𝑁

𝑖=1

𝜎𝑡=√1

𝑁 ∑ (𝑦𝑖−

𝑁𝑡=1 𝜇𝑡 )2

𝛽𝑡 =1

𝑁∑(

𝑦𝑖 − 𝜇𝑡

𝜎𝑡

𝑁

𝑖=1

)3

Where N is the number of samples in the IMF µt is

the mean, ơt is the variance and t𝛽𝑡 is skewness of the corresponding IMF.

D. Spectral statistics of analytic IMFs:

EMD has the features to perform, a spectral analysis of the signals. A frequency based

analysis can therefore be useful for feature extraction from EEG signals. The EMD

helps to decompose a signal into number of components(IMFs) which are response to

filters having narrow pass bands. The spectral analysis is done using the calculation of

instantaneous frequencies (IF).The calculation of IF has the physical meaning only for

monocomponent signals. The discrimination power of the PSD features can be

analysed by their respective plots for three IMFs from the no

𝑝(𝑤) = ∑ 𝑟𝑦 [𝑛]𝑒−𝑗𝑤𝑛

𝛼

−𝛼

where ry[n] represents the autocorrelation of y[n], defined as ry[n] = E(y[m]y_[m]).

Visual analysis of the PSD of IMFs shows that the statistics of the PSD can be used as

relevant features for feature extraction.

1. Spectral centroid

The researchers have shown that thecentroid frequencies of the IMFs extracted from

EEG signalsform distinct groups when supervised clustering is applied onthe EEG

signals . These respective groups are indicativeof the seizure and non-seizure EEG

signals. The centroidfrequency is therefore a distinctive feature that can be usedfor the

characterization of EEG signals.

The discrimination power of thePSD features can be visually analysed by their

respective plots for three IMFs from the normal and pathological EEG signals. The

PSD can be calculated as follows:


𝑐𝑠 =∑ 𝑤 𝑤𝑝(𝑤)

∑ 𝑤 𝑝(𝑤)

where P (w) is the amplitude of wth frequency bin in thespectrum.

2. Variation coefficient:

Since the spectral variation in theIMFs is different for normal and pathological EEG

signals,therefore it can be used for their characterization. This variation can be

calculated as follows:

𝜎2𝑠=

∑ 𝑤(𝑤−𝑐𝑠)2𝑝(𝑤)∑ 𝑤𝑝(𝑤)

where Cs is the spectral centroid.

3. Spectral skew:

Skewness is the third order moment andit measures the symmetry/asymmetry of a

distribution. EEG signals differs thus potentially yielding a useful featurefor the

classification of EEG signals. Skewness of the PSDcan be calculated as:

𝜷𝒏=

∑ 𝒘(𝒘−𝒄

𝝈 )𝟐𝒑(𝒘)

∑ 𝒘 𝒑(𝒘)

After the extraction of temporal and spectral features of each IMF, its feature vector

can be obtained by

𝐹 = [ 𝜇𝑡 𝜎𝑡 𝛽 𝑡 𝑐𝑠 𝜎𝑠 𝛽𝑠 ]

The feature vectors obtained from several IMFs can than be used for classification

purposes.

E. Classification

Feature extraction is followed by the classification of EEG signals using S-SVM and

LS-SVM. The S-SVM is the generalization of SVM .whereas SVM depends upon the

binary classification, multiclass classification and regression,while the S-SVM allows

training of a classifier for general structured output labels. It depends upon the correct

and previous output samples.Least sqaures support vector machine is the set of related

supervised learning method that helps to analyze data and recognize patterns,which is

used for the classification and regression analysis.it is the class of kernal based

learning methods.kernel methods are also called as kernal functions. This function


determine the similarity in the input traning data and split out the training data by

using hyperplane.

III. EXPERIMENTAL RESULTS

The performance of the proposed methodology for feature extraction from EEG

signals is studied using standard measures such as overall accuracy and area under

receiver operating characteristics (ROC) curve.

A. Performance and analysis of proposed system.

(a) (b)

(c) (d)

(e)

Fig 2 (a) Original input signal (b)calculated IMFs from the original signal (c) power

spectral density (d) ROC Curve for S-SVM (e) Receiver Operating Characteristic

curve for LS-SVM.


In this paper 50 signals are tested for calculating the accuracy. The accuracy of a test

is its ability to differentiate the normal and epilepsy signal correctly.to estimate the

accuracy of the test,we should calculate the proportion of true positive and true

negative in all evaluated cases. True positive and true negative is also called as

sensitivity and specificity.

Mathematically, this can be started as

ACCURACY = TP+TN/TN+TP+FP+FN

SIGNAL ACCURACY FOR S-SVM ACCURACY FOR LS-SVM

Normal signal

88

80

Abnormal signal

100

94

Total accuracy

94%

87%

IV. DISCUSSON AND CONCLUSION

This project paper presented an EEG data classification algorithm , which based on a

large number of feature extraction after wavelet transform .The foundation of this

method lies on the extraction of temporal and spectral features from Empirical Mode

Decomposition (EMD) of the EEG signals. The usage of EMD is motivated by the

fact that EEG signals are non-stationary and EMD is a data dependent method

exhibiting a better adaptability towards non-stationary in the EEG signals. The main

advantage of the algorithm are (a) the ability of the algorithm to run robustly in a

clinical setting with noised EEG;(b) feature extraction with highly meaningful

wavelet transform because hidden EEG information can be revealed and the noise

effort reduced as certain data under some scales are omitted;(c) simplicity and low

computational cost guaranteeing real clinical application;(d)very good sensitivity and

specificity. Both the S-SVM and LS-SVM are used for the classification of EEG

signals but according to performance and accuracy S-SVM is more suitable for EEG

than LS-SVM . S-SVM helps to find the normal and abnormal signals in a accurate

way.


REFERENCES

[1] Farhan Riaz1, Ali Hassan, Saad Rehman, Imran Khan Niazi and Kim

Dremstrup, “EMD based Temporal and Spectral Features for the Classification

of EEG Signals Using Supervised Learning” IEEE Transactions on Neural

Systems and Rehabilitation Engineering, 1534-4320 (c) 2015 IEEE.

[2] Mahmoud E. A. Abdel-Hadi, Reda A. El-Khoribi, M. I. Shoman, and M. M.

Refaey “Classification of motor imagery tasks with LS-SVM inEEG-based

self-paced BCI” 978-1-4673-6832-2©2015 IEEE

[3] N. Mak and J. R. Wolpaw, “Clinical applications of brain-computer interfaces:

current state and future prospects,” IEEE Rev Biomed Engg, vol. 2, pp.187-

199, 2009.

[4] H. Fukuyama, Y. Ouchi, S. Matsuzaki, Y. Nagahama, H. Yamauchi, M.

Ogawa, J. Kimura, and H. Shibasaki, “Brain functional activity during gait in

normal subjects: a SPECT study,” Neuroscience letters, vol. 228(3), pp.183-

186, 1997.

[5] V. Dietz, “Spinal cord pattern generators for locomotion,” Clinical Neurophys,

vol. 114(8), pp. 1379-1389, 2003.

[6] T. Hanakawa, “Neuroimaging of standing and walking: special emphasis on

Parkinsonian gait ,” Parkinsonism and Related Disorders, vol. 12(2), pp.S70-

S75, 2006

[7] M. Teplan, “Fundamentals of EEG measurement,” Measurement Sc. Rev, vol.

2, pp. 1-11, 2002.

[8] Morteza Behnam, Hossein Pourghassem ,”Singular Lorenz Measures Method

for Seizure Detection using KNN-Scatter Search Optimization Algorithm,

978-1-5090-0139-2/15/$31.00 ©2015 IEEE

[9] W. Jia, N. Kong, F. Li, X. Gao, S. Gao, G. Zhang, Y. Wang, and F.Yang, “An

epileptic seizure prediction algorithm based on second-order complexity

measure,” Physiol. Meas, vol. 26, pp.609-625, 2005.

[10] V. M. Bedeeuzzaman, O. Farooq, and Y. U. Khan, “Automatic Seizure

Detection Using Higher Order Moments,” in Proc. of Intl. Conf. on Recent

Trends in Information, Telecomm and Computing, pp. 159-163,

[11] A. T. Tzallas, M. G. Tsipouras, and D. I. Fotiadis, “Epileptic Seizure

Detection in EEGs Using Time-Frequency Analysis,” IEEE Trans. On

Information Technology in Biomedicine, vol. 13, no. 5, pp. 703-710,2009.

[12] F. Takens, “Detecting strange attractors in turbulence”, Dynamical Systems


and Turbulence, Lecture notes in Math., Springer-Verlag, Heidelburg, pp.

366-381, 1981. Classification of Lower Limb Motor Imagery Using

[13] Sanniv Bhaduri1,a, Anwesha Khasnobish2,b, Rohit Bose1,c, D. N.

Tibarewala2,d K Nearest Neighbor and Naive-Bayesian Classifier, 3rd Int’l

Conf. on Recent Advances in Information Technology, 978-1-4799-8579-

1/16/$31.00 ©2016 IEEE

[14] A. Wolf, J. B. Swift, H. L. Swinney, and J. A. Vastano, “Determining

Lyapunov Exponents from a time series,” Physical D, vol. 16, pp.285-317,

1985.

[15] P. Grassberger and I. Procaccia, “Characterization of strangeattractors,” Phys.

Rev. Lett, no. 50, pp. 346-349, 1983.

[16] F. Takens, “Numerical determination of the dimension of an

attractor”,Dynamical Systems and bifurcations, Lecture notes in Math.

Springer,Berlin, vol. 1125, pp. 99-106, 1985.

[17] V. Srinivasan, C. Eswaran, and N. Sriraam, “Artificial neural network based

epileptic detection using time-domain and frequency-domain features,” J.

Med. Syst., vol. 29, no.6, pp. 647–660, 2005.

[18] K. Polat and S. Günes, “Classification of epileptiform EEG using a hybrid

system based on decision tree classifier and fast Fourier transform,” Appl.

Math. Comput, vol. 187, no.2, pp. 1017–1026, 2007

[19] N. Mammone, F. La Foresta, and F. C. Morabito,“Automatic artifact rejection

from multichannel scalp EEG by wavelet ICA,” IEEE Sensors J., vol. 12, no.

3, pp. 533–542, Mar. 2012.

[20] R. B. Pachori and P. Sircar, “EEG signal analysis using FB expansion and

second-order linear TVAR process,” SignalProcess, vol. 88, no. 2, pp. 415–

420, 2008.

[21] S. F. Liang, H. C. Wang, and W. L. Chang, “Combinationof EEG complexity

and spectral analysis for epilepsy diagnosis and seizure detection,” EURASIP

J. Adv. Signal Process, vol. 2010, p. 853434, 2010.

[22] Meier R, Dittrich H, Schulze-Bonhage A and Aertsen A (2008), “Detecting

epileptic seizures in long-term human EEG: A new approach to automatic

online and real-time detection and classification of polymorphic seizure

patterns”, Clin. Neurophysiol., vol. 25, no. 3, pp. 119–131.

[23] Marwan N, Romano M.C, Thiel M and Kurths J (2007), “Recurrence plots for

the analysis of complex systems”, Phys. Rep., vol. 438.


[24] Andrzejak R, Lehnertz K, Mormann F, Rieke C, David P and Elger C (2001),

“Indications of nonlinear deterministic and finite-dimensional structures in

time series of brain electrical activity: Dependence on recording region and

brain state”, Phys. Rev. E, vol. 64, no. 6, pp.1–8.

[25] Durka P.J (2003), “From wavelets to adaptive approximations: Time-

frequency parameterization of EEG”, BioMed. Eng. OnLine, vol. 2, no. 1, pp.

1(1)–1(30).

[26] Elham Hosseini and Abolfazl Falahati (2013), “Improving Water-Filling

Algorithm to Power Control Cognitive Radio System Based Up on Traffic

Parameters and QoS”, Proceeding of Wireless Pers Commun, DOI

10.1007/s11277 -012-0778-0.

Classify and Compare Using S-SVM and LS-SVM for EMD Based ... · 946 M.Dhivya and Ms.I.Manju I....

Documents

Transcript of Classify and Compare Using S-SVM and LS-SVM for EMD Based ... · 946 M.Dhivya and Ms.I.Manju I....