You are on page 1of 27

Accepted Manuscript

Features based on analytic IMF for classifying motor imagery EEG signals in
BCI applications

Sachin Taran, Varun Bajaj, Dheeraj Sharma, Siuly Siuly, A. Sengur

PII: S0263-2241(17)30702-9
DOI: https://doi.org/10.1016/j.measurement.2017.10.067
Reference: MEASUR 5069

To appear in: Measurement

Received Date: 12 January 2017


Revised Date: 29 September 2017
Accepted Date: 31 October 2017

Please cite this article as: S. Taran, V. Bajaj, D. Sharma, S. Siuly, A. Sengur, Features based on analytic IMF for
classifying motor imagery EEG signals in BCI applications, Measurement (2017), doi: https://doi.org/10.1016/
j.measurement.2017.10.067

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers
we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and
review of the resulting proof before it is published in its final form. Please note that during the production process
errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
Features based on analytic IMF for classifying motor
imagery EEG signals in BCI applications

Sachin Tarana , Varun Bajaja , Dheeraj Sharmaa , Siuly Siulyb , A. Sengurc


a Discipline of Electronics and Communication Engineering, PDPM Indian Institute of
Information Technology, Design and Manufacturing Jabalpur, Jabalpur 452005, India
b Centre for Applied Informatics, College of Engineering and Science, Victoria University,

Melbourne, Australia
c Firat University, Technology Faculty, Electrical and Electronics Engineering Dept., Elazig,

TURKEY

Abstract

Brain-computer interface (BCI) system works as a reliable support system for


disabled people to communicate with real world. The augmentation in reliabil-
ity of BCI systems is possible by successful classification of different motor im-
agery (MI) tasks. In this work, the analytic intrinsic mode functions (AIMFs)
based features are proposed for classification of electroencephalogram (EEG)
signals of different MI tasks. The AIMFs are obtained by applying empirical
mode decomposition (EMD) and Hilbert transform on EEG signal. The fea-
tures namely: raw moment of first derivative of instantaneous frequency, area,
spectral moment of power spectral density, and peak value of PSD are com-
puted from AIMFs. The features are normalized to reduce the biased nature
of the classifier. The normalized features are applied as inputs to least squares
support vector machine (LS-SVM) classifier and performance parameters are
computed using different kernel functions of LS-SVM classifier. The radial basis
kernel function for IMF1 provides better MI task classification accuracy 97.56%,
sensitivity 96.45%, specificity 98.96%, positive predicted value 99.2%, negative
predictive value 95.2%, and minimum error rate detection 4.28%. The propose

∗ Corresponding author
Email address: sachin.taran@iiitdmj.ac.in, bajajvarun056@yahoo.co.in,
dheeraj@iiitdmj.ac.in, siuly.siuly@vu.edu.au, ksengur@gmail.com (Sachin Tarana ,
Varun Bajaja , Dheeraj Sharmaa , Siuly Siulyb , A. Sengurc )

Preprint submitted to Journal of LATEX Templates November 1, 2017


method shows better performance as compared to state-of-the-art methods.
Keywords: Electroencephalogram (EEG) signal, motor imagery (MI) tasks,
empirical mode decomposition, least squares support vector machine

1. Introduction

The brain-computer interface (BCI) is a system in which sequence of thoughts


initiated from the brain and used in form of commands to control specific equip-
ment. This imagining thought processes of a specific task in brain is known as
5 motor imagery (MI) task. Such type of MI-tasks based BCI systems find ap-
plications for disabled people to communicate with real world [1]. The utility
of BCI systems is relied on how accurately, the different MI-tasks are classified.
The use of electroencephalogram (EEG) signals for MI tasks classification is
most apt choice because of following reasons: EEG-based approaches are non-
10 invasive in nature [2, 3], EEG signals show good correlation with mental activity
and physiology of human brain for identification of sleep stages and epileptic
seizure [4, 5, 6], and EEG signals provide more practical way for real-time BCI
application such as muscular communication [7].
Due to ease of use and reliability of MI-based BCI systems for disabled peo-
15 ple, lots of research methodologies are proposed in this area. The methods,
which are most successful for separating MI tasks in BCI systems are based
on some kind of feature extraction and classification [8]. The sparsity-based
spatial filter is used to reduce the number of electrodes and variance of spa-
tial filter projected signal considered as input features to linear discriminant
20 analysis algorithm for classification [9]. The composite feature extraction and
classification approach, based on iterative spatiospectral patterns learning algo-
rithm is proposed for identification of MI tasks [10]. In [11], Bayesian method is
used for features extraction and spectrally weighted decision rule proposed for
classification. An improved version of common spatial-subspace decomposition
25 is used as feature extraction method and SVM is tested for classification [12]. In
[13], clustering technique is used for feature extraction and LS-SVM is employed

2
for classification. The multivariate decomposition and short time Fourier trans-
form (FT) analyzes the EEG signals and magnitude spectrum based features
classify the MI tasks EEG signals [14]. The Dual tree complex WT decomposed
30 EEG signals into different bands of real and imaginary coefficients. Energy of
respective bands coefficients are considered as features and tested on several
classification algorithms to classify left and right hand MI tasks EEG signals
[15].
The wavelet packet decomposition decomposed de-noising EEG signal into
35 sub-bands, which are used in the extraction of higher order statistical features
[16]. These features are fed as inputs to K-nearest neighbour classifier and
results 92.8% classification accuracy [16]. An optimum allocation based statis-
tical approach is applied for reduction in length of EEG sequences and these
sequences are used in the extraction of statistical features [17]. The extracted
40 features are applied as inputs to naive base classifier, which results in 96.3% ac-
curacy and 2.32% standard deviation (SD). The common spatial pattern (CSP)
method is used for feature extraction and z-score linear discriminant analy-
sis (Z-LDA) classifier provides the 81.1% classification accuracy and 18.2% SD
[18]. The cross-correlation based approach is proposed for extraction of statis-
45 tical features, which are fed as inputs to LS-SVM classifier and the outcome
of methodology came up in form of 95.72% accuracy and 4.35% SD [19]. The
regularized-CSP with aggregation is proposed for classification of MI based EEG
signals, which provides 83.9% classification accuracy and 10.36% SD [20]. As
observed from the literature survey, no one has been applied analytic intrinsic
50 mode functions (AIMFs) based feature extraction method for classification of
MI tasks and BCI system can be improved for identification of different MI
tasks.
In this work, AIMFs based features are proposed for automatic classification
of MI tasks dependent EEG signals. The multichannel recorded EEG signals
55 show non-linear and non-stationary nature. The complexity of such EEG sig-
nals is explored by AIMFs-based features. The AIMFs from raw EEG signal are
obtained by applying empirical mode decomposition (EMD) and Hilbert trans-

3
form (HT). The features based on AIMFs are: raw moment of the first derivative
of instantaneous frequency (RMFDIF), area, spectral moment of power spec-
60 tral density (SMPSD), and peak value of PSD. Further, the unit length feature
normalization method is applied on extracted features to performed unbiased
classification. The normalized features are applied as inputs to LS-SVM classi-
fier to evaluate the MI-tasks classification performance of proposed method.
The sequence of remaining paper is as follows. The proposed methodology
65 described in Section 2, under the following subsections: data set, decomposition
technique, AIMF based features extraction, feature normalization, and classi-
fication. Section 3 presents the results of proposed method and discussion is
given in Section 4. The conclusion of paper is given in Section 5.

2. Methodology

70 The EEG signal flow of proposed methodology is presented by block diagram


in Figure 1.

2.1. Dataset

The proposed methodology validated over BCI competition-III, IVa EEG


data-set [21]. This dataset was recorded from five healthy subjects (labelled aa,
75 al, av, aw, ay) who performed right hand (RH) (class1) and right foot (RF)
(class2) MI tasks. The original EEG recording is done at 1000 Hz sampling
frequency and later it is down sampled at 100 Hz. In this experiment, EEG
signals of first 10 channels are used for evaluation of proposed method. The
detailed description of used dataset is available in references [20, 21].

80 2.2. Empirical mode decomposition (EMD) and Hilbert transform (HT)

The EMD decomposes non-stationary EEG signal into stationary basis func-
tions which are called as intrinsic mode functions (IMFs) [22]. These IMFs have
to satisfy two basic characteristics. (i) In the complete data set, the count of
maximum and minimum values must be equal or differ by at most one from the
85 number of baseline crossing. (ii) The mean value of envelopes defined by local

4
Figure 1: EEG based MI tasks classification process.

maxima and minima are zero. First condition improves, the stationary char-
acteristics and second improves, the definition of instantaneous frequency (IF).
After applying EMD, the non-stationary EEG signal x[n] can be decomposed
into M IMFs as:
M
X
x[n] = cm [n] + rM [n] (1)
m=1

90 where, cm [n] is the mth IMF and rM [n] is final residue. The RH MI task EEG
signal and its IMFs are shown in Figure 2.
Hilbert transform (HT) provides the analytical representation of IMF, which
is an extension of real-valued IMF to complex valued IMF. The analytic IMF
can be represented as [22]:

z[n] = c[n] + jcH [n] = A[n]ejφ[n] (2)

95 where, cH [n] is the HT of IMF c[n], A[n] is time dependent amplitude, and φ[n]
is instantaneous phase of a AIMF.

5
Figure 2: The IMFs of right hand (RH) MI task EEG signal.

2.3. Analytic IMF based features extraction

The AIMFs assessed the following characteristics of EEG signals. (i) The
AIMFs characterizes the time-dependent amplitude and frequency component
100 corresponding to the variation of EEG signal. (ii) The AIMFs of EEG signal
rotate in a specific direction around the unique center. (iii) AIMFs have multi-
resolution property, so proposed features are able to extract different levels of
information from EEG signals. In this work following AIMFs based features are
extracted for classification of different MI-tasks EEG signals.

105 2.3.1. Raw moment of first derivative of IF (RMFDIF)


The RMFDIF feature represents the weighted successive difference of IF
of an IMF. It asses the frequency variability characteristics of MI-based EEG
signals by including extreme values. For computing RMFDIF, the IF from phase

6
of mth AIMF (Φm ) and its difference (δf ) is defined as:

1
fi = dif f {Φm }, δf = |dif f (fi )|. (3)

110 The IFs of IMFs of Figure 2 is shown in Figure 3. IF difference (δf ) defined in
Eq. (3), is used for the computation of RMFDIF feature as [23]:

Figure 3: The IFs of different IMFs of RH MI task EEG signal.

N −1
1 X
RM F DIF = δf [n] (4)
N − 1 n=1

where, N is number of samples in IF.

2.3.2. Area computation from analytic IMF


The center tendency measure (CTM) is used to recognize the variability
115 present in EEG signals. The radius of an area is equivalent to 95% of CTM.
The CTM of an AIMF z[n] can be defined as [24]:
PN
n=1 δd [n]
CT M = (5)
N

7
 0.5
 1 if [<{z[n]}]2 + [={z[n]}]2 <r 123
δd [n] = (6)
 0 otherwise
where, N is a total number of samples in AIMF and the area of AIMF computed
by the radius (r) of circular region to define 95% CTM.

120 2.3.3. Spectral moment of power spectral density (SMPSD)


The power spectral density (PSD) is used to find the power and dominant
frequency components present in EEG signal. In this paper, Welch’s method is
used for computation of PSD. The PSD of an AIMF z[n] can be represented as:
 2 
 1 XT 
Sz (f ) = limT −→∞ z[n]e−j2πf n (7)

 2T


n=−T

The PSD of different IMFs is shown in Figure 4.

Figure 4: The PSD of different IMFs of RH MI task EEG signal.

125 The spectral moment of PSD is used to define greater order shape of EEG data
points, which can be defined as [25]:
L
X
SM P SD = k.P SDk (8)
k=1

8
where, L is the number of point in PSD.

2.3.4. Peak value of PSD


The peak value of PSD is selected as a feature, which represents the highest
130 overshoot of PSD and it can be computed as [25]:

P SDpeak = max{Sz (f )} (9)

2.4. Feature normalization

Feature normalization controls the biased nature of the classifier, towards a


particular set of attributes, it works when feature size is large or there is a lot of
variation in each feature for different classes [26, 27]. The same condition is met
135 in this work, with extracted features. The linear scaling to unit range along the
norm or length of the feature vector is used as a feature normalization method.
The following steps are used to get
 normalized feature matrix. 
a a12 . . . . a1q
 11 
 
 a21 a22 . . . . a2q 
1. Let the feature matrix is, F =



 . . . . . . . 
 
ap1 ap2 . . . .apq
where, p represents the number of feature vectors and q is the instances.
140 2. Get the length of each feature vector.
3. Find out the inverse of norm or length along the feature instances of each
feature vector and construct thetransformation vector
 as:
√ 1

 a11 +a12 ...a1q   w1q 
 √ 1
   

 a21 +a22 ...a2q   w2q 
W = =


  ...
.....
 
   
 
√ 1 wpq
ap1 +ap2 ...apq
4. Multiply each element of row vector {aij ∈ i = (1, p), j = (1, q)} to their
145 corresponding transformation factor wiq . This will gives the normalized feature
vector as:

9
 
a11 ∗ w1q a12 ∗ w1q . . . . a1q ∗ w1q
 
 a21 ∗ w2q a22 ∗ w2q a2q ∗ w2q 
 
. . . .
FN = 



 . . . . . . . 
 
ap1 ∗ wpq ap2 ∗ wpq . . . .apq ∗ wpq
where, FN is unit length normalized feature matrix.

2.5. Least-squares support vector machine (LS-SVM)

150 The proposed AIMFs based features are tested on LS-SVM classifier for the
classification of RH and RF MI tasks EEG signals. The final decision function
for the LSSVM classifier is given as [28]:
"N #
X T

f (x) = sign αi yi K(x, xi ) + b (10)


i=1

where xi is the ith input feature vector of d dimensions, yi (which is either +1 or


-1) is the class label of xi , NT is the number of feature vectors, b is the bias term,
155 αi (i = 1, 2, ...., NT ) denotes Lagrange multipliers also called as support values,
and K(x, xi ) is a kernel function. In this work, Linear, Polynomial, and Radial
basis function (RBF) kernels are tested, which can be represented, respectively
as [28]:
K(x, xi ) = xT xi (11)

K(x, xi ) = (xT xi + 1)l (12)


160
−||x−xi ||2
K(x, xi ) = exp 2σ 2 (13)

where, l is degree of polynomial kernel and σ is width of RBF kernel. The


detailed description of the LS-SVM algorithm could be found in reference [28].

3. Results

The analytic-IMFs based features are extracted for classification of RH and


165 RF MI tasks, EEG signals. The unit length feature normalization method is
applied on extracted features to avoid their dominating behaviour. The statis-
tical significance of normalized features is assessed by probabilistic (p)−value

10
of Kruskal-Wallis statistical test. Tables 1-4 present, the p−values of extracted
features RMFDIF, area, SMPSD, and peak value of PSD, respectively. In these
170 tables, the lesser p−values of all features show their statistical significance for
discrimination of different MI tasks. In Tables 1-4, higher frequency IMFs based
features are more statistically significant for all channels as compared to lower
frequency IMFs based features. Hence, first four IMFs-based features are consid-
ered for the classification of different MI tasks. The ranges (mean and standard
175 deviation) of these IMFs-based features for RH and RF MI tasks EEG signals
are presented in Tables 5-8. Table 5 shows that, the range of RMFDIF feature
is higher for RH MI task EEG signals as compared to RF MI task EEG signals,
for all IMFs. This range variation might be shown because of the larger rate of
change of IF corresponding to RH MI task EEG signals as compared to RF MI
180 tasks EEG signals. Table 6 presents the variation of area feature, the IMF1 and
IMF2 based area feature values show larger variation corresponding to RF MI
task EEG signals. It could be due to, the larger amplitude of IMF1 and IMF2
of RF MI task EEG signals as compared to RH MI task EEG signals. The
opposite behaviour is observed for IMF3 and IMF4, in Table 6. The variation
185 in SMPSD feature is presented in Table 7, the SMPSD shows higher values for
IMF1 and IMF2 of RF MI task EEG signals. It might be happen due to the
larger amplitude of IMF1 and IMF2 of RF MI task EEG signals as compared to
RH MI task EEG signals. The opposite nature is observed in Table 7, for IMF3
and IMF4. The peak value of PSD is simply the maximum overshoot of PSD,
190 which depends on the maximum value of IMF. In Table 8, the larger values of
the peak of PSD feature observed corresponding to RH MI task EEG signals.
It could be due to the larger maximum value of IMFs corresponding to RH MI
task EEG signals as compared to RF MI task EEG signals.
The proposed features in Tables 1-8 clearly show the differences in electrical
195 activity of brain corresponding to RH and RF MI tasks, EEG signals. The
dimension of IMFs-wise feature-sets, for RH and RF MI tasks EEG signals are
40 × 200 and 40 × 250, respectively. In these dimensions, 40 rows represent the
four features extracted for 10-channels, while 200 and 250 columns represent

11
the number of observation of RH and RF MI tasks EEG signals, respectively.
200 The IMF-wise feature set is applied as input to LS-SVM classifier for classifi-
cation of RH and RF MI tasks, EEG signals. The classification performance
of LS-SVM classifier is evaluated in terms of following performance parameters:
accuracy (ACC), sensitivity (SEN), specificity (SPE), positive predictive value
(PPV), negative predictive value (NPV), and error rate detection (ERD). These
205 parameters are computed using 10-cross validation approach and different ker-
nel functions for LS-SVM classifier. In this work, the used kernel functions and
their parameter are as follows: linear kernel is parameter independent, polyno-
mial kernel of degree l=5, RBF kernel of width σ=10. Table 9 presents, the
IMFs wise summary of evaluated performance parameters with different Kernel
210 functions of the LS-SVM classifier. The overall best performance parameters are
obtained for IMF1 with RBF kernel in Table 9 as: ACC 97.56%, SEN 96.45%,
SPE 98.96%, PPV 99.2%, and NPV 95.2%, and ERD 4.28%.

4. Discussions

This works proposes analytic-IMF based features for classification of differ-


215 ent MI tasks, EEG signals. The proposed features namely: RMFDIF, area,
SMPSD, and peak value of PSD. These specific features are selected because of
their smaller p−values and sufficient range variation shown in Tables 1-8 for RH
and RF MI tasks, EEG signals. The extracted features are tested on LS-SVM
classifier with different kernel functions. In Table 9, the RBF kernel provides
220 maximum MI tasks classification accuracy for IMF1, because IMF1-based fea-
tures are more statistically significant as compared to other IMFs-based features
in Tables 1-4.
As mentioned in dataset section the data was recorded from five subjects.
The all subjects average distribution of true positive rate (TPR) and true nega-
225 tive rate (TNR) with different folds for IMF1 are shown in Figure 5. Here, TPR
signifies the successful detection rate for the RH MI task class EEG signals and
TNR signifies the successful detection rate for the RF MI task class EEG sig-

12
Table 1: The p−values of RMFDIF feature for different IMFs.

Channels IMF1 IMF2 IMF3 IMF4 IMF5 IMF6

Channel1 9.75 ∗ 10−06 1.54 ∗ 10−12 8.89 ∗ 10−07 7.18 ∗ 10−05 0.0024 0.0588

Channel2 0.2075 1.62 ∗ 10−12 3.37 ∗ 10−07 0.0201 0.3505 0.5581

Channel3 0.0286 2.44 ∗ 10−13 1.85 ∗ 10−09 7.05 ∗ 10−07 0.2929 0.0875

Channel4 0.0016 5.48 ∗ 10−17 1.3 ∗ 10−08 9.89 ∗ 10−05 0.0758 0.4047

Channel5 0.0003 3.55 ∗ 10−16 9.17 ∗ 10−09 0.0002 0.0007 0.0761

Channel6 6.89 ∗ 10−19 2.36 ∗ 10−14 1.14 ∗ 10−07 0.0002 0.053 0.0163

Channel7 0.0008 7.43 ∗ 10−18 1.08 ∗ 10−10 1.8 ∗ 10−05 0.0913 0.0105

Channel8 4.35 ∗ 10−05 2.94 ∗ 10−19 3.25 ∗ 10−12 2.59 ∗ 10−06 0.0018 0.0095

Channel9 4.61 ∗ 10−11 2.06 ∗ 10−19 1.49 ∗ 10−07 8.16 ∗ 10−06 0.0013 0.0256

Channel10 3.21 ∗ 10−10 2.52 ∗ 10−18 3.31 ∗ 10−12 3.92 ∗ 10−57 0.0472 0.001

Table 2: The p−values of area feature for different IMFs.

Channels IMF1 IMF2 IMF3 IMF4 IMF5 IMF6

Channel1 1.4 ∗ 10−48 1.6 ∗ 10− 19 1.97 ∗ 10−77 0.0014 4.34 ∗ 10−40 0.001

Channel2 2.36 ∗ 10−10 0.4199 2.77 ∗ 10−35 8.24 ∗ 10−13 3.53 ∗ 10−41 0.7949

Channel3 1.14 ∗ 10−08 0.002 7.51 ∗ 10−63 9.89 ∗ 10−09 0.6643 0.0032

Channel4 1.44 ∗ 10−12 0.0005 1.21 ∗ 10−06 5.83 ∗ 10−19 0.9388 0.0003

Channel5 1.95 ∗ 10−10 8.31 ∗ 10−05 1.66 ∗ 10−25 0.0088 0.3345 0.7772

Channel6 4.04 ∗ 10−06 3.65 ∗ 10−08 0.0016 0.0154 0.0003 0.0385

Channel7 6.38 ∗ 10−12 2.91 ∗ 10−08 2.07 ∗ 10−29 5.39 ∗ 10−17 1.13 ∗ 0−06 9.18 ∗ 10−64

Channel8 4.41 ∗ 10−82 4.41 ∗ 10−82 4.41 ∗ 10−82 1.75 ∗ 10−81 2.2 ∗ 10−80 8.37 ∗ 10−71

Channel9 0.1549 7.08 ∗ 10−19 2.19 ∗ 10−11 9.57 ∗ 10−23 1.75 ∗ 10−12 0.0006

Channel10 1.92 ∗ 10−14 3.71 ∗ 10−09 3.54 ∗ 10−57 2.96 ∗ 10−07 4.49 ∗ 10−44 0.0008

13
Table 3: The p−values of SMPSD feature for different IMFs.

Channels IMF1 IMF2 IMF3 IMF4 IMF5 IMF6

Channel1 6.53 ∗ 10−82 4.64 ∗ 10−65 8.26 ∗ 10−78 1.03 ∗ 10−59 1.36 ∗ 10−07 0.6416

Channel2 0.4061 0.0354 8.39 ∗ 10−46 5.59 ∗ 10−34 0.1894 1.14 ∗ 10−08

Channel3 2.17 ∗ 10− 21 0.089 4.73 ∗ 10−08 5.33 ∗ 10−67 0.6933 3.46 ∗ 10−27

Channel4 4.69 ∗ 10−07 2.27 ∗ 10−05 0.0041 9.61 ∗ 10−54 2.47 ∗ 10−09 7.1 ∗ 10−08

Channel5 2.04 ∗ 10−60 2.32 ∗ 10−09 0.0004 7.85 ∗ 10−80 0.1713 5.84 ∗ 10−08

Channel6 0.3283 0.0011 0.6061 4.43 ∗ 10−41 0.2224 0.4881

Channel7 0.0002 1.07 ∗ 10−27 4.1 ∗ 10−17 2.36 ∗ 10−50 1.88 ∗ 10−10 1.57 ∗ 10−10

Channel8 4.41 ∗ 10−82 4.41 ∗ 10−82 1.12 ∗ 10−80 2.4 ∗ 10−62 4.41 ∗ 10−82 6.46 ∗ 10−82

Channel9 1.26 ∗ 10−25 8.32 ∗ 10−19 2.82 ∗ 10−61 2.16 ∗ 10−77 0.0022 9.55 ∗ 10−08

Channel10 2.61 ∗ 10−58 3.53 ∗ 10−09 6.43 ∗ 10−63 0.0007 0.0487 0.0472

Table 4: The p−values of peak of PSD feature for different IMFs.

Channels IMF1 IMF2 IMF3 IMF4 IMF5 IMF6

Channel1 3.17 ∗ 10−41 4.18 ∗ 10−70 0.007 8.84 ∗ 0−07 5.29 ∗ 10−23 3.11 ∗ 10−55

Channel2 1.93 ∗ 10−51 3.53 ∗ 10−36 1.24 ∗ 10−06 1.8 ∗ 10−09 0.0009 0.3332

Channel3 0.05 3.96 ∗ 10−68 0.047 8.55 ∗ 10−09 0.0001 0.4431

Channel4 0.038 5.21 ∗ 10−37 2.84 ∗ 10−13 2.44 ∗ 10−17 0.0008 1.04 ∗ 10−6

Channel5 0.1968 5.25 ∗ 10−74 0.0018 2.02 ∗ 10−06 0.4026 0.4314

Channel6 1.26 ∗ 10−07 1.85 ∗ 10−17 0.0133 0.0015 0.0011 0.5787

Channel7 8.72 − 10−20 2.24 ∗ 10−31 4.9 ∗ 10−05 2.44 ∗ 10−08 0.0003 0.0038

Channel8 8.57 ∗ 10−81 1.59 ∗ 10−69 4.4 ∗ 10−82 4.52 ∗ 10−82 4.4 ∗ 10−82 4.4 ∗ 10−82

Channel9 8.53 ∗ 10−34 2.1 ∗ 10−78 2.08 ∗ 10−15 9.54 ∗ 10−26 0.0013 0.004

Channel10 3.99 ∗ 10−73 1.07 ∗ 10−58 0.0009 0.0475 0.0002 0.3777

14
Table 5: Ranges (Mean±Standard Deviation) of RMFDIF feature for IMFs of RH and RF
MI task EEG signals.

Channels MI Tasks IMF1 IMF2 IMF3 IMF4

Mean±std Mean±std Mean±std Mean±std

Channel1 RH 0.0693 ± 0.0143 0.0695 ± 0.0130 0.0676± 0.0209 0.0655 ± 0.0266

RF 0.0619 ± 0.0129 0.0623 ± 0.0112 0.0611 ± 0.0165 0.0589 ± 0.0231

Channel2 RH 0.0637 ± 0.0308 0.0698± 0.0115 0.0685± 0.0176 0.0659 ± 0.0258

RF 0.0575 ± 0.0264 0.0625 ± 0.0099 0.0615 ± 0.0149 0.0593 ± 0.0220

Channel3 RH 0.0681 ±0.0192 0.0694 ± 0.0135 0.0676 ± 0.0206 0.0658 ± 0.0259

RF 0.0609 ± 0.0171 0.0620 ± 0.0125 0.0602± 0.0194 0.0582 ± 0.0249

Channel4 RH 0.0685 ± 0.0174 0.0698 ± 0.0114 0.0678 ± 0.0200 0.0657± 0.0262

RF 0.0612 ± 0.0159 0.0623 ± 0.0106 0.0609 ± 0.0172 0.0592 ± 0.0224

Channel5 RH 0.0687 ±0.0170 0.0696 ± 0.0128 0.0678 ± 0.0203 0.0651 ± 0.0277

RF 0.0614 ±0.0152 0.0622± 0.0115 0.0605 ± 0.0184 0.0588 ± 0.0232

Channel6 RH 0.0699 ±0.0109 0.0697 ±0.0121 0.0681 ± 0.0191 0.0665 ± 0.0242

RF 0.0626 ± 0.0093 0.0624 ± 0.0103 0.0608± 0.0174 0.0594 ± 0.0218

Channel7 RH 0.0688 ± 0.0163 0.0699 ± 0.0110 0.0685 ±0.0176 0.0664 ± 0.0245

RF 0.0616 ± 0.0144 0.0625 ± 0.0100 0.0608± 0.0173 0.0592 ± 0.0222

Channel8 RH 0.0691 ± 0.0149 0.0699± 0.0106 0.0683± 0.0183 0.0658 ± 0.0260

RF 0.0617 ± 0.0141 0.0623 ± 0.0109 0.0606 ± 0.0182 0.0585 ± 0.0242

Channel9 RH 0.0694 ± 0.0134 0.0700 ± 0.0104 0.0675 ± 0.0210 0.0655 ± 0.0267

RF 0.0621 ± 0.0119 0.0626 ± 0.0092 0.0608 ± 0.0175 0.0586 ± 0.0239

Channel10 RH 0.0696 ±0.0124 0.0699 ± 0.0109 0.0689 ± 0.0159 0.0673 ± 0.0218

RF 0.0622 ± 0.0112 0.0625 ± 0.0099 0.0612 ± 0.0159 0.0603 ± 0.0192

15
Table 6: Ranges of area feature for IMFs of RH and RF MI task EEG signals.

Channels MI Tasks IMF1 IMF2 IMF3 IMF4

Mean±std Mean±std Mean±std Mean±std

Channel1 RH 0.0136 ± 0.0696 0.0225 ± 0.0672 0.0486± 0.0514 0.0430 ± 0.0563

RF 0.0272 ± 0.0572 0.0402 ± 0.0489 0.0369 ± 0.0514 0.0352 ± 0.0526

Channel2 RH 0.0208 ± 0.0678 0.0225 ± 0.0672 0.0453 ± 0.0545 0.0490 ± 0.0511

RF 0.0213 ± 0.0597 0.0466 ± 0.0428 0.0444 ± 0.0451 0.0327 ± 0.0542

Channel3 RH 0.0219 ± 0.0674 0.0413 ± 0.0576 0.0494 ±0.0507 0.0470 ±0.0529

RF 0.0270 ± 0.0573 0.0416 ± 0.0478 0.0387 ± 0.0501 0.0306 ± 0.0555

Channel4 RH 0.0225 ±0.0672 0.0398 ± 0.0586 0.0496 ±0.0505 0.0507 ± 0.0495

RF 0.0287 ± 0.0565 0.0434 ± 0.0461 0.0374 ± 0.0511 0.0284 ± 0.0566

Channel5 RH 0.0204 ± 0.0679 0.04 ± 0.0585 0.0488 ± 0.0513 0.0327 ± 0.0628

RF 0.0272 ± 0.0572 0.0392 ± 0.0498 0.036 ± 0.0521 0.0309 ± 0.0553

Channel6 RH 0.0216 ± 0.0675 0.0287 ± 0.0648 0.0456 ± 0.0541 0.0292 ± 0.0645

RF 0.0269 ± 0.0574 0.0373 ± 0.0512 0.0366 ± 0.0517 0.0282± 0.0567

Channel7 RH 0.0256 ± 0.0661 0.0372 ± 0.0603 0.0450 ± 0.0547 0.0464 ± 0.0535

RF 0.0305 ±0.0555 0.0431 ± 0.0464 0.0328 ± 0.0542 0.0268 ± 0.0574

Channel8 RH 0.0270 ± 0.0655 0.0439 ± 0.0556 0.0482 ± 0.0518 0.0416 ± 0.0573

RF 0.0041 ± 0.0632 0.0041 ± 0.0632 0.0041 ± 0.0632 0.0043 ± 0.0632

Channel9 RH 0.0260 ± 0.0659 0.0338 ± 0.0622 0.0473 ± 0.0527 0.0385 ± 0.0595

RF 0.0210 ± 0.0598 0.0323 ± 0.0545 0.0294 ± 0.0561 0.0192± 0.0604

Channel10 RH 0.0241 ± 0.0666 0.0366 ± 0.0607 0.0478 ± 0.0522 0.0357 ± 0.0612

RF 0.0279 ± 0.0569 0.0430 ± 0.0464 0.0382 ±0.0505 0.0282 ± 0.0567

16
Table 7: Ranges of SMPSD feature for IMFs of RH and RF MI task EEG signals.

Channels MI Tasks IMF1 IMF2 IMF3 IMF4

Mean±std Mean±std Mean±std Mean±std

Channel1 RH 0.0051 ± 0.0707 0.0056 ± 0.0707 0.0364 ± 0.0608 0.0223 ± 0.0673

RF 0.0342 ± 0.0533 0.0239 ± 0.0587 0.0151 ± 0.0615 0.0257 ± 0.0579

Channel2 RH 0.0357 ± 0.0612 0.0263 ± 0.0658 0.0286 ± 0.0649 0.0347 ± 0.0618

RF 0.0329 ± 0.0541 0.0284 ±0.0566 0.0294 ± 0.0561 0.0216 ± 0.0595

Channel3 RH 0.0159 ± 0.0691 0.0355 ± 0.0613 0.0332 ± 0.0626 0.0415 ± 0.0574

RF 0.0295 ± 0.0561 0.0267 ± 0.0575 0.0301 ± 0.0557 0.0140± 0.0618

Channel4 RH 0.0188 ± 0.0683 0.0216 ± 0.0675 0.0384 ± 0.0595 0.0356 ± 0.0612

RF 0.0238 ± 0.0587 0.0114 ± 0.0623 0.0170 ± 0.0610 0.0241 ± 0.0586

Channel5 RH 0.0059 ± 0.0706 0.0276 ± 0.0653 0.0301 ± 0.0642 0.0280 ± 0.0651

RF 0.0169 ± 0.0611 0.0151 ± 0.0615 0.0219 ± 0.0595 0.0199 ± 0.0601

Channel6 RH 0.0233 ± 0.0669 0.0302 ± 0.0641 0.0320 ± 0.0632 0.0209 ± 0.0677

RF 0.0242 ± 0.0586 0.0216 ± 0.0596 0.0236 ± 0.0588 0.0226 ± 0.0592

Channel7 RH 0.0271 ± 0.0655 0.0096 ± 0.0702 0.0280 ± 0.0651 0.0303 ± 0.0641

RF 0.0333 ± 0.0539 0.0210 ± 0.0598 0.0154 ± 0.0615 0.0178 ± 0.0608

Channel8 RH 0.0220 ± 0.0674 0.0345 ± 0.0619 0.0326 ± 0.0629 0.0324 ± 0.0630

RF 0.0040 ± 0.0632 0.0040 ± 0.0632 0.0040 ± 0.0632 0.0040 ± 0.0632

Channel9 RH 0.0071 ± 0.0705 0.0157 ± 0.0691 0.0312± 0.0636 0.0267 ± 0.0656

RF 0.0117 ± 0.0623 0.0279 ± 0.0569 0.0233 ± 0.0589 0.0196± 0.0602

Channel10 RH 0.0078 ± 0.0705 0.0190 ± 0.0683 0.0254 ± 0.0661 0.0252 ± 0.0662

RF 0.0297 ± 0.0560 0.0277 ± 0.0570 0.0208 ± 0.0599 0.0204 ± 0.0600

17
Table 8: Ranges of peak of PSD feature for IMFs of RH and RF MI task EEG signals.

Channels MI Tasks IMF1 IMF2 IMF3 IMF4

Mean±std Mean±std Mean±std Mean±std

Channel1 RH 0.0144 ± 0.0694 0.0092 ± 0.0703 0.0442 ± 0.0553 0.0410 ± 0.0577

RF 0.0439 ± 0.0456 0.0393 ± 0.0496 0.0339 ± 0.0535 0.0283 ± 0.0567

Channel2 RH 0.0245 ± 0.0665 0.0343± 0.0620 0.0423± 0.0568 0.0422 ± 0.0569

RF 0.0208 ± 0.0598 0.0353 ± 0.0526 0.0304 ± 0.0556 0.0307 ± 0.0554

Channel3 RH 0.0485 ± 0.0516 0.0421 ± 0.0569 0.0435 ± 0.0559 0.0421 ± 0.0569

RF 0.0410 ± 0.0483 0.0390 ± 0.0499 0.0349 ± 0.0528 0.0265 ± 0.0575

Channel4 RH 0.0500 ± 0.0501 0.0402± 0.0583 0.0374 ± 0.0602 0.0445 ± 0.0551

RF 0.0433 ± 0.0462 0.0240 ± 0.0587 0.0194 ± 0.0603 0.0246 ± 0.0584

Channel5 RH 0.0384 ± 0.0595 0.0424 ± 0.0567 0.0399 ± 0.0585 0.0339 ± 0.0622

RF 0.0416 ±0.0478 0.0379 ± 0.0507 0.0314 ± 0.0550 0.0267 ± 0.0574

Channel6 RH 0.0460 ± 0.0538 0.0317± 0.0634 0.0415 ± 0.0574 0.0358 ± 0.0611

RF 0.0408 ± 0.0484 0.0302 ± 0.0557 0.0335 ± 0.0537 0.0302 ± 0.0557

Channel7 RH 0.0531 ± 0.0469 0.0342± 0.0621 0.0411± 0.0577 0.0413 ± 0.0575

RF 0.0457 ± 0.0439 0.0358 ± 0.0522 0.0303 ± 0.0556 0.0298 ± 0.0559

Channel8 RH 0.0530± 0.0469 0.0317 ± 0.0633 0.0277 ± 0.0652 0.0287 ± 0.0648

RF 0.0040 ± 0.0632 0.0040 ± 0.0632 0.0040 ± 0.0632 0.0041 ± 0.0632

Channel9 RH 0.0412 ± 0.0576 0.0374± 0.0602 0.0380 ± 0.0598 0.0352 ± 0.0615

RF 0.0375 ± 0.0510 0.0354 ± 0.0525 0.0188 ± 0.0605 0.0153 ± 0.0615

Channel10 RH 0.0516 ± 0.0485 0.0405 ± 0.0581 0.0383 ± 0.0596 0.0288 ± 0.0647

RF 0.0445± 0.0450 0.0347 ± 0.0530 0.0289 ± 0.0564 0.0290 ±0.0563

18
Table 9: The performance parameters of proposed method using different kernels of LS-SVM
classifier.

Kernel Function IM F s ACC SEN SPE PPV NPV ERD

(Parameters) (%) (%) (%) (%) (%) (%)

IM F1 91.78 91.76 91.79 93.60 89.50 14.51

Linear Kernel IM F2 93.33 94.00 92.50 94.00 92.50 12.00

IM F3 94.89 95.22 94.47 95.60 94.00 9.16

IM F4 85.78 82.98 90.48 93.60 76.00 22.70

IM F1 94.89 97.10 92.34 93.60 96.50 9.54

Polynomial Kernel IM F2 95.11 94.19 96.35 97.20 92.50 8.53

(l=5) IM F3 88.89 84.97 95.73 97.20 78.50 17.48

IM F4 88.22 85.05 93.49 95.60 79.00 18.86

IM F1 97.56 96.45 98.96 99.2 95.2 4.28

RBF Kernel IM F2 96.22 95.33 97.41 98.00 94.00 6.61

(σ =10) IM F3 95.33 100.00 90.50 91.60 100.00 9.17

IM F4 93.11 91.95 94.71 96.00 89.50 11.88

Figure 5: The TPR and TNR patterns of proposed method.

19
Table 10: The performance report of proposed method with other existing methods in terms
of subjects (aa, al, av, aw, ay) wise accuracy (%), average accuracy (AACC), and standard
deviation (SD).

Authors Methods aa al av aw ay AACC SD

Proposed Approach AFE and FN based approach 97.78 98.89 96.67 98.89 95.56 97.56 1.45

Kevric et al. [16] MSPCA, WPD, HOS and k-NN 96 92.3 88.9 95.4 91.4 92.8 2.93

Wang et al. [17] OA and NB based approach 97.92 97.88 98.26 94.47 93.26 96.36 2.32

Zhang et al. [18] Z-LDA 77.7 100.0 68.4 99.60 59.9 81.1 18.2

Siuly and Li [19] CC based LS-SVM 97.88 99.17 98.75 93.43 89.36 95.72 4.35

Lu et al. [20] R-CSP with aggregation 76.80 98.20 74.50 92.90 77.00 83.90 10.86

Where, AFE analytic feature extraction and FN feature normalization, MSPCA multi-scale principal component
analysis, WPD wavelet packet decomposition, HOS higher order statistics, and K-NN K-nearest neighbor, OA op-
timum allocation and NB naive bayes, Z-LDA Z-score linear discriminant analysis, CC cross correlation, R-CSP
regularized common spatial pattern.

nals. Figure 5 clearly shows that both class detection rates are closer to highest
100% detection rate. Hence, Table 9 and Figure 5 evident that proposed method
230 provides a fairly accurate and stable solution, for the classification of RH and
RF MI tasks, EEG signals. Table 10 presents, the performance comparison re-
port of proposed method with other existing methods. The proposed method
provides consistent subjects-wise performance in a range of 95.56-98.89%, while
the other methods show fluctuating results in subject-wise performance, which
235 is also reflected in terms of their larger value of standard deviation. In Table 10,
the proposed method achieved highest MI tasks classification accuracy 97.56%
and minimum standard deviation 1.45% as compared to other existing methods.
The results of proposed method show its deftness for classification of different
MI tasks in BCI system. Therefore, the proposed method can be used for devel-
240 oping practical application of RH and RF MI tasks in BCI systems, like robotic
arm and wheelchair controlling etc.

20
5. Conclusion

In this work, the AIMFs-based features are proposed for classification of


different MI tasks, EEG signals. The AIMFs are obtained by applying EMD
245 and HT on raw EEG signal. The obtained AIMFs are employed for extraction of
features. The dominant nature of specific class features is controlled by applying
unit length feature normalized method. The normalized features are applied as
inputs to LS-SVM classifier and performance parameters are computed to assess
the performance of proposed method. The contribution of proposed framework
250 can be summarized as follows: (i) Introduce AIMFs-based features namely:
RMFDIF, area, SMPSD, and peak value of PSD for extracting representative
information from EEG signals, (ii) the proposed features have the capability to
yield better classification results as compared to other existing methods, (iii)
the obtained results of proposed method as, maximum classification accuracy
255 97.56% and minimum standard deviation 1.45%.
The results of proposed method show its adroitness for developing practical BCI
applications of RH and RF MI tasks, like robotic arm and wheelchair controlling.
In near future, the proposed method can be used to classify other physiological
and pathological states of the brain using EEG signals.

260 References

[1] C. Neuper, G. Müller, A. Kübler, N. Birbaumer, G. Pfurtscheller, Clinical


application of an EEG-based brain–computer interface: a case study in
a patient with severe motor impairment, Clinical neurophysiology 114 (3)
(2003) 399–409.

265 [2] A. R. Hassan, M. I. H. Bhuiyan, An automated method for sleep staging


from EEG signals using normal inverse gaussian parameters and adaptive
boosting, Neurocomputing 219 (2017) 76–87.

[3] A. R. Hassan, S. Siuly, Y. Zhang, Epileptic seizure detection in EEG sig-

21
nals using tunable-Q factor wavelet transform and bootstrap aggregating,
270 Computer methods and programs in biomedicine 137 (2016) 247–259.

[4] A. R. Hassan, M. I. H. Bhuiyan, Automated identification of sleep states


from EEG signals by means of ensemble empirical mode decomposition
and random under sampling boosting, Computer Methods and Programs
in Biomedicine 140 (2017) 201–210.

275 [5] A. R. Hassan, M. I. H. Bhuiyan, Automatic sleep scoring using statistical


features in the EMD domain and ensemble methods, Biocybernetics and
Biomedical Engineering 36 (1) (2016) 248–255.

[6] A. R. Hassan, A. Subasi, Automatic identification of epileptic seizures from


EEG signals using linear programming boosting, computer methods and
280 programs in biomedicine 136 (2016) 65–77.

[7] M. Arns, J. Gunkelman, S. Olbrich, C. Sander, U. Hegerl, EEG vigilance


and phenotypes in neuropsychiatry: Implications for intervention, Neuro-
modulation and neurofeedback: Techniques and applications (2010) 79–
123.

285 [8] F. Lotte, M. Congedo, A. Lécuyer, F. Lamarche, B. Arnaldi, A review of


classification algorithms for EEG-based brain–computer interfaces, Journal
of neural engineering 4 (2) (2007) R1.

[9] X. Yong, R. K. Ward, G. E. Birch, Sparse spatial filter optimization for


EEG channel reduction in brain-computer interface, in: Acoustics, Speech
290 and Signal Processing, 2008. ICASSP 2008. IEEE International Conference
on, IEEE, 2008, pp. 417–420.

[10] W. Wu, X. Gao, B. Hong, S. Gao, Classifying single-trial EEG during


motor imagery by iterative spatio-spectral patterns learning (ISSPL), IEEE
Transactions on Biomedical Engineering 55 (6) (2008) 1733–1743.

22
295 [11] H.-I. Suk, S.-W. Lee, A novel bayesian framework for discriminative fea-
ture extraction in brain-computer interfaces, IEEE Transactions on Pattern
Analysis and Machine Intelligence 35 (2) (2013) 286–299.

[12] M. Li, C. Lu, The recognition of EEG with CSSD and SVM, in: Intelligent
Control and Automation (WCICA), 2012 10th World Congress on, IEEE,
300 2012, pp. 4741–4746.

[13] Y. Li, P. P. Wen, et al., Clustering technique-based least square support


vector machine for EEG signal classification, Computer methods and pro-
grams in biomedicine 104 (3) (2011) 358–372.

[14] S. K. Bashar, A. R. Hassan, M. I. H. Bhuiyan, Motor imagery movements


305 classification using multivariate EMD and short time fourier transform, in:
India Conference (INDICON), 2015 Annual IEEE, IEEE, 2015, pp. 1–6.

[15] S. K. Bashar, A. R. Hassan, M. I. H. Bhuiyan, Identification of motor


imagery movements from EEG signals using dual tree complex wavelet
transform, in: Advances in Computing, Communications and Informatics
310 (ICACCI), 2015 International Conference on, IEEE, 2015, pp. 290–296.

[16] J. Kevric, A. Subasi, Comparison of signal decomposition methods in clas-


sification of EEG signals for motor-imagery BCI system, Biomedical Signal
Processing and Control 31 (2017) 398–406.

[17] H. Wang, Y. Zhang, et al., Detection of motor imagery EEG signals employ-
315 ing naı̈ve bayes based learning process, Measurement 86 (2016) 148–158.

[18] R. Zhang, P. Xu, L. Guo, Y. Zhang, P. Li, D. Yao, Z-score linear discrim-
inant analysis for EEG based brain-computer interfaces, PloS one 8 (9)
(2013) e74433.

[19] S. Siuly, Y. Li, Improving the separability of motor imagery EEG sig-
320 nals using a cross correlation-based least square support vector machine
for brain–computer interface, IEEE Transactions on Neural Systems and
Rehabilitation Engineering 20 (4) (2012) 526–538.

23
[20] H. Lu, H.-L. Eng, C. Guan, K. N. Plataniotis, A. N. Venetsanopoulos, Reg-
ularized common spatial pattern with aggregation for EEG classification in
325 small-sample setting, IEEE transactions on Biomedical Engineering 57 (12)
(2010) 2936–2946.

[21] B. Blankertz, K.-R. Muller, D. J. Krusienski, G. Schalk, J. R. Wolpaw,


A. Schlogl, G. Pfurtscheller, J. R. Millan, M. Schroder, N. Birbaumer, The
BCI competition III: Validating alternative approaches to actual BCI prob-
330 lems, IEEE transactions on neural systems and rehabilitation engineering
14 (2) (2006) 153–159.

[22] N. E. Huang, Z. Shen, S. R. Long, M. C. Wu, H. H. Shih, Q. Zheng, N.-C.


Yen, C. C. Tung, H. H. Liu, The empirical mode decomposition and the
Hilbert spectrum for nonlinear and non-stationary time series analysis, in:
335 Proceedings of the Royal Society of London A: mathematical, physical and
engineering sciences, Vol. 454, The Royal Society, 1998, pp. 903–995.

[23] Y. Ohsawa, S. Tsumoto, Chance discoveries in real world decision mak-


ing: data-based interaction of human intelligence and artificial intelligence,
Vol. 30, Springer, 2006.

340 [24] R. B. Pachori, V. Bajaj, Analysis of normal and epileptic seizure EEG
signals using empirical mode decomposition, Computer methods and pro-
grams in biomedicine 104 (3) (2011) 373–381.

[25] S.-M. Zhou, J. Q. Gan, F. Sepulveda, Classifying mental tasks based on


features of higher-order statistics from EEG signals in brain–computer in-
345 terface, Information Sciences 178 (6) (2008) 1629–1640.

[26] M. Shanker, M. Y. Hu, M. S. Hung, Effect of data standardization on neural


network training, Omega 24 (4) (1996) 385–397.

[27] S. Kotsiantis, D. Kanellopoulos, P. Pintelas, Data preprocessing for su-


pervised leaning, International Journal of Computer Science 1 (2) (2006)
350 111–117.

24
[28] J. A. Suykens, J. Vandewalle, Least squares support vector machine clas-
sifiers, Neural processing letters 9 (3) (1999) 293–300.

25
Highlights:

 Features are extracted from analytic intrinsic mode functions of EEG signals.
 Features used as input to LS-SVM classifier for classification of MI tasks EEG signals.
 Proposed features with LS-SVM classifier provide better performance as compared to existing methods.

You might also like