Endro Yulianto et al. / International Journal of Engineering Science and Technology (IJEST)
CHARACTERISTIC CORRELATION OF EEG SIGNAL THE MOTOR MOVEMENT TO MEXICAN HAT WAVELET 1
ENDRO YULIANTO
Student of the Post-Graduate, Department of Electronical Engineering and Information Technology, Engineering Faculty, Gadjah Mada University, Yogyakarta, Indonesia Email address :
[email protected] 2
Dr.ADHI SUSANTO
Department of Electronical Engineering and Information Technology, Engineering Faculty, Gadjah Mada University, Yogyakarta, Indonesia 3
Dr.THOMAS SRI WIDODO
Department of Electronical Engineering and Information Technology, Engineering Faculty, Gadjah Mada University, Yogyakarta, Indonesia 4
Dr.SAMEKTO WIBOWO
Medical Faculty, Gadjah Mada University, Yogyakarta, Indonesia
Abstract A motor movement occurs for the command of brain through neuromuscular channel. Brain Computer Interface (BCI), refers to a system that is able to interpret the brain signal to make human able to have an interaction with environment without a need of neuromuscular channel. Event–Related Synchronization/ Desynchronization (ERS/ERD) is one of EEG signal types that can be used in the system of BCI. The signal occurred as a result of this motor movement is characterized by a significant increase and a decrease of amplitude compared to amplitude when brain is in a resting state. The emergence of ERS/ERD in this research is stimulated through a motor movement to turn the simulation of steering wheel to the right and left direction. In order to obtain the signal, signal processing is used that is in the form of centering, band-pass filter at 4- 20 Hz, signal correlation and Eigen value decomposition (EVD). The characteristic of EEG signal from motor movement of “turn right” and “turn left” has an equal form compared to Mexican Hat Wavelet. Hence, the process of classification is conducted using the method of testing data correlation towards Mexican Hat Wavelet. The correlation is performed to the testing data of 22 volunteers simultaneously and 4 volunteers in a different time. The outcome of the research then shows that the high value of correlation is obtained from 4 volunteers that are correlated compared to 22 volunteers in the same time. This shows that the correlation of the testing data using Mexican Hat will result in a high value of correlation if done in each volunteer with a different scale. Keywords : BCI, ERD/ERS, centering, wavelet transform, signal correlation, EVD 1. Introduction Amyotrophic lateral sclerosis (ALS) and stroke refer to muscle diseases that eventually can cause disorders on neuromuscular channels leading the ability of the brain in controlling a voluntary movement in lost. Further, these diseases might stimulate the incapability of muscles to have any movements including the movement of eyes, respiratory or even all parts of body. To enable the control of voluntary movement, a solution can be used through a
ISSN : 0975-5462
Vol. 4 No.06 June 2012
2981
Endro Yulianto et al. / International Journal of Engineering Science and Technology (IJEST)
direct communication between the brain and the external environment without a media of the neuromuscular channels [Bai (2008)]. At this point, technology that can use the signals to the control or communication with external devices is known as BCI. EEG (electroencephalography) - for being done using a non-invasive method by placing electrodes on the scalp surface without any surgery on the brain commonly referred to an invasive method - becomes the most widely used bioelectrical measuring and recording device in the application of BCI. This, additionally, is due to the accessibility and affordability of EEG compared to other devices such as magnetoensefalograf (MEG) and Functional Magnetic Resonance Imaging (fMRI) [Benimeli (2007)]. One type of EEG signals used in BCI systems is ERS/ERD, a signal that appears as a result of motor movement in body. It refers to a pattern of EEG signal that occurs when someone makes a motion of limbs or simply imagines performing a movement that will be maximally measured at the motor cortex of the brain [Kooi (2010)]. As a signal of EEG, ERS/ERD is characterized by a significant increase and decrease of amplitude compared with the EEG signal amplitude in motionless state of human. ERS amplitude and ERD amplitude are respectively to be greater and smaller than the amplitude of the signal during the resting phase [Bai (2008)]. In order to obtain a signal, ERS / ERD require a signal processing. It is due to the loss of the signal in the background signal, other bioelectrical signals or noise signals. One of the major sources of noise and artifacts in the EEG signal is the influence of other bioelectrical devices such as electrooculogram (EOG), electrocardiogram (ECG), electromyogram (EMG) and the most importantly, the background signal from the brain activity itself [Atry]. Several studies have been conducted on the processing and classification of EEG signals associated with motor movement. Benimeli and Sharman (2007) in a study observed the mental state of volunteers toward the emerging of stimuli on computer monitor. Here, to obtain the filter on EEG data, the 8th order of inverse Chebyshev high-pass filter with the frequency at 1 Hz and 49th order inverse Chebyshev low-pass filter with that of at 9 Hz were used since their frequency range still covered a frequency range in the characteristic of EEG having a low frequency such as slow cortical potential (SCP) and high frequency such as ERS/ERD. Data of EEG subsequently was processed using Discrete Wavelet Transform (DWT) with the types of Daubechies-4 wavelet with decomposition level 4 and 5. Wavelet coefficients obtained from the DWT finally was used as input to the support vector machine (SVM) for the process of classification. Cortes et al. (2010) conducted a research on the signal of P300 from the brain frequently coming up in the centralparietal part of the brain. In this case, P300 was generated by giving the volunteer a visual stimulus. In this research an image of a ship would randomly appear on the screen and at the same time the volunteer was required to press a button in each sudden emergence of the image. Data of EEG were taken from 14 electrodes using a 10-20 point system including AF3, F7, F3, FC5, T7, P7, O1, O2, P8, T8, FC6, F4, F8, and AF4. To eliminate the noise and separate the potential evocation of artifact such as the bioelectrical signal from the eye blink, respiration, head movement or heart signal, an initial processing of EEG signal was applied using Independent Component Analysis (ICA). The outcome of this processing then was processed through the sub-band coding method of DWT with 11 levels of decomposition using the type of Daubechies-4 wavelet. Frequency range containing the most significant energy as a result of the wavelet decomposition was used as an input for the Adaptive Neuro Fuzzy Inference Systems (ANFIS) for the classification process. Darvishi and Ani (2007) in a study meanwhile conducted an analysis on EEG signals through a movement of the right and left hand taken from the electrodes of C3, Cz and C4. Filters on EEG were performed in the frequency range at 0.5 to 30 Hz. This was done for the existence of the EEG signals related to the motor movement in a range of alpha frequency (8-13 Hz) and beta frequency (18-25 Hz). In response, Continuous Wavelet Transform (CWT) method, having a wavelet of Morlet which is in accordance with the frequency range of alpha and beta waves, was used in the performance of the feature extraction. Afterwards, the obtained wavelet coefficient was performed to obtain the average mean energies, each of which with the most significant difference of mean and variance between the movement of the right hand and left hand is used as inputs to ANFIS for a classification process. Gareis et al. (2010) researched the EEG signals to control the movement of wheelchairs in which the EEG data were taken from a point electrode FZ, Cz, Pz, Oz, C3, C4 at 1024 Hz sampling frequency. To reduce the amount of data
ISSN : 0975-5462
Vol. 4 No.06 June 2012
2982
Endro Yulianto et al. / International Journal of Engineering Science and Technology (IJEST)
without any loss of information, the electrodes were downsampled by a factor of 4 in which 6 electrodes were to be reduced and selected only for two electrodes - Oz and Cz. The selection of those electrodes was due to the emergence of the potential of the brain signals with different polarities on them. To generate the potential of the brain, a stimulus was given, in this case, in the form of icon emerging on the monitor to control the movement of the wheelchair. The signal processing was conducted using the 8th order Chebyshev low-pass filter and Discrete Dyadic Wavelet Transform (DDWT) to decompose the signal into approximation and detail is performed. The types of wavelet used here included Biorthogonal 3.9, Daubechies 9, Coiflet 4 and Symlet 8. At last, an obtained wavelet coefficient was used as an input for the process of classification in measuring accuracy, sensitivity and specificity. The aim of this research is to characterize the correlation of EEG signal feature with the motor movement of “turn right” and ‘turn left” using mother wavelet that has a proper form to the feature. In order to obtain the characteristic of the EEG signal to the movement, several processes of signal processing; namely centering, bandpass filter, signal correlation and EVD are performed. The range of bandpass filter used in this research is at 4-20 Hz with a consideration that this range still covers the range of mu-rythm (~10 Hz) and beta (~20 Hz) in which the characteristic of EEG with the motor movement probably exist [Neuper et al,(2006)]. Centering is used to eliminate the noise background coming from the mental condition of the signal volunteer in the process of data collection. Centering in turn will eliminate the average of the signal subsequently resulting in a signal with zero average. There is a need to eliminate the noise background as this signal is not a result of a motor movement. The correlation of signal is functioned to determine the level of inter-related signals. EVD on the other side purposively is to perform transformation towards the linearly mixed signals, which, as a result, will create a new signal that is not correlated and has various variances. Data collection is performed using stimulus that is by moving the steering wheel to the right and the left side based on the arrow emerging on the screen. Electrode EEG is positioned in the area of motor cortex; namely C3, C4, P3, and P4 (Central and Parietal). Motor cortex refers to an interrelated part of the brain and has relevance with motor movement done by the parts of the body.
2.
Fundamental
BCI refers to a system taking and analyzing the signal from the brain nerves at the aim of creating communication between the brain and the computer. In a word, BCI is a brain signal and computer interface. 2.1. Eigenvalue decomposition EVD functions to do linear transformation towards the vector of observation in order to obtain a new vector in which its components are not correlated and its variants are various, Simply, the covariance matrix from is equal to the identity matrix: E xx Τ
I
(1)
EVD is from the covariance matrix E xx Τ EDEΤ where E refers to orthogonal matrix from the eigen vector Τ E xx and D refers to diagonal matrix from the eigen value, D diag d , … , d .. The transformation of EVD from the blending matrix results in a new matrix. Α Here, an equation is obtained: x
ED
⁄
E Τ Αs
Αs
(2)
The benefit of this method is that the new blending matrix Α is orthogonal. This can be seen from the equation below. E xx Τ
Τ
ΑE ss Τ Α
Τ
ΑΑ
I
(3)
EVD is also very functional in decreasing the dimension of data. Then, as frequently occurred in the statistical technique of principal component analysis (PCA), the eigen valued is seen from E xx Τ and the values considered to be too little are dropped. This then could affect on the decrease of noise. 2.2. Wavelet transform
ISSN : 0975-5462
Vol. 4 No.06 June 2012
2983
Endro Yulianto et al. / International Journal of Engineering Science and Technology (IJEST)
Short-Time Fourier Transform (STFT) refers to a method that functions to change the signal in a domain of time to be the one in time and frequency domain. It is the method of Fourier Transform (FT) with the windowing technique. In fact, one thing becoming a weak point from STFT is the size of the window that has an equal size for all ranges of frequency either as the high or low one. Moreover, the basic function of STFT is particularly equal to FT that is the function of a sinusoid. In response, CWT is developed as an alternative approach in coping with the problems in STFT. CWT has various sizes of window for all ranges of frequency that will be analyzed. For an analysis on high frequency, CWT for example will use a window with a narrow size and for analysis of low frequency; a wide window will be used. If STFT simply has a base of a sinusoid function, CWT will have a number of families from wavelet bases such as Haar, daubechies, Biortogonal, Coiflet, Symlets, Morlet, Mexican Hat and Meyer. Furthermore, CWT is defined as an integral signal multiplied by the scaled and shifted wavelet function (scale, position, time). C scale, position
ψ
CWT τ, s
∞ f ∞
Ψψ τ, s
t Ψ scale, position, t dt
| |
x t ψ∗
τ
dt
(4)
(5)
In equality of Eq. (4) and Eq.(5), refers to a shift, s refers to scale, and t refers to the transformation function, commonly called as mother wavelet. The results of CWT are a number of wavelet coefficients C, which is a function of scale and position. In this research, the calculation towards 22 volunteers was performed using the unipolar technique of 10-20 system at the points of C3, C4, P3 and P4.
Fig. 1. Technique of Installing the EEG electrode of 10-20 system (Sanei, 2007)
3.
Research Methodology
The data retrieval is performed using Biosignal Measurement Instrument K&H Type KL-710.
ISSN : 0975-5462
Vol. 4 No.06 June 2012
2984
Endro Yulianto et al. / International Journal of Engineering Science and Technology (IJEST)
Fig. 2. Bio-signal Measurement Instrument K&H Type KL-710
Fig. 3. Steering Wheel and Turning Sign
ISSN : 0975-5462
Vol. 4 No.06 June 2012
2985
Endro Yulianto et al. / International Journal of Engineering Science and Technology (IJEST)
C4
C3
P3
P4
Fig. 4. The points of EEG electrodes on volunteer
EEG data retrieval was performed by volunteers from the following stages: (1) EEG data measurements were carried out to 22 volunteers in fit condition and not suffering from any muscle paralysis. All the volunteers must be the ones actively moving their right side of the body – in a word, they were not the left-handed people. (2) Electrode was installed in the area of the motor cortex of the brain at the point C3, C4, P3 and P4 using the 1020 international system of measurement methods with unipolar right and left ear as a reference point. The installation of EEG electrode was carried out by nurses competent in the installation and operation of the electrode EEG equipments. (3) To make the volunteers stay comfortable during the process of data collection, the simulation of the steering wheel was placed on the table with the height and position adjustable to the condition of each volunteer. The stimuli in the form of arrows mark to the right and left directions were appeared on the screen approximately 6 meters at the distance of the volunteers - aimed to avoid the occurrence of brain electrical potential arising as a result of stimulation in the form of light emitted from computer screens. (4) When the stimulus appeared on the monitor screen in the form of an arrow to the right or to the left direction, the volunteers were asked to turn the steering wheel to the right or to the left. Once the arrow disappeared from the screen, the position of the steering wheel was returned to its initial position. For the volunteers, the data collection was taken in 5 stages on each in which one stage would contain EEG data by randomly presenting the stimuli of 5 times for turning right and other 5 times for turning left. Duration for each emergence of stimulus direction on the screen was for 3 seconds and the distance among the emergence of each stimulus was for 3 seconds. Segmentation was subsequently performed based on the time that had been obtained from the software. It was purposively to make the feature extraction process easier and merely was focused on the signal ERS / ERD occurred at 200-400 ms after stimulus by picking out the 1-second data before reaching 2 seconds once the stimulus in the form of direction of turning right and left were given. By doing so, the data segmented into 25 turnings to the right and 25 turnings to the left would be obtained from each volunteer. (5) Classifying the data from EEG signal measurement was done based on both the direction of the movement of turn right and turn left and the position of electrode C3, C4, P3 and P4.
ISSN : 0975-5462
Vol. 4 No.06 June 2012
2986
Endro Yulianto et al. / International Journal of Engineering Science and Technology (IJEST)
Fig. 5. The condition of the volunteer in the data retrieval
The block diagram of this research is presented as follows:
ISSN : 0975-5462
Vol. 4 No.06 June 2012
2987
Endro Yulianto et al. / International Journal of Engineering Science and Technology (IJEST)
Group of Data of Turning Right/Left: C3, C4, P3, P4
Data Segmentation 1 s before – 2 s after Stimulus
Bandpass Filter and Centering
Correlation Data By Shifting Hamming Window 200 Data Overlap 10 Data
The Biggest Correlation Coefficient Each Pair of Signal in Each Pair of Window
Group of Signal with the Biggest Correlation Coefficient
Eigenvalue Decomposition
Signal with biggest eigenvalue
Features of EEG Signal with Motor Movement of Turning Right and Left
Correlation of Characteristic and Mexican Hat Fig. 6. Diagram of Research Block
ISSN : 0975-5462
Vol. 4 No.06 June 2012
2988
Endro Yulianto et al. / International Journal of Engineering Science and Technology (IJEST)
Segmentation subsequently was performed after the classification of EEG signal measurement based on the direction of turn right and turn left and the electrode positions C3, C4, P3 and P4. This performance was intended to condition that the signals merely occurred as a result of motor movement of turn right and turn left to be processed at the following stage. Segmentation was performed by picking up the 1-second data before reaching 2 seconds once the stimulus appeared on the screen. Bandpass filter at 4-20 Hz and centering were used to reduce noise and artifact from another bioelectrical measuring devices such as electrooculogram (EOG), electrocardiogram (ECG), electromyogram (EMG) and background signals and other signals not needed in the next stage. Data with a length 600 were obtained from the EEG signal consisting of 25 data for turn right and other 25 data for turn left from each volunteer after segmenting process. Windowing technique, in this case Hammnig Window, was applied to the segmented data that had 200 data in length. Hamming window was to reduce the function of sinc on each signal in the window. Signal correlation was performed in each group of signal turn right and turn left by shifting the window along 600 data with 10 data overlapped in each shift.
w
0.54
0.46 cos
π
(6)
1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0
0
20
40
60
80
100
120
140
160
180
200
Fig. 7. Hamming window EEG Signal Without Hamming Window 0.02 0.01 0 -0.01 -0.02
0
20
40
60
80
100
120
140
160
180
200
160
180
200
EEG Signal With Hamming Window 0.02 0.01 0 -0.01 -0.02
0
20
40
60
80
100
120
140
Fig. 8. Signal of EEG with and without Hamming window
ISSN : 0975-5462
Vol. 4 No.06 June 2012
2989
Endro Yulianto et al. / International Journal of Engineering Science and Technology (IJEST)
The correlation of each signal in the group of 25 data of turn right and that of turn left from each volunteer separately was performed. Here, the 1st data was correlated to the 2nd orderly up to the 25th data as well as the 2nd data to the 3rd one up to 25th data and at last 24th data to 25th data. The 1st window from the 1st data was correlated to the 1st window up to the last window of the 2nd data. Subsequently, the 2nd window of the 1st data was also correlated to the 1st window up to the last window of the 2nd data. It then lasted to the last window of the 1st data correlated to the 1st window until the last window of the 2nd data. The correlation of each window shift from the 1st and 2nd data would result in a correlation coefficient. Furthermore, the largest correlation coefficient was sought from the correlation of the 1st data to the 2nd data. The signal from the window couple with the highest correlation coefficient was then included in the signal characteristics that in this case was carried out by the 1st data to the 2nd data: 25, the 2nd data to the 3rd data: 25, the 3rd data to the 4th: 25 and so on until the 24th data to the 25th data. From this, a group of data consisting of the signals that were a group of signal characteristics with a motor movement of “turn right” and “turn left” would be obtained. Correlation coefficient refers to a measure of the correlation signal. A large correlation coefficient indicates that the signals are correlated with each other, while a small correlation coefficient indicates that the signals are not correlated with each other. In this study the correlation coefficient was obtained by comparing the first eigenvalue to the second ones from the covariance matrix of the signal in a pair of window. The groups of correlated signals from the group of motor movement of turn right and turn left from each volunteer were then processed using EVD. The signal with the largest eigenvalue of the covariance matrix of the group of the correlated signal was then used as a characteristic EEG signals for motor movement that, in turn, could result in a potential change in brain signal called ERS / ERD. ERD refers to a characteristic of EEG signals for motor movement in the form of a decrease of amplitude compared to the signal in a resting state. Meanwhile ERS is in the form of increase of signal amplitude compared to the signals in the resting state [Huang, et al,( 2011)]. Hence, the signal characteristic the largest eigenvalue obtained from EVD method from 22 volunteers was grouped and its maximum peak point was sought subsequently to be altogether moved into one point. The average calculation was performed towards the group of signals with the maximum peak point that has been put together. This average signal was used as a characteristic of the motor movement of turn right and turn left.
4. Result of Research and Discussion At the phase of signal processing, an averaged signal with the point of maximum peak for the turning right and left is obtained as well as one averaged signal with the point of maximum peak for turning right and turning left from 22 volunteers in each point of electrode C3, C4, P3 and P4.
ISSN : 0975-5462
Vol. 4 No.06 June 2012
2990
Endro Yulianto et al. / International Journal of Engineering Science and Technology (IJEST)
EEG Signal Before Filtered -0.15 -0.2 -0.25 -0.3 -0.35
0
100
200
300
400
500
600
700
800
600
700
800
EEG Signal After Filtered 4-20 Hz 0.02 0.01 0 -0.01 -0.02
0
100
200
300
400
500
Fig. 9. EEG Signal before and after being filtered by 4-20 Hz (line = stimulus point) ELEKTRODE C3 0.15 RIGHT LEFT 0.1
0.05
0
-0.05
-0.1
0
20
40
60
80
100
120
140
160
180
200
Fig. 10. Feature of EEG Signal of the average result of 22 Volunteer in Electrode C3
ISSN : 0975-5462
Vol. 4 No.06 June 2012
2991
Endro Yulianto et al. / International Journal of Engineering Science and Technology (IJEST)
ELEKTRODE C4 0.15 RIGHT LEFT 0.1
0.05
0
-0.05
-0.1
0
20
40
60
80
100
120
140
160
180
200
Fig. 11. Feature of EEG Signal of average result of 22 Volunteers at Electrode C4 ELEKTRODE P3 0.25 RIGHT LEFT
0.2
0.15
0.1
0.05
0
-0.05
-0.1
0
20
40
60
80
100
120
140
160
180
200
Fig. 12. Feature of EEG signal of the average result of 22 volunteers at the electrode P3 ELEKTRODE P4 0.15 RIGHT LEFT 0.1
0.05
0
-0.05
-0.1
0
20
40
60
80
100
120
140
160
180
200
Fig. 13. Feature of EEG signal of the average result of 22 volunteers at the electrode P4
Once obtained, the feature of the signal of EEG of motor movement of “turning right” and “turning left” is correlated to Mexican Hat having a similar form with that feature.
ISSN : 0975-5462
Vol. 4 No.06 June 2012
2992
Endro Yulianto et al. / International Journal of Engineering Science and Technology (IJEST)
Fig. 14. Mexican Hat
The wavelet testing was conducted on a group of testing data that has been prepared. The testing data refers to the data also included in the signal processing in order to find out the signal characteristic of motor movement of turn right and turn left , yet the segmentation was performed with different data length. Segmentation on the testing data was performed within 3 seconds before reaching 3 seconds after the stimulus was given. This testing therefore will result in the testing data with 1200 data in length. This testing was conducted by correlating Mexican Hat to the testing data; namely the group of testing data of turn right and turn left in each position of electrodes C3, C4, P3 and P4. In the BCI system there are two conditions used to measure the accuracy of the system: Intentional Control (IC) and No-Control (NC). IC is a condition in which the BCI system is switched to control or communicate with external devices, while NC is a state where the system is in an inactive state [Faradji, et al,( 2010)]. In this study the IC period was determined in the range of 1 second to 2 seconds after the stimulus was given. Accuracy in turn was determined by calculating the number of signals with the highest correlation coefficient between the group of the testing data and the Mexican Hat in IC period. The following is the table of the result of correlation of the EEG data of turning right and turning left with Mexican Hat using the wavelet scale of 1:200. The correlation is performed to 22 volunteers simultaneously with an equal scale and correlation to 4 volunteers is performed separately. The numbers shown in the following table are the maximum and minimum values from the result of correlation from scale 1 to 200. The numbers shown in the following table cover “Peak Max” which is the correlation with the highest peak of the signal in the measurement period of the testing data to Mexican Hat. “Peak Min” meanwhile means correlation to the lowest peak of the signal and “Peak Max + Min” comes to be correlation with the highest and lowest peak and only one of two that is to be chosen. DIFF refers to a difference from the number of correlated signals. The highest value for DIFF with a positive value (+) shows that the scale of the wavelet is suitable for the data of turning right. Meanwhile, the highest value for DIFF with a negative value (-) shows that the scale of wavelet is suitable for the data of turning left.
ISSN : 0975-5462
Vol. 4 No.06 June 2012
2993
Endro Yulianto et al. / International Journal of Engineering Science and Technology (IJEST)
Table 1. Correlation of 22 volunteers (in %) SCALE 1 2 3 4 5 . . . . 195 196 197 198 199 200 MAX MIN
RIGHT 53.091 54.182 54.727 54.727 52.727 . . . . 12.364 12.545 14 12.727 13.091 13.818 54.727 0
PEAK MAX LEFT 55.273 55.818 55.273 56.909 54.182 . . . . 13.273 14.545 14.909 14.909 14.545 17.091 56.909 0
DIFF -2.182 -1.636 -0.546 -2.182 -1.455 . . . . -0.909 -2 -0.909 -2.182 -1.454 -3.273 3.637 -4
RIGHT 55.636 56.364 56.545 54.364 56.545 . . . . 14 14 15.273 15.273 16.727 15.273 56.545 0
PEAK MIN LEFT 53.636 55.455 56.364 56.364 56.909 . . . . 10.364 14.727 13.273 14.727 14.909 16.545 56.909 0
DIFF 2 0.909 0.181 -2 -0.364 . . . . 3.636 -0.727 2 0.546 1.818 -1.272 4 -3.273
RIGHT 67.091 70.364 74.909 72.727 72.909 . . . . 25.455 25.818 27.818 26.727 29.273 28.545 74.909 0.18182
PEAK MAX + MIN LEFT 66.545 69.455 71.818 72.364 71.273 . . . . 22.727 28.364 27.455 28.182 28.545 31.455 72.364 0.18182
DIFF 0.546 0.909 3.091 0.363 1.636 . . . . 2.728 -2.546 0.363 -1.455 0.728 -2.91 4 -4.182
Correlation to Mexican Hat as seen in Table 1 is performed to 22 volunteers for all positions of electrode C3, C4, P3 and P4. The correlation afterwards is also performed to 4 volunteers separately. The following is the table showing a MAX and MIN values from all positions of electrodes to 22 volunteers and 4 volunteers. Table 2. Correlation of 22 volunteers and 4 Volunteers for the MAX Value (in %)
C3
C4
P3
P4
22 VOLUNTEER VOLUNTEER 1 VOLUNTEER 2 VOLUNTEER 3 VOLUNTEER 4 22 VOLUNTEER VOLUNTEER 1 VOLUNTEER 2 VOLUNTEER 3 VOLUNTEER 4 22 VOLUNTEER VOLUNTEER 1 VOLUNTEER 2 VOLUNTEER 3 VOLUNTEER 4 22 VOLUNTEER VOLUNTEER 1 VOLUNTEER 2 VOLUNTEER 3 VOLUNTEER 4
ISSN : 0975-5462
RIGHT 54.727 64 64 72 72 54.182 68 68 76 56 53.455 64 80 56 80 57.818 60 64 72 60
PEAN MAX LEFT 56.909 64 56 68 68 53.273 60 68 68 60 55.091 68 64 56 64 58.545 68 68 68 64
DIFF 3.637 12 28 20 20 5.091 32 16 32 24 5.454 24 20 32 32 5.818 28 24 20 24
RIGHT 56.545 68 60 72 64 55.455 68 64 72 60 55.636 60 68 64 64 57.636 76 60 68 68
PEAK MIN LEFT 56.909 68 60 72 68 54.182 72 60 72 72 55.273 68 56 72 64 56.182 72 52 72 64
Vol. 4 No.06 June 2012
DIFF 4 16 24 36 20 5.091 40 24 28 20 7.091 20 40 24 24 6 28 28 32 12
PEAK MAX + MIN RIGHT LEFT DIFF 74.909 72.364 4 84 72 12 72 76 20 88 84 32 84 84 16 69.091 69.636 7.091 84 72 28 72 84 16 84 80 28 68 80 20 72.364 70.182 5.637 80 80 24 92 80 32 68 76 28 92 76 20 73.636 71.818 6.728 84 84 32 76 72 28 84 88 28 76 72 16
2994
Endro Yulianto et al. / International Journal of Engineering Science and Technology (IJEST)
Table 3 Correlation of 22 Volunteers and 4 Volunteers for the MIN Value (in %)
C3
C4
P3
P4
22 VOLUNTEER VOLUNTEER 1 VOLUNTEER 2 VOLUNTEER 3 VOLUNTEER 4 22 VOLUNTEER VOLUNTEER 1 VOLUNTEER 2 VOLUNTEER 3 VOLUNTEER 4 22 VOLUNTEER VOLUNTEER 1 VOLUNTEER 2 VOLUNTEER 3 VOLUNTEER 4 22 VOLUNTEER VOLUNTEER 1 VOLUNTEER 2 VOLUNTEER 3 VOLUNTEER 4
RIGHT 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
PEAN MAX LEFT 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.18182 0 0 0 0
DIFF -4 -24 -28 -16 -24 -5.272 -16 -24 -24 -28 -3.455 -20 -16 -12 -20 -6 -16 -28 -24 -28
RIGHT 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
PEAK MIN LEFT 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
DIFF -3.273 -24 -24 -20 -28 -4.909 -24 -24 -16 -28 -4.181 -36 -24 -20 -20 -5.818 -24 -20 -28 -24
PEAK MAX + MIN RIGHT LEFT DIFF 0.18182 0.18182 -4.182 4 4 -24 4 4 -24 4 4 -16 4 4 -36 0.18182 0.18182 -7.272 4 4 -32 4 4 -32 4 4 -16 4 4 -24 0.18182 0.18182 -3.272 4 4 -32 4 4 -20 4 4 -20 4 4 -16 0.18182 0.18182 -6.182 4 4 -32 4 4 -28 4 4 -36 4 4 -28
Table 2 and Table 3 show that the value of DIFF for the correlation of Mexican Hat towards the testing data of “turning right” and “turning left” from 22 volunteers simultaneously with an equal scale will result in a little percentage value. Meanwhile, the result of data correlation for each volunteer shows a higher value of DIFF. This then shows that each volunteer has a different scale. 5. Conclusion From this research, several conclusions could be taken: 1.
2. 3.
To obtain the feature of EEG signal for the motor movement in the form of ERD/ERS, the signal processing could be used that is in the form of centering, bandpass filter, signal correlation and eigen value decomposition now that ERD/ERS is hidden in the background signal and bioelectrical signal. ERD/ERS as a result of a motor movement occurs in the area of motor cortex from brain. The result of Mexican Hat correlation shows that each volunteer has a different scale of wavelet transform in order to obtain a high correlation value, even for the motor movement.
6. References [1] [2]
[3] [4]
[5] [6] [7]
Atry. F, Amir H. O, S. Kamaledin. S, Model Based EEG Signal Purification to Improve the Accuracy of the BCI Systems, Control and Intelligent Processing Centre of Excellence, ECE Department, Faculty of Engineering, University of Tehran, Tehran, Iran Bai .O, Peter . L, Sherry .V, Mary .K.F, Noriaki .H, Mark .H, 2008, A high performance sensorimotor beta rhythm-based brain–computer interface associated with human natural motor behavior, IOP Publishing, Journal of Neural Engineering 5 (2008), 24-35, doi:10.1088/17412560/5/1/003 Berger. H, Jena, R. Caton, Elektroensefalograf EEG History, Biocybernaut Institute Mountain View, California. Cortes .J.M.R, V.A. Aquino, G.R. Cholula, P.G. Gil, J.E. Ambrosio, 2010, P-300 Rhythm Detection Using ANFIS Algorithm and Wavelet Feature Extraction in EEG Signals, Proceedings of the World Congress on Engineering and Computer Science Vol 1, WCECS, San Francisco, USA Darvishi .S and A.A. Ani, 2007, Brain-Computer Interface Analysis using Continous Wavelet Transform and Adaptive Neuro-Fuzzy Classifier, Proceedings of the 29th Annual International Conference of the IEEE EMBS Cite’ Internationale, Lyon, France Faradji .F, R.K. Ward and G.E. Birch, ” A Simple Approach to Find the Best Wavelet Basis in Classification Problem”, International Conference on Pattern Recognation, 1051-4651/10 IEEE, DOI 10.1109/ICPR.2010.162, 2010 Gabriel .J.F, ”Fisika Kedokteran, Buku Kedokteran ”, EGC, 1996
ISSN : 0975-5462
Vol. 4 No.06 June 2012
2995
Endro Yulianto et al. / International Journal of Engineering Science and Technology (IJEST)
[8]
[9] [10]
[11]
[12] [13] [14] [15] [16] [17] [18] [19]
Gareis .I, G. Gentiletti, R. Acevedo, L. Rufiner, 2010, Feature Extraction on Brain Computer Interface using Discrete Dyadic Wavelet Transform: Preliminary Result, Reseach Center for Signals, Systems and Computational Intelligence, Journal of Physics : Conference Series (IOP) Guyton .A.C and J.E. Hall ,” Buku Ajar Fisiologi Kedokteran,” 9th Edition, Buku Kedokteran ,EGC, 1997 Huang. D, Kai. Q, Simon .O, Ding-Yu .F, Ou .B, 2011, Event-Related Desynchronization / Synchronization-Based Brain-Computer Interface towards Volitional Cursor Control in a 2D Center-Out Paradigm, Institute of Electronics and Electrical Engineers (IEEE) Symposium Series on Computational Intelligence. Cognitive Algorithms, Mind and Brain. Conference Proceedings. Pgs. 151-158 Kooi . O.V.D, 2010, Identifying Individuals using Event Related Synchronization and Desynchronization, 13thTwente Student Conference on IT, Enschede, The Netherlands. Copyright 2010, University of Twente, Faculty of Electrical Engineering, Mathematics and Computer Science. Matlab, Version 7.0.1.24704 (R14) Service Pack1, 2004 Neuper.C, Michael. W, Gert. P, 2006, ERD/ERS Patterns Reflecting Sensorimotor Activation and Deactivation, Neuper & Klimesch (Eds.) Progress in Brain Research, Vol. 159 ISSN 0079-6123 Ochoa .J.B, G.G. Molina, T. Ebrahimi, “EEG Signal Classification for Brain Computer Interface Application”, Ecole Polytechnique Federale De Lausanne, 2002 Parra .L and P. Sadja, “Blind Source Separation via Eigen Value Decompisition”, Journal of Machine Learning Research 4, 2003 Polikar .R, “The Wavelet Tutorial”, Dept. of Electrical and Computer Engineering, Rowan University., 1996 Sanei . S and J.A. Chambers, 2007, EEG Signal Processing, Centre of Digital Signal Processing, Cardiff University, UK, John Wiley & Sons, Ltd. Wasilewski. F,” PyWavelets-DiscreteWavelet Transform in Python”, Copyright 2006 – 2011 Products K&H MFG. CO., LTD Designed by Dtell, Copyright © 2010 (2010, December,29th) Available: www.kandh.com.tw/products
Born in Banyubiru, 17 July 1976, Diploma of Academy of Electro-medical Engineering of Health Department, Indonesia, 1997, Graduate of Nuclear Engineering of Universitas Gadjah Mada, Yogyakarta, Indonesia, 1999, Master in Institut Teknologi Sepuluh Nopember Surabaya, Indonesia, 2004, From 2007 – present the candidate of Doctor in the Electronics Engineering of Engineering Faculty, Universitas Gadjah Mada, Yogyakarta, Indonesia.
Dr. Adhi Susanto, born in Banjar Indonesia, 1940 obtained M.Sc (1966) and PhD (1988) from University of California Davis. He is a professor at Electronics Engineering Department andInformation Technology of Engineering Faculty, Universitas Gadjah Mada, Yogyakarta, Indonesia. The current research field is Electronics Engineering, Image Processing, Signal Processing, Adaptive System, Classification and Pattern Recognition Techniques. He is also a member of IEEE and Planetary Society.
Dr. Thomas Sri Widodo, born in Klaten, Indonesia, 1950. A professor in Electronics and Technology and Information Department, Engineering Faculty, Universitas Gadjah Mada, Yogyakarta, Indonesia. Dipl. Ing. ENSERG, Grenoble, France, 1985, Doctorate d’Etudes Approfondies, Univ. Montpellier 2, France, 1986 Docteur de l’Universite, Univ. Montpellier2, France, 1988. The current research field is system, signal and electronics
Dr. Samekto Wibowo, Medical Faculty University of Gadjah Mada, Yogyakarta, Indonesia, 1969, Master in Universitas Gadjah Mada Yogyakarta in 1975, Doctor from Universitas Gadjah Mada, Yogyakarta, Indonesia in 1998. He is a professor in Neurology of Medical Faculty of Universitas Gadjah Mada, Yogyakarta, Indonesia. The research field is about neurology.
ISSN : 0975-5462
Vol. 4 No.06 June 2012
2996