Printer Friendly

EEG signal classification for brain computer interface using SVM for channel selection.

Introduction

The technology that helps paralyzed people to communicate with the external world is Brain Computer Interface (BCI). Electroencephalogram (EEG) based BCI is used to translate the changes in the brain signals into operative control signals. In the analysis of the brain signals, Motor Imagery (MI) is the state in which the depicted particular motor action is internally reactivated within the working memory without any overt motor output. It is governed by the principles of motor control [1]. Event-related desynchronization / Synchronization (ERD / ERS) patterns are the measurable changes made by Motor Imagery (MI) in the EEG signals.

In MI-based BCIs high inter subject and intra subject variability are resulted due to the time, frequency, and spatial non-stationarity of these patterns. One of the most effective algorithms for MI-BCI is based on common spatial pattern (CSP) technique [2] [3]. In BCI application, the proper selection of subject specific frequency bands plays a vital role for the success of CSP. For automatically choosing the optimal frequency band, in the literature, common sparse spectral spatial pattern (CSSSP) [4] sub band CSP (SBCSP) [5]; Filter bank CSP (FBCSP) [6] and adaptive FBCSP [7] have been proposed.

To effectively choose the subject-specific features, the FBCSP [8] uses CSP features from a set of fixed band pass filters and feature selection algorithm based on mutual information. During the selection process features are selected from the relevant frequency components. The proposed method uses a subject-specific FB selection before feature extraction to enhance the accuracy of the FBCSP framework as the subject-specific frequency components carry distinct features. The core of a BCI, in which the EEG signals are mapped into the space of epochs, is Classification algorithm. They are then classified using decision functions learned on the training set composed of labeled signals. The Performance of classification depends on the choice of the pre-processing techniques. [9] A large training session is required to be beneficial to lay down the decision rules that allow the classification of the user's intention [10].

The energy distribution over uniform frequency sub bands given by the Fourier transform is an example of apriori choice of signal feature. In previous studies [11] [12], it had been proposed that the marginal of the Discrete Wavelet Transform (DWT) for feature extraction and the feature space were selected by optimizing the mother wavelet of the decomposition. The DWT marginal reflects the average signal intensity over dyadic sub bands. The dyadic decomposition is well suited to describe and discriminate signals whose discriminative information is mainly at low frequencies since the frequency resolution is higher for low frequencies than for high frequencies.

In this paper we propose to measure energy of specific motor imageries in the brain signal using Fast Hartley transform along with the Chebyshev filter and selecting the ideal channels for the classification problem using the proposed support vector machine. The resultant data obtained was classified using Boosted Decision tree. This paper is organized as follows. Section 2 describes the features of the data set used in this paper. Sections 3 and 4 describe the preprocessing techniques and the classification algorithms analyzed in this study respectively. Section 5 analyzes our results.

Dataset

Data set IV A dataset used in the brain computer interface competition provided by Intelligent Data Analysis Group [13] is used in this research. This data set consists of EEG recordings from five individuals. Visual cues indicated for 3.5 s which of the following 3 motor imageries the subject should perform: (L) left hand, (R) right hand, (F) right foot. The presentation of target cues was intermitted by periods of random length, 1.75 to 2.25 s, in which the subject could relax. Given are continuous signals of 118 EEG channels and markers that indicate the time points of 280 cues for each of the 5 subjects (aa, al, av, aw, ay). Subject aa was used in our study.

Preprocessing of EEG Signals

The regular Hartley transform's kernel is based on the cosine-and-sine function, defined [14] as

cas (vt) = cos(vt) + sin(vt).

Hartley transform compared to Fourier transforms is a real function. The Hartley transform pair can be defined as follows:

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

A very important property of Hartley Transform is its symmetry

H {f (t))} = H(v), H {H (t)} = f(v)

This has the advantage of using the same operation for computing the transform and its inverse. Another important feature is that the transform pairs are both real which provides good computational advantages for Hartley Transform (HT) over the Fourier transform (FT).

Many of the familiar complex relations in the Fourier domain have very similar counter parts in the Hartley domain. Let F (co) and H (v) be the FT and HT of a function f (t) the n it is to verify the following.

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

Where [real part],[imaginary part],[epsilon],o denote real, imaginary, even and odd parts. Other properties in the Hartley domain are:

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

HT's discrete formulation DHT is given by:

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

Which is applied to the discrete-time function x (n) with period N. The properties of the DHT are similar to those of the discrete Fourier transform (DFT) and Fast Hartley transform (FHT) [15] which is similar to the familiar Fast Fourier Transform (FFT). Some of the properties of DHT are listed:

af (n) + bg (n) [??] aF (k) + bG (k), f (-n)[??] F (-k),

Chebyshev filters are used to capture the energies in the 5-15 Hz band. This frequency band was selected primarily to capture the Beta waves in the EEG signal which is closely linked to motor behavior and is generally attenuated during active movements. Chebyshev filters are fast because they are carried out by recursion rather than convolution. The design of these filters is based on the z-transform.Once the energy for each channel is computed, the values of each channel are subjected to the proposed support vector machine algorithm with respect to the class to rank the importance of the channel. Scoring over 75 was selected in this work. Support Vector Machines (SVMs) are a set of related supervised learning methods. They are used to analyze data and recognize patterns and are extensively used for classification and regression analysis. The SVM algorithm is based on the statistical learning theory and the Vapnik-Chervonenkis (VC) dimension.

A support vector machine in a high dimensional space, is used to construct a hyperplane or a set of hyperplanes which can be used for classification, regression or other tasks. IThe hyperplane which provides the largest distance to the nearest training data points of any class is chosen, in general, as the larger the margin the lower the generalization error of the classifier. The data in the dimensional space may not be discriminated or linearly separable even though the initial problem is stated in a finite dimensional space. Mapping the original dimensional space to a higher dimensional space is solution to overcome this problem and thus making the separation easier.

Across the optimal hyperplane each channel is checked with respect to the class label. Error function is computed for each channel using

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

The error is minimized using

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

where C is the capacity constant, w is the vector of coefficients, b a constant and [[xi].sub.i] are parameters for handling non separable data (inputs). The index i label the N training cases.

Classification Algorithms

Boosting trees used in classification follow the principle of boosting methods applied in regression trees. Boosting trees compute a sequence of simple trees built in each stage based on the prediction residuals of the preceding tree. Hence boosting trees build many binary trees. In boosting trees additive weighted expansions produce excellent fit of the predicted values even if the dependent variable of interest is non linear in nature.

Experimental Results

An application was built using labview for the proposed method. The epoch occurring for a time period of 3.5 at a sampling rate of 100Hz was input to the application. The minimum, maximum and average energy was computed. Screen shots of output are shown in figure 1and figure 2.

[FIGURE 1 OMITTED]

[FIGURE 2 OMITTED]

The energies were computed for 118 EEG electrodes for 168 instances of motor imagery cues of right hand and right foot. The energy values from each electrode were used as attributes for predicting the class label. SVM was used to rank the channel and electrodes with ranking of over 75 was selected. The obtained attribute values were used as inputs to the boosting tree algorithm. The summary of the boosted tree, lift chart and gains chart is shown in appendix. The classification accuracy is shown in table I.

Conclusion

The energy from the preprocessed EEG epoch was extracted using a combination of Fast Hartley transform and Chebyshev filter. The energies from each channel was subject to SVM and ranked. Boosting tree was used to classify the selected attributes. The classification result so obtained was tabled in the previous section. The proposed technique gives good results in terms of classification keeping in mind the goal of reducing the preprocessing time. Further work need to be done to improve the classification accuracy to bridge the man-machine gap by understanding the human semantic factor

References

[1] N. Sharma, V.M. Pomeroy, and J. C. Baron, "Motor imagery: A backdoor to themotor imagery system after stroke?" Stroke, vol. 37, pp. 1941-1952, 2006.

[2] H.Ramoser, J. Muller-Gerking, and G. Pfurtscheller, "Optimal spatial filtering of single trial EEG during imagined hand movement," IEEE Trans. Rehabil. Eng., vol. 8, no. 4, pp. 441-446, Dec. 2000.

[3] C.Guger, H. Ramoser, and G. Pfurtscheller, "Real time EEG analysis with subject specific spatial patterns for a brain-computer interface," IEEE Trans. Rehabil. Eng., vol. 8, no. 4, pp. 447-456, Dec. 2000.

[4] G. Dornhege, B. Blankertz, M. Krauledat, F. Losch, G. Curio, and K.-R. Muller, "Combined optimization of spatial and temporal filters for improving brain-computer interface," IEEE Trans. Biomed. Eng., vol. 53,no. 11, pp. 2274-2281, Nov. 2006.]

[5] Q. Novi, C. Guan, T. H. Dat, and P. Xue, "Sub band common spatial pattern for brain-computer interface," in Proc. 3rd Int. Conf. Neural Eng.IEEE Eng. Med. Biol. Soc. (EMBS), May 2007, pp. 204-207.

[6] K. K. Ang, Z. Y. Chin, H. Zhang, and C. Guan, "Filter bank common spatial pattern (FBCSP) in brain-computer interface," in Proc. IEEE Int. Joint Conf. Neural Netw., Jun. 2008, pp. 2390-2397.

[7] P. T. Kavitha, C. Guan, C. T. Lau, and A. P. Vinod, "An adaptive filter bank for MI based brain-computer interface," in Proc. 30th Annu. Int. Conf. IEEE Eng. Med. Biol., Aug. 2008, pp. 1104-1107.

[8] K. K. Ang, Z. Y. Chin, H. Zhang, and C. Guan, "Filter bank common spatial pattern (FBCSP) in brain-computer interface," in Proc. IEEE Int. Joint Conf. Neural Netw., Jun. 2008, pp. 2390-2397.

[9] Denis Vautrin, Xavier Artusi, Marie-Franchise Lucas, and Dario Farina, A Novel Criterion of Wavelet Packet Best Basis Selection for Signal Classification with Application to Brain-Computer Interfaces, IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 56, NO. 11, NOVEMBER 2009.

[10] N. Birbaumer, A. R. Murguialday, and L. Cohen, "Brain-computer interface in paralysis," Curr. Opin. Neurol., vol. 21, no. 6, pp. 634-638, Dec. 2008.

[11] O. F. Do Nascimento and D. Farina, "Movement-related cortical potentials allow discrimination of rate of torque development in imaginary isometric plantar flexion," IEEE Trans. Biomed. Eng., vol. 55, no. 11, pp. 2675-2678, Nov. 2008.

[12] D. Farina, O. F. Do Nascimento, M. F. Lucas, and C. Doncarli, "Optimization of wavelets for classification of movement-related cortical potentials generated by variation of force-related parameters," J. Neurosci. Methods, vol. 162, no. 1, pp. 357-363, May 2007.

[13] (Klaus-Robert Muller, Benjamin Blankertz), and Campus Benjamin Franklin of the Charite-University Medicine Berlin, Department of Neurology, Neurophysics Group (Gabriel Curio).

[14] Hartley Transform and its Applications Saied Hosseini Khayat_ Electrical Engineering Department Ferdowsi University of Mashhad, Iran.

[15] R. N. Bracewell, "The fast Hartley transform," Proc. IEEE 72 (8), 1010-1018 (1984).

V. Baby Deepa (1), P. Thangaraj (2) and S. Chitra (3)

(1) Department of Software Engineering, M. Kumarasamy College of Engineering, Karur, Tamil Nadu, India

(2) Department of Computer Science & Engineering, Bannari Amman Institute of Technology, Sathyamangalam, Erode, India

(3) Vice Principal, M. Kumarasamy College of Engineering, Karur, Tamil Nadu, India E-mail: deepamct@gmail.com; schitra3@gmail.com
Table I: Individual class label classification accuracy.

 Predicted hand Predicted foot

Label - Hand 35 45
Prediction % 43.75% 56.25%
Label - Foot 17 71
Prediction % 19.32% 80.68%
Hand & Foot 52 116
Total Percent 30.95% 69.05%
COPYRIGHT 2011 Research India Publications
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 2011 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Author:Deepa, V. Baby; Thangaraj, P.; Chitra, S.
Publication:International Journal of Computational Intelligence Research
Date:Apr 1, 2011
Words:2106
Previous Article:Information fusion based on multiplicity of data preprocessing boosts the AdaBoost classifier.
Next Article:A novel programming approach for DNA computing.
Topics:

Terms of use | Privacy policy | Copyright © 2020 Farlex, Inc. | Feedback | For webmasters