Welcome to Scribd, the world's digital library. Read, publish, and share books and documents. See more
Standard view
Full view
of .
Look up keyword
Like this
0 of .
Results for:
No results containing your search query
P. 1
06 Paper 30041335 IJCSIS Camera Ready Pp. 39-46

06 Paper 30041335 IJCSIS Camera Ready Pp. 39-46

Ratings: (0)|Views: 10|Likes:
Published by ijcsis

More info:

Published by: ijcsis on Jul 02, 2013
Copyright:Attribution Non-commercial


Read on Scribd mobile: iPhone, iPad and Android.
download as PDF, TXT or read online from Scribd
See more
See less





(IJCSIS) International Journal of Computer Science and Information Security,Vol. 11, No. 5, May 2013
A Review Based on Function Classification of EEGSignals
Rajesh Singla
Dept. of Instrumentation and Control EngineeringDr. B. R Ambedkar National Institute of TechnologyJalandhar, Indiarksingla1975@gmail.com
Neha Sharma
Dept. of Electrical EngineeringDAV Institute of Engineering and TechnologyJalandhar, Indianeha.nitj@gmail.com
Navleen Singh Rekhi
Dept. of Electronics and Communication EngineeringDAV Institute of Engineering and TechnologyJalandhar, Indianavleenr@yahoo.com
— For Electroencephalography (EEG) based BCI, motorimagery is considered as one of the most effective ways .Thispaper presents review on the results of performance measures of different classification algorithms for brain computer interfacebased on motor imagery tasks such as left hand, right hand, footand wrist moment . Based on the literature, we give a brief comparison of accuracy of various classifications algorithms interms of their certain properties consisting of feature extractiontechniques which involves FBCSP, CSP, ICA, Wavelets etc andclassifiers such as SVM, LDA, ANN.
 Keywords-BCI; EEG; Wavelet Transform; LDA; SVM; NN 
A Brain-Computer Interface (BCI) is a communicationsystem capable of transforming the person’s cognitivefunctions into control commands that let the user interact withexternal devices [64], [65]. The basic operation of a BCI is torecord the cerebral bioelectric activity through electrodes inorder to differentiate between several mental tasks. This kind of systems creates a natural way of human-machinecommunication because they translate intentions into orders tointeract with the environment without performing any physicalmovement. Thus, the BCI systems are of great interest topeople with severe disabilities or mobility limitations. They canimprove their quality of life and assist them in various dailytasks.A BCI is divided in different modules: preprocessing,feature extraction, classification and feedback. Various signalsare used in BCI systems, but our experiences were based inEEG signals, which can vary in time. Therefore, adaptationmodules like feature extraction or/and classification is a veryimportant issue in BCI research Among these approaches, inorder to effectively extract the components of differentfrequency bands from EEG recordings, a well-designed filter isgenerally needed in BCI system, which is one of the importantissues for the classification performance of EEG signals in BCIsystem [15]. The traditional filters such as Butterworth filterand FIR based on window functions could not adapt to thecharacteristics of EEG data flexibility. Thus, it is necessary todevelop more effective filtering method and technique forimproving the accuracy of classification for intentionalactivities.Electroencephalographic (EEG) activity has been discussedin relation with functional neuronal mechanisms. In this regard,it is of major interest to investigate how EEG changes duringpathological or physiological brain states or by external andinternal stimulation [44].The ongoing electroencephalographic signals (EEG)contain information associated to movements, mental tasks ormental responses related to some stimuli. These signals areanalyzed and processed through several mathematicaltechniques to extract useful information represented in the formof feature vectors, which are then translated into meaningfulcontrol commands. An important purpose of a direct BCI is toallow individuals with motor disabilities such as locked-insyndrome, which can be caused by amyotrophic lateralsclerosis, high-level spinal cord injury, or some other severehealth conditions, to have some control over external devices[46].The goal of this paper is to review of classificationalgorithms used for BCI, their properties and their evaluation.The outline of the paper is as follows: section 2 depicts abrief description of the pattern recognition system andemphasizes the role of classification. Section 3 surveys theclassification algorithms used for BCI and finally, section 4concludes the study.II.
Feature Extraction
The original EEG signals potentials recorded from the scalpare very complex so they are needed to be processed and
39http://sites.google.com/site/ijcsis/ISSN 1947-5500
(IJCSIS) International Journal of Computer Science and Information Security,Vol. 11, No. 5, May 2013
desired components are needed to be extracted for furthercontrolling of devices.
In autoregressive (AR) techniques, a model iscreated where a current voltage can be predicted from N pastvoltages where the model order is N [60]. Thus the model canbe represented as:
 N ieieiei
i xa x
is the ith order AR coefficient for electrode e.These AR coefficients can be used as features. To obtain thesecoefficients, EEG data is generally windowed into blocks of data with more than N samples. Then, as the value of t isshifted through the window of data, we obtain numerous modelequations which allow us to compute optimum AR coefficients.Thus, these AR coefficients can be used to represent the mentalstate during that window of time.
To date, little has been published usingwavelets as a feature extraction method for a BCI system.However, they have been used in a variety of other EEGpattern recognition work [50, 51] including neural networks[52,53]. Wavelets are essentially a compromise between time-domain and frequency-domain since they allow the user toview change in frequency bands over time (with less resolutionthan just time-domain or frequency-domain). The DiscreteWavelet Transform (DWT) can be computed as a series of filters. To date, little has been published using wavelets as afeature extraction method for a BCI system. However, theyhave been used in a variety of other EEG pattern recognitionwork, including neural networks. Wavelets are essentially acompromise between time-domain and frequency-domain sincethey allow the user to view change in frequency bands overtime (with less resolution than just time-domain or frequency-domain).
Common Spatial Filter:
Common spatial patterns (CSP)method was firstly suggested for classification of multi-channelEEG during imagery hand movements by Ramoser et al.[41].The main idea is to use a linear transform to project the multi-channel EEG data into a low-dimensional spatial subspace witha projection matrix, of which each row consists of weights forchannels. This transformation can maximize the variance of two-class signal matrices. CSP method is based on thesimultaneous diagonalization of the covariance matrices of both classes.
Experimental results suggested that ICA is a usefuland feasible method for spatial filtering and feature extractionin motor imagery based multi-class BCIs. When using EEGrecordings as the input signals of a BCI system, the researchermay face a problem of extracting features used forclassification in the presence of artifacts such aselectrooculogram (EOG) or electromyogram (EMG). Theamplitude of the disturbances may be higher than that of brainsignals. This requires an efficient method to separate brainsignals from artifacts. ICA happens to be a suitable approach tocarry out the separation. This approach is based on theassumption that the brain activity and the artifacts areanatomically and physiologically separate processes, and thisseparation is reflected in the statistical independence betweenthe electrical signals generated by those processes [16].
Classification Algorithms
The original EEG signals potentials recorded from the scalpare very complex so they are needed to be processed anddesired components are needed to be extracted for furthercontrolling of devices.
: LD classifier is one of the linear classificationmethods that require fewer examples in order to obtain areliable classifier output [59] It is also a simpler andcomputationally attractive as compared to other classifiers
LDwas used to classify different combinations of mental.
: An SVM also uses a discriminant hyper plane toidentify classes[56]. However, concerning SVM, the selectedhyper plane is the one that maximizes the margins, i.e., thedistance from the nearest training points. Maximizing themargins is known to increase the generalizationcapabilities[56]. As RFLDA, an SVM uses a regularizationparameter C that enables accommodation to outliers andallows errors on the training set. Such an SVM enablesclassification using linear decision boundaries and is known aslinear SVM. This classifier has been applied, always withsuccess, to a relatively large number of synchronous BCIproblems[57,58]. However, it is possible to create nonlineardecision boundaries, with only a low increase of theclassifier’s complexity, by using the ‘kernel trick’. It consistsin implicitly mapping the data to another space, generally of much higher dimensionality, using a kernel function K(x, y).The kernel generally used in BCI research is the Gaussian orradial basis function (RBF).
 Neural Networks:
Neural networks (NN) are, togetherwith linear classifiers,[55] the category of classifiers mostlyused in BCI research. Let us recall that an NN is an assemblyof several artificial neurons which enables us to producenonlinear decision boundaries.
The k-nearest neighbor (k-NN) [54] is a classifierthat assigns the class label of a new data based on the classwith the most occurrences in a set of k nearest training datapoints usually computed using a distance measure such as theEuclidean distance.
 Multilayer Perception:
An MLP is composed of severallayers of neurons: an input layer, possibly one or severalhidden layers and an output layer. Each neuron’s input isconnected with the output of the previous layer’s neuronswhereas the neurons of the output layer determine the class of the input feature factor.
40http://sites.google.com/site/ijcsis/ISSN 1947-5500
(IJCSIS) International Journal of Computer Science and Information Security,Vol. 11, No. 5, May 2013
Neural networks and thus MLP are universalapproximators, i.e., when composed of enough neurons andlayers, they can approximate any continuous function. The factthat they can classify numerous classes makes NN very flexibleclassifier that can adapt to a great variety of problems.Consequently, MLP, which are the most popular NN used inclassification, have been applied to almost all BCI problemssuch as binary [46] or multiclass synchronous [48] orasynchronous [49] BCI. However, the fact that MLP areuniversal approximators makes these classifiers sensitive toovertraining, especially with such noisy and non-stationary dataas EEG, e.g., [47]. Therefore, careful architecture selection andregularization is required.
K-nearest neighbours:
The aim of this technique is toassign to an unseen point the dominant class among its k nearest neighbors within the training set [61]. For BCI, thesenearest neighbors are usually obtained using a metric distance.With a sufficiently high value of k and enough trainingsamples, kNN can approximate any function which enables itto produce nonlinear decision boundaries.KNN algorithms are not very popular in the BCIcommunity, probably because they are known to be verysensitive to the curse-of-dimensionality which made them failin several BCI experiments [42].
 Mahalanobis distance:
Mahalanobis distance basedclassifiers assume a Gaussian distribution N (
c,Mc) for eachprototype of the class c. Then, a feature vector x is assigned tothe class that corresponds to the nearest prototype, according tothe so-called Mahalanobis distance dc(x)[62]. This leads to asimple yet robust classifier, which even proved to be suitablefor multiclass or asynchronous BCI systems [62].
Pre-processing Features Classification Accuracy (%) References
Finger-The BCICompetition IIIdataset IVaFilter Bank CommonSpatial Pattern(FBCSP)NBPWFLDSVM90.3±0.7%89.9±0.9%90.0±0.8%[7]]Finger-ondifferent dataFilter Bank CommonSpatial Pattern(FBCSP)NBPWFLDSVM81.1±2.2%80.9±2.1%81.1±2.2%[7]Muscle/ Data set Iof BCICompetition IIIBand Pass(8-30Hz)CSP FDA 90% [9]facial functions FBCSP decisionthreshold-basedclassifier87.1±0.76% [11]ECoG signal CSP SVM 90% [21]LDA 82%Discrimination b/wwrist and fingerICA BD MD 65 % [26]ANN 71 % [26]
are LEFT
Protocol Preprocessing Features Classification Accuracy (%) References
On differentEEG data AARparameters,logarithmic BPestimates andtheconcatenationof bothadaptivequadratic andlineardiscriminantanalysisaccuracy of 72%for a two targettask and 45%for a fourtarget task,within 10minutes.[1]BCI competitionIII dataset IVaCSP SVM 90%, [9]On differentEEG data1-40 Hzband-passfilterICA (LDA) 89.52 [10]
41http://sites.google.com/site/ijcsis/ISSN 1947-5500

You're Reading a Free Preview

/*********** DO NOT ALTER ANYTHING BELOW THIS LINE ! ************/ var s_code=s.t();if(s_code)document.write(s_code)//-->