Paper Info
Title
A novel data reduction method: Distance based data reduction and its application to classification of epileptiform EEG signals
Abstract
Results To validate and test the proposed data reduction, the classification accuracy, sensitivity, and specifity analysis, computation time, 10-fold cross-validation, and 95% confidence intervals have been used in this study. Six different combined methods have been used to classify the epileptiform EEG signal. These methods are (i) combining DFT and C4.5 decision tree classifier (DCT), (ii) combining DFT, distance based data reduction, and C4.5 DCT, (iii) combining AR and C4.5 DCT, (iv) combining AR, distance based data reduction, and C4.5 DCT, (v) combining DWT and C4.5 DCT, and (vi) combining DWT, distance based data reduction, and C4.5 DCT. The classification accuracies and computation times obtained by these methods are 99.02% – 79 s, 99.12% – 47 s, 99.32% – 65 s, 98.94% – 45 s, 92.00% – 52.06 s, and 89.50% – 29.9 s. Conclusions These results have shown that the proposed distance based data reduction method has produced very promising results with respect to both classification accuracy and computation time for classifying the epileptiform EEG signals. Also, proposed hybrid systems can be used to detect the epileptic seizure. Keywords EEG signals Distance based data reduction AR spectral analysis Discrete Fourier transform Discrete wavelet transform (DWT) C4.5 decision tree classifier Epileptic seizure 1 Introduction The electroencephalogram (EEG) signal is widely used clinically to investigate brain disorders. The study of the brain electrical activity, through the electroencephalographic records, is one of the most important tools for the diagnosis of neurological diseases [1–3] . Large amounts of data are generated by EEG monitoring systems for electroencephalographic changes, and their complete visual analysis is not routinely possible. Computers have long been proposed to solve this problem and thus automated systems to recognize electroencephalographic changes have been under study for several years [4–7] . There is a strong demand for the development of such automated devices, due to the increased use of prolonged and long-term video EEG recordings for proper evaluation and treatment of neurological diseases and prevention of the possibility of the analyst missing (or misreading) information [6,8] . In this paper, we have proposed a new data reduction method called distance based data reduction to reduce the number of new data points obtained when AR or DFT spectral analysis techniques were applied to raw epileptiform EEG signals. The entire process used to classify epileptiform EEG signals can generally be divided into a number of disjoint processing subsystems: preprocessing, feature extraction, data selection (reduction), and classification ( Fig. 1 ). Signal/image acquisition, artifact removing, averaging, thresholding, signal/image enhancement, and edge detection are the main operations in the course of preprocessing. The accuracy of signal/image acquisition is of great importance since it contributes significantly to the overall classification result. The markers are subsequently processed by the feature extraction module. The module of data selection is an optional stage, whereby the data vector is reduced in size including only, from the classification viewpoint, what may be considered as the most relevant data required for discrimination. The classification module is the final stage in automated diagnosis. It examines the input feature vector and based on its algorithmic nature produces a suggestive hypothesis [9] . In the literature, various studies have been considered related with classifying the EEG signals. While Guler et al. [23] obtained 96.79% classification accuracy using recurrent neural networks to detect the epileptic seizure from EEG signals, Kannathal et al. [24] achieved 95% classification accuracy using ANFIS classifier. Subasi obtained 95% and 93.6% classification accuracies using combination wavelet transform and mixture of experts and combination wavelet transform and multilayer perceptron neural network, respectively [10] ; 98.68% accuracy was obtained using combination wavelet transform and ANFIS classifier by Guler et al. [14] . In the last work of Guler et al., 95.53% and 98.60% classification accuracies were achieved by combining eigenvector methods – mixture of experts (ME) and modified mixture of experts (MME), respectively [9] . To validate and test the proposed data reduction method, the classification accuracy, sensitivity and specifity analysis, computation time, 10-fold cross-validation, and 95% confidence intervals have been used in this study. Four different combined methods have been used to classify the epileptiform EEG signal. These methods are (i) combining DFT and C4.5 decision tree classifier (DCT), (ii) combining DFT, distance based data reduction, and C4.5 DCT, (iii) combining AR and C4.5 DCT, (iv) combining AR, distance based data reduction, and C4.5 DCT, (v) combining DWT and C4.5 DCT, and (vi) combining DWT, distance based data reduction, and C4.5 DCT. Also, hybrid systems proposed can be used to detect the epileptic seizure. We have compared the used spectrum analysis methods according to the power spectrum densities obtained from AR, DFT, and DWT spectrum analysis methods. 2 Material and method 2.1 Data selection and recording We used the publicly available data described in [11] . In this section, we restrict ourselves to only a short description and refer to [11] for further details. The complete dataset consists of five sets (denoted A–E) each containing 100 single channel EEG segments. These segments were selected and cut out from continuous multi-channel EEG recordings after visual inspection for artifacts, e.g., due to muscle activity or eye movements. Sets A and B consisted of segments taken from surface EEG recordings that were carried out on five healthy volunteers using a standardized electrode placement scheme. Volunteers were relaxed in an awake state with eyes open (A) and eyes closed (B), respectively. Sets C, D, and E originated from EEG archive of presurgical diagnosis. EEGs from five patients were selected, all of whom had achieved complete seizure control after resection of one of the hippocampal formations, which was therefore correctly diagnosed to be the epileptogenic zone. Segments in set D were recorded from within the epileptogenic zone, and those in set C from the hippocampal formation of the opposite hemisphere of the brain. While sets C and D contained only activity measured during seizure free intervals, set E only contained seizure activity. Here, segments were selected from all recording sites exhibiting ictal activity. All EEG signals were recorded with the same 128-channel amplifier system, using an average common reference. The data were digitized at 173.61 samples per second using 12 bit resolution. Band-pass filter settings were 0.53–40 Hz (12 dB/oct). In this study, we used two datasets (A and E) of the complete dataset. Typical EEGs are depicted in Fig. 2 [10] . 2.2 Proposed method In this study, spectral analysis methods including AR, DFT, or DWT, distance based data reduction method, and C4.5 decision tree classifier have been combined to classify epileptiform EEG signals. The block diagram of the combined method is given in Fig. 3 . The proposed method has been explained in the following subsections. 2.3 The spectral analysis methods for EEG signals: feature extraction process We have used the AR, DFT, and DWT based spectral analysis methods for detection of epileptic seizure using EEG signals. We explain the used methods as follows. 2.3.1 Spectral analysis of EEG signals and welch method Welch method of power spectrum estimation was applied on the EEG data. Acquired EEG data were grouped in frames of 128 data points and the method was applied on these frames. Welch’s method is one among the classical methods of spectrum estimation based on FFT. FFT based Welch method is defined as a classical (non-parametric) method. It is made the second modification of periodogram spectral estimator, which is to window data segments prior to computing the periodogram [12,13] . If available information on the signal consists of the samples { x ( n ) } n = 1 N , the periodogram spectral estimator is given by (1) P ^ PER ( f ) = 1 N ∑ n = 1 N x ( n ) exp ( - j 2 π fn ) 2 . Here, P ^ PER ( f ) is the estimation of periodogram. In the Welch method, signals are divided into overlapping segments, each data segment is windowed, periodograms are calculated and then the average of periodograms is found. { x l ( n )}, l = 1, … , S are data segments and each segment’s length equals M . Note that the overlap is often chosen to be 50%. The Welch spectrum estimate is given by (2) P ^ w ( f ) = 1 S ∑ l = 1 s P ^ l ( f ) and P ^ l ( f ) = 1 M 1 P ∑ n = 1 M v ( n ) x l ( n ) exp ( - j 2 π fn ) 2 , where P ^ l ( f ) is the periodogram estimate of l TH segment, v ( n ) is the data-window, P is the total average of v ( n ) and given as P = 1 / M ∑ n = 1 M | v ( n ) | 2 , P ^ w ( f ) is the Welch PSD estimate, M is the length of each signal segment and S is the number of segments. Then, evaluation of P ^ w ( f ) at the frequency samples basically requires the computation of the following discrete Fourier transform (DFT): (3) X ( k ) = ∑ n = 1 N x ( n ) exp - j 2 π N nk , k = 0 , … , N - 1 , where X ( k ) is expressed as the discrete Fourier coefficient, N is the length of available data and x ( n ) is the input signal on the time domain. The procedure that computes Eq. (3) is called as FFT algorithm. The Welch PSD can be efficiently computed by the FFT algorithm. Variance of an estimator is one of the measures often used to characterize its performance. For 50% overlap and triangular window, variance for the Welch method is given by (4) var ( P ^ w ( f ) ) = 9 8 S var ( P ^ l ( f ) ) , where P ^ w ( f ) is the Welch PSD estimate and P ^ l ( f ) is the periodogram estimate of each signal interval [12–14] . 2.3.2 Spectral analysis of EEG signals and AR method Based on the AR model, the current value of an EEG signal x ( n ) can be described by a linear combination of previous values of the same EEG signal and a white noise input [15] . Many discrete-time signals encountered in practice may be presented using rational transfer function models. The AR model of order p is defined as follows: (5) xn = - ∑ k = 1 p a k x n - k + e ( n ) , where x n = output sequence (EEG signal), a k = parameters of the model, e ( n ) = driving signal (white noise process). A linear filter based on the above relationship can be defined in the Z -domain. The output power of filter evaluated on the unit circle, driven by a white noise of zero mean and variance s 2 , is equal to the power spectral density of the process x n : (6) P ( f ) = s 2 Δ t 1 + ∑ k = 1 p a k exp ( - j 2 π kf Δ t ) 2 , where Δ t denotes the sampling period, f the frequency, model parameters, p the model order and s 2 the total squared error of the model [16] . Among several methods of estimation of the AR model parameters (Yule Walker equation, Burg algorithm, least squares algorithm), the Burg method for estimating the AR parameters is used in this study. The Burg AR method is based on the minimization of the forward and backward prediction errors and on the estimation of the reflection coefficient [17] . In this study for spectral analysis, AR modeling on the EEG data grouped in frames of 128 data points. One of the most important aspects of the use in model-based methods is the selection of the model order. Much work has been done by various investigators on this problem and many experimental results have been given in the literature [17,18] . There are a number of criteria to optimize the AR model order. One of the better known criteria for selecting the model order has been proposed by Akaike [19] , called the Akaike information criterion (AIC), which is based on selecting the order that minimizes (7) AIC ( p ) = ln σ ⌢ 2 + 2 p / N , where σ ⌢ 2 is the estimated variance of the linear prediction error. In this study, model orders of the AR methods were taken as 25 by using Eq. (7) . 2.3.3 Spectral analysis of EEG signals and DWT method Wavelet transform is a spectral estimation technique in which any general function can be expressed as an infinite series of wavelets. The basic idea underlying wavelet analysis consists of expressing a signal as a linear combination of a particular set of functions (wavelet transform, WT), obtained by shifting and dilating one single function called a mother wavelet. The decomposition of the signal leads to a set of coefficients called wavelet coefficients. Therefore, the signal can be reconstructed as a linear combination of the wavelet functions weighted by the wavelet coefficients. To obtain an exact reconstruction of the signal, adequate number of coefficients must be computed. The key feature of wavelets is the time-frequency localization. It means that most of the energy of the wavelet is restricted to a finite time interval. Frequency localization means that the Fourier transform is band limited. When compared to STFT, the advantage of time-frequency localization is that wavelet analysis varies the time-frequency aspect ratio, producing good frequency localization at low frequencies (long time windows), and good time localization at high frequencies (short time windows). This produces a segmentation, or tiling of the time-frequency plane that is appropriate for most physical signals, especially those of a transient nature. The wavelet technique applied to the EEG signal will reveal features related to the transient nature of the signal which are not obvious by the Fourier transform. In general, it must be said that no time-frequency regions but rather time-scale regions are defined [10] . All wavelet transforms can be specified in terms of a low-pass filter g , which satisfies the standard quadrature mirror filter condition (8) G ( z ) G ( z - 1 ) + G ( - z ) G ( - z - 1 ) = 1 , where G ( z ) denotes the z -transform of the filter g . Its complementary high-pass filter can be defined as (9) H ( z ) = zG ( - z - 1 ) . A sequence of filters with increasing length (indexed by i ) can be obtained (10) G i + 1 ( z ) = G ( z 2 i ) G i ( zz ) , H i + 1 ( z ) = H ( z 2 i ) G i ( z ) , i = 0 , … , I - 1 , with the initial condition G 0 ( z ) = 1. It is expressed as a two scale relation in time domain (11) g i + 1 ( k ) = [ g ] ↑ 2 i g i ( k ) , h i + 1 ( k ) = [ h ] ↑ 2 i g i ( k ) , where the subscript [.] ↑ m indicates the up-sampling by a factor of m , and k is the equally sampled discrete time. The normalized wavelet and scale basis functions φ i , l ( k ), ψ i , l ( k ) can be defined as (12) φ i , l ( k ) = 2 i / 2 g i ( k - 2 i l ) , ψ i , l ( k ) = 2 i / 2 h i ( k - 2 i l ) , where the factor 2 i /2 is an inner product normalization, i and l are the scale parameter and the translation parameter, respectively. The DWT decomposition can be described as (13) a ( i ) ( l ) = x ( k ) ∗ φ i , l ( k ) , d ( i ) ( l ) = x ( k ) ∗ ψ i , l ( k ) , where a ( i ) ( k ) and d ( i ) ( l ) are the approximation coefficients and the detail coefficients at resolution i , respectively [10,25,26] . The discrete wavelet transform (DWT) is an adaptable signal processing tool that finds many engineering and scientific applications. One area in which the DWT has been particularly successful is the epileptic seizure detection because it captures transient features and localizes them in both time and frequency content accurately. DWT analyzes the signal at different frequency bands, with different resolutions by decomposing the signal into a coarse approximation and detailed information. DWT employs two sets of functions called scaling functions and wavelet functions, which are related to low-pass and high-pass filters, respectively. The decomposition of the signal into the different frequency bands is merely obtained by consecutive high-pass and low-pass filtering of the time domain signal. The procedure of multi-resolution decomposition of a signal x ( n ) is schematically shown in Fig. 4 . Each stage of this scheme consists of two digital filters and two down-samplers by 2. The first filter, h [.], is the discrete mother wavelet, high-pass in nature, and the second filter, g [.], is the discrete mother wavelet, high-pass in nature, and the second of the first high-pass and low-pass filters provides the detail, D1, and the approximation, A1, respectively. The first approximation A1 is further decomposed and this process is continued as shown in Fig. 4 [27] . Selection of suitable wavelet and the number of decomposition levels is very important in the analysis of signals using the DWT. The number of decomposition levels is chosen based on the dominant frequency components of the signal. The levels are chosen such that those parts of the signal that correlate well with the frequencies necessary for classification of the signal are retained in the wavelet coefficients. In the present study, since the EEG signals do not have any useful frequency components above 30 Hz, the number of decomposition levels was chosen to be 5. Thus, the EEG signals were decomposed into details D1–D5 and one final approximation, A5. Usually, tests are performed with different types of wavelets and the one which gives maximum efficiency is selected for the particular application. The smoothing feature of the Daubechies wavelet of order 4 (db4) made it more appropriate to detect changes of EEG signals. Hence, the wavelet coefficients were computed using the db4 in the present study. To investigate the effect of other wavelets on classifications accuracy, tests were carried out using other wavelets also. Apart from db4, Symmlet of order 10 (sym10), Coiflet of order 4 (coif4), and Daubechies of order 2 (db2) were also tried. It was noticed that the Daubechies wavelet gives better accuracy than the others, and db4 is slightly better than db2. Fig. 5 shows approximation (A5) and details (D1–D5) of an epileptic EEG signal. Fig. 6 shows approximation (A5) and details (D1–D5) of a normal EEG signal. These approximation and detail records are reconstructed from the Daubechies 4 (DB4) wavelet filter. Wavelet transform acts like a mathematical microscope, zooming into small scales to reveal compactly spaced events in time and zooming out into large scales to exhibit the global waveform patterns [1] . As can be seen from Figs. 5 and 6 , D2 detail coefficient can be used to discriminate the epileptic seizure from healthy subjects. So we have chosen the D2 detail coefficient to distinguish epileptic seizure. 2.4 Distance based data reduction method: data pre-processing Data reduction techniques are a very important step affecting both performance and computation time of classification systems in pattern recognition applications such as medical decision making systems, intelligent control, and data clustering. In the proposed method, data reduction process is done for all attributes of each data sample according to the distance between two data points in the same attribute. The data reduction process is conducted for each attribute separately. For example, all of the values of the first attribute among the whole data are reconsidered and changed according to the distance based data reduction method and then the same process is conducted for the second attribute values of the data samples and so on. The process that is done in determining one attribute value of one data sample can be explained simply like this: Let j be the first data sample label in the i th attribute and m be the second data sample in the i th attribute. As can be seen from Fig. 7 , the first thing we must do is to calculate the distances in case of groups double, respectively. Here, a simple absolute difference is utilized as a distance measure (Eq. (14) ): (14) d ( x i ( j ) , x i ( m ) ) = | x i ( j ) - x i ( m ) | , where x i ( j ) is the j th data sample of i th attribute, while x i ( mk ) is the m th data sample of i th attribute. After the calculation of the distances, they are compared to the threshold that is in the range of 0 and 1 entered by the user. If distance between two data points in the same attribute is bigger than threshold, new data value of two data points will be the average of these data points. Otherwise, one of these data points is chosen as new data value. The process is conducted in the same manner for each attribute value of the same data sample. If two data points in the same attribute are very close to each other, distance between these points will approximate to zero. In this study, we have chosen the threshold value as 0.1 to determine the closeness of data more accurately. 2.5 C4.5 decision tree classifier: classification process C4.5 decision tree learning is one of the most widely used and practical methods for inductive inference. It is a method for approximating discrete-valued functions that is robust to noisy data and capable of learning disjunctive expressions [20,21] . C4.5 decision tree learning is a method for approximating discrete-valued functions, in which the learned function is represented by a decision tree. Learned trees can also be-represented as sets of if–then rules to improve human readability. These learning methods are among the most popular of inductive inference algorithms and have been successfully applied to a broad range of tasks from learning to diagnose medical cases to learning to assess credit risk of loan applicants. Decision tree learning is a heuristic, one-step lookahead (hill climbing), non-backtracking search through the space of all possible decision trees [20–22] . The aim of C4.5 decision tree learning is recursively partition data into sub-groups. Working of decision tree learning is as follows: • Select an attribute and formulate a logical test on attribute. • Branch on each outcome of test, move subset of examples (training data) satisfying that outcome to the corresponding child node. • Run recursively on each child node. • Termination rule specifies when to declare a leaf node. Definitions that used training of C4.5 decision tree learning are explained as follows: • Selection : used to partition training data. • Termination condition : determines when to stop partitioning. • Pruning algorithm : attempts to prevent overfitting. 3 Experimental results 3.1 Used statistics measures We have used four methods for performance evaluation of epileptic seizure diagnosis. These methods are classification accuracy, sensitivity and specifity analysis, and k -fold cross-validation, and 95% confidence intervals. 3.2 Results and discussion In this study, we have proposed a new data reduction method called distance based data reduction method both to increase the classification accuracy of classifier system and to decrease the computation time of classifier on the classification of epileptiform EEG signals. We have chosen C4.5 decision tree classifier that is well known and effective classifier as classification system. Also, autoregressive, discrete Fourier transform, and discrete wavelet transform based spectral analysis methods have been compared with respect to power spectral density and feature extraction. The proposed system has been evaluated with respect to classification accuracy, sensitivity and specifity values, and 95% confidence intervals. The 100 EEG time series of 4096 samples for each class was windowed by a rectangular window. Later, autoregressive (AR), discrete Fourier transform (DFT), and discrete wavelet transform (DWT) methods were applied to these each vectors. Distance based data reduction (DBDR) method was used to reduce from 3200 vectors showing the total EEG signals (1600 vectors for each class) to 1600 vectors. The distribution of both epileptiform EEG signals without data reduction and with data reduction is shown in Table 1 . Power spectral densities (PSDs) of a subject that has eye open and subject that has epileptic seizure subject from the EEG signals using AR, Welch, and DWT methods without/with distance based data reduction method are obtained. While Fig. 8 presents the power spectrum densities of a subject that has eye open and subject that has epileptic seizure subject from the EEG signals using DFT (Welch) method without distance based data reduction, Fig. 9 shows the power spectrum densities of a subject that has eye open and subject that has epileptic seizure subject using DFT (Welch) method with distance based data reduction. Fig. 10 presents the power spectrum densities of a subject that has eye open and subject that has epileptic seizure subject using AR (Burg) method without distance based data reduction. Fig. 11 presents the power spectrum densities of a subject that has eye open and subject that has epileptic seizure subject using AR (Burg) method with distance based data reduction. While Fig. 12 presents the power spectrum densities of a subject that has eye open and subject that has epileptic seizure subject using DWT method without distance based data reduction, Fig. 13 denotes the power spectrum densities of a subject that has eye open and subject that has epileptic seizure subject using DWT method with distance based data reduction. These results show that the best distinguishable methods are AR (Burg) method and DFT (Welch) methods without/with data reduction for detection of epileptic seizure, respectively. In this work, six different combined methods have been used to classify the epileptiform EEG signal. These methods are (i) combining DFT and C4.5 decision tree classifier (DCT), (ii) combining DFT, distance based data reduction, and C4.5 DCT, (iii) combining AR and C4.5 DCT, (iv) combining AR, distance based data reduction, and C4.5 DCT, (v) combining DWT and C4.5 DCT, and (vi) combining DWT, distance based data reduction, and C4.5 DCT. While Table 2 presents the results obtained by combining DFT and C4.5 decision tree classifier (DCT) for 10-fold cross-validation, Table 3 shows the results obtained by combining DFT, distance based data reduction, and C4.5 DCT. Table 4 gives the results obtained by combining AR and C4.5 DCT using 10-fold cross-validation. Table 5 displays the results obtained by combining AR, distance based data reduction, and C4.5 DCT using 10-fold cross-validation. While Table 6 denotes the results obtained by combining DWT and C4.5 DCT using 10-fold cross-validation, Table 7 denotes the results obtained by combining DWT, distance based data reduction, and C4.5 DCT. These results have shown that the proposed distance based data reduction method has produced very promising results with respect to both classification accuracy and computation time for classifying of epileptiform EEG signals. Also, we have calculated the 95% confidence interval in the 10-fold cross-validation test. We have given the 95% confidence intervals for epileptiform EEG signals dataset shown in Table 8 . We compare our results with the previous results reported by earlier methods. Table 9 shows the classification accuracies of our method and previous methods. As we can see from these results, our method using 10-fold cross-validation obtains the highest classification accuracy, 99.32%, reported so far. The above results have shown that distance based data reduction method is superior to the application of classification of EEG signals. 4 Conclusion and future work It is a difficult task to detect epilepsy and requires observation of the patient, an EEG, and collection of additional clinical information. Combining feature extraction methods based on AR, DFT, and DWT, distance based data reduction, and C4.5 decision tree classifier that classifies subjects as having or not having an epileptic seizure provides a valuable diagnostic decision support tool for physicians treating potential epilepsy, since differing etiologies of seizures result in different treatments. In this work, we have proposed a new data reduction method called distance based data reduction method both to increase the classification accuracy of classifier system and to decrease the computation time of classifier on the classification of epileptiform EEG signals. We have chosen C4.5 decision tree classifier that is well known and effective classifier as classification system. Also, autoregressive, discrete Fourier transform, and discrete wavelet transform based spectral analysis methods have been compared with respect to power spectral density and feature extraction. As can be seen from the above results, the best distinguishable methods are AR (Burg) method and DFT (Welch) methods without/with data reduction for detection of epileptic seizure, respectively. The proposed system has been evaluated with respect to classification accuracy, sensitivity and specifity values, and 95% confidence intervals. The obtained results have shown that distance based data reduction method is superior to the application of classification of EEG signals. Acknowledgement This study is supported by the Scientific Research Projects of Selcuk University. References [1] H. Adeli Z. Zhou N. Dadmehr Analysis of EEG records in anepileptic patient using wavelet transform J. Neurosci. Meth. 123 1 2003 69 87 [2] N. Hazarika J.Z. Chen A.C. Tsoi A. Sergejew Classification of EEG signals using the wavelet transform Signal Process. 59 1 1997 61 72 [3] O.A. Rosso A. Figliola J. Creso E. Serrano Analysis of wavelet-filtered tonic-clonic electroencephalogram recordings Med. Biol. Eng. Comput. 42 4 2004 516 523 [4] J.R. Glover Jr. N. Raghaven P.Y. Ktonas J.D. Frost Jr. Context-based automated detection of epileptogenic sharp transients in the EEG: elimination of false positives IEEE Trans. Biomed. Eng. 36 5 1989 519 527 [5] A.J. Gabor M. Seyal Automated interictal EEG spike detection using artificial neural networks Electroencephalogr. Clin. Neurophysiol. 83 5 1992 271 280 [6] W.R.S. Webber B. Litt R.P. Lesser R.S. Fisher I. Bankman Automatic EEG spike detection: what should the computer imitate? Electroencephalogr. Clin. Neurophysiol. 87 6 1993 364 373 [7] V.P. Nigam D. Graupe A neural-network-based detection of epilepsy Neurol. Res. 26 1 2004 55 60 [8] I. Guler E.D. Ubeyli Adaptive neuro-fuzzy inference system for classification of EEG signals using wavelet coefficients J. Neurosci. Meth. 148 2005 113 121 [9] I. Guler E.D. Ubeyli Features extracted by eigenvector methods for detecting variability of EEG signals Pattern Recognit. Lett. 28 5 2007 592 603 [10] A. Subasi EEG signal classification using wavelet feature extraction and a mixture of expert model Expert Syst. Appl. 32 4 2007 1084 1093 [11] R.G. Andrzejak K. Lehnertz F. Mormann C. Rieke P. David C.E. Elger Indications of nonlinear deterministic and finite-dimensional structures in time series of brain electrical activity: dependence on recording region and brain state Phys. Rev. E 64 2001 061907 [12] D. Evans Doppler signal analysis Ultrasound Med. Biol. 26 Suppl. 1 2000 S13 S15 [13] P.J. Vaitkus R.S.C. Cobbold K.W. Johnston A comparative study and assessment of Doppler ultrasound spectral estimation techniques part II: methods and results Ultrasound Med. Biol. 14 1988 673 688 [14] I. Guler F. Hardalac M. Kaymaz Comparison of FFT and adaptive ARMA methods in transcranial Doppler signals recorded from the cerebral vessels Comp. Biol. Med. 32 2002 445 453 [15] C.K. Yeh P.C. Li Doppler angle estimation of pulsatile flows using AR modeling Ultrasonic Imaging 24 2002 135 146 [16] K. Kaluzynski Order selection in Doppler blood flow signal spectral analysis using autoregressive modeling Med. Biol. Eng. Comput. 27 1989 89 92 [17] S.M. Kay S.L. Marple Spectrum analysis—a modern perspective Proc. IEEE 69 11 1981 1380 1419 [18] F.S. Schlindwein D.H. Evans Selection of the order of autoregressive models for spectral analysis of Doppler ultrasound signals Ultrasound Med. Biol. 16 1 1990 81 91 [19] H. Akaike A new look at the statistical model identification IEEE Trans. Autom. Control AC-19 1974 716 723 [20] M.T. Mitchell Machine Learning 1997 McGraw-Hill Singapore [21] J.R. Quinlan Induction of decision trees Mach. Learn. 1 1986 81 106 [22] K. Polat S. Güneş The effect to diagnostic accuracy of decision tree classifier of fuzzy and k-NN based weighted pre-processing methods to diagnosis of erythemato-squamous diseases Digital Signal Process. 16 6 2006 922 930 [23] N.F. Guler E.D. Ubeyli I. Guler Recurrent neural networks employing Lyapunov exponents for EEG signals classification Expert Syst. Appl. 29 2005 506 514 [24] N. Kannathal Min Lim Choo U. Rajendra Acharya P.K. Sadasivana Entropies for detection of epilepsy in EEG Comp. Meth. Programs Biomed. 80 2005 187 194 [25] M. Akay Wavelet applications in medicine IEEE Spectrum 34 5 1997 c50 c56 [26] N.F. Guler E.D. Ubeyli Wavelet-based neural network analysis of ophthalmic artery Doppler signals Comp. Biol. Med. 34 7 2004 601 613 [27] A. Subasi Epileptic seizure detection using dynamic wavelet network Expert Syst. Appl. 29 2005 343 355
Year
DOI
Venue
2008
10.1016/j.amc.2007.12.028
Applied Mathematics and Computation
Keywords
Field
DocType
EEG signals,Distance based data reduction,AR spectral analysis,Discrete Fourier transform,Discrete wavelet transform (DWT),C4.5 decision tree classifier,Epileptic seizure
Decision tree,Discrete cosine transform,Discrete wavelet transform,Artificial intelligence,Cluster analysis,Mathematical optimization,Pattern recognition,Algorithm,Feature extraction,Discrete Fourier transform,Mathematics,Decision tree learning,Data reduction
Journal
Volume
Issue
ISSN
200
1
0096-3003
Citations 
PageRank 
References 
8
0.73
10
Authors
2
Name
Order
Citations
PageRank
Kemal Polat1134897.38
Salih Güneş2126778.53