Engineering, Technology & Applied Science Research Vol. 8, No. 4, 2018, 3093-3097 3093 www.etasr.com Tzimourta et al.: Evaluation of Window Size in Classification of Epileptic Short-Term EEG Signals Using … Evaluation of Window Size in Classification of Epileptic Short-Term EEG Signals Using a Brain Computer Interface Software Katerina D. Tzimourta Medical Physics Laboratory, University of Ioannina Ioannina, Greece ktzimourta@cc.uoi.gr Loukas G. Astrakas Medical Physics Laboratory, University of Ioannina Ioannina, Greece astrakas@uoi.gr Anna Maria Gianni Computer Engineering Dpt, Technological Educational Institute of Epirus, Kostakioi, Arta, Greece anna.maria.gianni@gmail.com Alexandros T. Tzallas Computer Engineering Dpt, Technological Educational Institute of Epirus, Kostakioi, Arta, Greece tzallas@teiep.gr Nikolaos Giannakeas Computer Engineering Dpt, Technological Educational Institute of Epirus, Kostakioi, Arta, Greece giannakeas@teiep.gr Ioannis Paliokas Information Technologies Institute, Centre for Research and Technology Hellas Thessaloniki, Greece ipaliokas@iti.gr Dimitrios G. Tsalikakis Informatics and Telecommunications Eng. Dpt, University of Western Macedonia, Kozani, Greece dtsalikakis@uowm.gr Markos G. Tsipouras Informatics and Telecommunications Eng. Dpt, University of Western Macedonia, Kozani, Greece mtsipouras@uowm.gr Abstract— The complexity of epilepsy created a fertile ground for further research in automated methods, attempting to help the epileptologists’ task. Over the past years, great breakthroughs have emerged in computer-aided analysis. Furthermore, the advent of Brain Computer Interface (BCI) systems has facilitated significantly the automated seizure analysis. In this study, an evaluation of the window size in automated seizure detection is proposed. The EEG signals from the University of Bonn was employed and segmented into 24 epochs of different window lengths with 50% overlap each time. Statistical and spectral features were extracted in the OpenViBE scenario and were used to train four different classifiers. Results in terms of accuracy were above 80% for the Decision Tree classifier. Also, results indicated that different window sizes provide small variations in classification accuracy. Keywords-epilepsy; EEG; seizure detection; window size; brain computer interface I. INTRODUCTION Epilepsy is a devastating brain disorder followed by seizures, which are repetitive episodes of temporary interruption or disturbance of communication between neurons. Many people experience seizures without a clear cause and almost one-third of the epileptic patients suffer from refractory seizures [1]. The latest facts render epilepsy a life-threatening disorder and a major factor responsible for mortality in developed and developing countries [2]. Depending on the brain areas that participate during epileptic activity, seizures are divided into two fundamental types: partial (affects only a single brain area) and generalized (affects more than one region). These two main types are subdivided forming a bigger list of several seizure types [3]. The Electroencephalogram (EEG) is used to monitor and diagnose epilepsy. The brain activity is monitored through the EEG which is usually performed in a well-equipped hospital. The electrodes are either attached to the surface of the skull (scalp EEG – sEEG) or placed invasively inside the brain (intracranial EEG – iEEG). Furthermore, the EEG recording is usually performed after a seizure episode and before the next seizure occurrence (interictal period). Rarely an EEG recording captures the seizure onset (ictal period) and it usually happens in a 24-hour monitoring. The complexity of EEG recordings and the huge amount of data, led to the development of methods for detecting different patterns of brain activity [4] and automated seizure analysis [5]. Generally, these methods follow a pattern recognition approach, which contains feature extraction and classification. The signal is usually decomposed in epochs of specific duration in an attempt to better capture the transient occurrences of the EEG. Several time-frequency analysis methods have been proposed such as Discrete Wavelet Transform [6-8], Wigner-Ville distribution [9], Empirical Mode Decomposition [10-11] etc. Significant features are extracted from the decomposed signal and are then used to train a classifier. Usually, the raw signal is initially analyzed in epochs of small duration and the window size is of primary importance in automated seizure detection. Recent breakthroughs in computer-aided analysis initiated the development of Brain Computer Interface (BCI) systems, which are more user-friendly and provide direct communication with the user’s brain in real-time without any possible movement [12]. In this work, a method for seizure det ana ava dif ov cal cas tra sig tes op com mi win Un dec fea fea Fig an tra dep A. ord of EE vo gro con dur Engineerin www.etasr tection and t alysis is pres ailable databa fferent lengths erlap. Signifi lculated and e ses were creat ained and teste In this work gnal processing sting various en-source so mmunication inimum of pro ndow size in niversity of composed in atures were e atures were u gure 1 a flowc example of aining parame picted in Figur Fig. 1. A The Databas The databas der to investig EEG segment EG recording lunteers and ouped in five ntaining of 10 ration marked ng, Technology r.com T the investigat sented. The E ase was used s, ranging fro ficant statistic employed to t ted in the open d the four clas II. MATERI k, the OpenVi g and the WE classifiers. T oftware desig with the brain ogramming ski n EEG signa Bonn was u various epoch extracted in t used to train chart of the pro f the OpenVi eters which u re 2. A flowchart of the se e University gate the windo ts that were se gs, which w 5 epileptic p subsets, deno 00 single-chann d as Z, O, N y & Applied Sci Tzimourta et al. tion of wind EEG recordin and decomp om 1 to 24 cal and spec train four clas n-source softw ssifiers. IALS AND MET iBE BCI soft EKA environm The OpenViBE gned to be n, by a broad ills. To investi l analysis, th used. The E hs of differen the OpenViB four remark oposed method iBE scenario used to train proposed three-s of Bonn [14] ow length. Th elected and cu were acquired patients. The oted as A, B, nel EEG segm N, F and S, ience Research : Evaluation of dow size in ngs from a p posed in epoc seconds with tral features ssifiers. In tot ware OpenViB THOD tware was use ment for trainin E [13] is an used for range of users igate the role he database o EEG signals nt sizes and s BE scenario. kable classifie d is presented. that contain n the classifie step methodology ] was employ he database co ut out of conti d from 5 h EEG segmen , C, D and E ments of 23.6-s , respectively h V f Window Size i signal public chs of h 50% were tal, 24 E that ed for ng and easy, direct s with of the of the were everal These ers. In Also, ns the ers is yed in onsists inuous ealthy nts are E each second y. The sam wer pre emp set cha as g pro Sub lab Z O N F S B. dec sec epo epo eac seg 13 Vol. 8, No. 4, 20 in Classificatio mpling frequen re visually re sented in Tab ployed to crea is a class (Z allenging and m good classifica oblem. Fig. 2. Part o TABLE bset bel Type of EEG recording Z sEEG O sEEG N iEEG F iEEG S iEEG Time Epochin In the Ope composed in e ond with 0.5 oching” was u ochs from 1 to ch time. In to gments ranging seconds overla 018, 3093-3097 on of Epileptic S ncy of the dat emoved. More ble I. In this st ate the proble Z-O-N-F-S). T many studies ation results as of the OpenViBE I. SUMMAR g Subjects healthy healthy epileptic epileptic epileptic ng enViBE Desi epochs of cons 5 seconds ove used to confi o 24 seconds w otal, 24 XML g from 1 secon ap. Short-Term EE ta is 173.61Hz e details abou tudy, all the 5 em of five cla This particular [15-17] have a s the ones obta scenario for the “ RY OF THE BONN D Subjec relaxed in an aw op relaxed in an aw cl Seizure-free fr hemisphere of z Seizure-free from z Seizure act epilepto igner the EE secutive secon erlap. The plu figure the dur with 50% seco L files were nd up until 24 3094 EG Signals Usin z and any art ut the databas 500 segments asses, wherein r problem is attempted to o ained from a b “F” EEG segment DATABASE ct’s state wake state with ey pened wake state with ey losed from the opposite the epileptogenic zone m the epileptogen zone tivity from the ogenic zone EG signals nds, starting fr ugin “Time b ration of the onds epoch ov created with seconds with ng … tifacts e are were each more obtain binary ts. es es c nic were rom 1 based EEG verlap EEG 0.5 - Engineering, Technology & Applied Science Research Vol. 8, No. 4, 2018, 3093-3097 3095 www.etasr.com Tzimourta et al.: Evaluation of Window Size in Classification of Epileptic Short-Term EEG Signals Using … C. Feature Extraction A set of 8 features was extracted in each epoch, forming the feature vector that used in the classification. The plugins “Univariate Statistics” and “Spectral Analysis” were employed in the OpenViBE scenario to compute four time-based features namely:  mean value,  variance,  range (maximum value – minimum value),  median value, and four spectral features based on Fast Fourier Transform being:  the spectrum amplitude (the power of the signal) in alpha band (8-12Hz),  the spectrum amplitude in beta band (12-25Hz),  the spectrum amplitude in theta band (4-8Hz),  the spectrum amplitude in delta band (1-4Hz), The resulting feature set was used to train four well-known classifiers, being Naïve Bayes, Multilayer Perceptron, Support Vector Machines and Decision Tree. D. Classification To evaluate the proposed method and the window size in classification results, four of the most sophisticated supervised classifiers were used. 1) Naïve Bayes (NB) Naïve Bayes (NB) is a simple classifier that combines a probability model with a decision rule. The classifier operates on the simplest assumption that the features are conditionally independent and is based on Bayes decision theory, wherein the posterior probability of each class is calculated by the likelihood and the prior probability [15]. The ultimate goal of the classifier is to minimize the probability of classification error and maximize the posterior probability. A small number of training data is needed for the classifier to estimate the necessary classification parameters and in conjunction with the less time complexity, this simple classifier has preferred in complex classification problems. 2) MultiLayer Perceptron (MLP) The Multilayer Perceptron (MLP) classifier is a neural network with at least three layers of nodes. MLP utilizes the backpropagation techniques for training and maps non-linear input data into a space, where it becomes linearly separable. In order to train a MLP classifier and perform correct pattern classification, the connection weights after each processing of data are adjusted, based on the comparison between the error in the output and the expected result [15]. 3) Support Vector Machines (SVM) Support Vector Machines is a machine learning technique for linear and non-linear classification problems. The non- linear input data is projected into a high-dimension feature space in order to be linearly separated. This projection is performed by the kernel function, which can be either a linear or a polynomial function, the radial basis function or the sigmoid kernels. The gap that separates the data is called hyperplane and the major goal of the algorithm is to find the optimal separating hyperplane that maximizes the distance between the data and minimizes the classification error [15]. In our experiments, the radial basis function was used. 4) Decision Tree (DT) A Decision Tree (DT) classifier is a straightforward classifier based on a series of decision rules. The root node of the tree is displayed at the top and is successively connected with other nodes through links or branches, until no further links to other nodes exist (leaf nodes). According to the DT classifier, only one link can be followed each time and the subsequent node becomes the root node of the next sub-tree. The procedure is repeated until a leaf node is reached, leading to no further decision and the category label is read [15]. III. RESULTS AND DISCUSSION The four classifiers were trained and tested according to the 10-fold cross-validation technique. To evaluate the classification results of the experiments and thus, the window size the accuracy was calculated from the correctly classified instances: = (1) The obtained statistical results of the 24 experiments for each classifier are presented in Table II. TABLE II. RESULTS Epoch (sec) Over- lapping (sec) Correctly Classified instances (%) NB MLP SVM DT 1 0.5 59.11% 72.41% 67.51% 85.95% 2 1.0 63.67% 75.56% 70.35% 79.06% 3 1.5 66.37% 78.02% 70.46% 82.05% 4 2.0 67.47% 78.73% 70.11% 83.50% 5 2.5 68.97% 79.47% 68.87% 83.60% 6 3.0 70.32% 81.72% 70.00% 84.87% 7 3.5 70.23% 81.10% 69.21% 85.45% 8 4.0 70.03% 80.17% 68.55% 85.17% 9 4.5 71.90% 80.91% 68.45% 85.19% 10 5.0 72.11% 77.75% 63.63% 83.83% 11 5.5 72.08% 83.15% 68.10% 85.48% 12 6.0 73.33% 81.95% 68.46% 85.39% 13 6.5 73.52% 81.03% 67.44% 83.72% 14 7.0 74.01% 81.23% 67.83% 84.25% 15 7.5 75.29% 81.98% 68.34% 82.71% 16 8.0 75.61% 83.35% 66.57% 84.51% 17 8.5 74.58% 83.53% 68.19% 83.75% 18 9.0 75.85% 82.39% 67.72% 82.31% 19 9.5 75.37% 82.18% 68.31% 82.60% 20 10.0 75.57% 83.00% 67.87% 81.23% 21 10.5 81.23% 82.79% 67.62% 84.46% 22 11.0 76.48% 82.04% 68.87% 82.82% 23 11.5 77.34% 82.24% 68.16% 83.77% 24 12.0 77.55% 81.70% 68.82% 85.63% Engineering, Technology & Applied Science Research Vol. 8, No. 4, 2018, 3093-3097 3096 www.etasr.com Tzimourta et al.: Evaluation of Window Size in Classification of Epileptic Short-Term EEG Signals Using … The best classifier for epochs of 1 second with 0.5 seconds overlap, is DT with 85.95% followed by MLP with 72.41%, SVM with 67.51% and NB with 59.11%. For epochs lasting 2 seconds with 1 second overlap, the best classifier is DT with 79.06%, followed by MLP (75.56%) and SVM (70.35%) and the worst classifier is NB with 63.67% accuracy. For windows of 3 seconds with 1.5 seconds overlap and 4 seconds with 2 seconds overlap, DT remains the best classifier (82.05% and 83.50% respectively) and the lowest accuracy (66.37% and 67.47% respectively) is obtained with NB as well. For the next 12 window sizes the DT outperforms (ranging from 82.71% to 85.48%) and the lowest values (ranging from 63.63% to 70%) are obtained with SVM. At the same time, MLP indicates accuracy ranging from 77.75% to 83.35%, whereas NB shows lower accuracy, ranging from 68.97% to 75.61%, and in some cases perform almost the same with SVM (for windows of 5sec and 6sec duration). For the next 4 window sizes the DT and the MLP presents almost the same accuracy and none of the two classifiers is considerably better than the other. Accuracy for NB is about 75% for these four window sizes and SVM does not exceed 68%. Finally, for the last four window sizes, DT is the best classifier (ranging from 81.23% to 85.63%) followed by MLP (ranging from 82.04% to 83%), NB (ranging from 75.57% to 81.23%) and SVM (ranging from 67.62% to 68.87%). The smooth and small changes of the obtained results for each classifier and the comparison between their performances are depicted in Figure 3. From Figure 3 it can be observed that while the window size is increasing, the classification results for NB, MLP and DT are slightly higher, without great variations (in most cases not greater that 3%). The lowest accuracy for NB and MLP is 59.11% and 72.41% and the highest 81.23% and 83.53%, respectively; however, despite the big difference between the lowest and highest values, the intermediate values range from 1% to 3%. The best classification accuracy is provided with DT with lowest value 79.06% (2 seconds window with 1 second overlap) and highest 85.63% (24 seconds with 12 seconds overlap) whereas SVM showed the worst classification accuracy (lowest value 63.63% and highest 70.46%). Also, the best accuracy for DT, MLP and NB (above 80%) is obtained for windows of 21 seconds with 10.5 seconds overlap, whereas the SVM seemed to be the weakest classifier compared to the others. The epoch duration has also been a focus point for several researchers. Recently [16], a method based on dynamic principal component analysis (DPCA) and energy was proposed. The authors evaluated four window sizes of the EEG segments, being 64, 128, 256 and 512 samples per window (approximately 0.37, 0.74, 1.47 and 2.95 seconds respectively). Results showed minor increase when signals were segmented in 512 samples epochs per window. In [17], authors conducted 4 experiments to find the optimal window size between 4 options. In the presented method the Permutation Entropy was calculated from nonoverlapping epochs of 0.25, 0.5, 0.75 and 1 second and was used to train a SVM classifier. Results in terms of F1 score showed small deviation (about 5% for overall F1 score). An extension of the above mentioned approach was presented in [18] wherein the same group of authors investigated among other, whether the overlapping plays significant role in seizure detection or not and they proposed a method based on Weighted Permutation Entropy (WPE). In this approach, 200 signals of the Bonn database were employed and segmented in epochs of approximately 0.35 seconds duration with overlap (OV) (128 samples per window) and without overlap (NOV) (164 samples per window). The method was evaluated on SVM and Artificial Neural Network (ANN) classifier and results indicated that epochs with 50% overlap provided slightly higher accuracy (2.25% overall accuracy for SVM and ANN). In our experiments, 24 window sizes are evaluated and the classification accuracy slightly arises as the window size increases for all classifiers. The best classifier is DT with accuracy ranging from 79.06% to 85.95%, followed by MLP ranging from 72.41% to 83.53%, SVM from 67.51% to 70.46% and NB from 59.11% to 75.61%. A comparison table is presented in Table III. TABLE III. COMPARISON TABLE Reference Window size Signal analysis Features Classifier Classification Problem Performance Metrics [16] 4 non-overlapping window sizes ranging from 0.37-2.95 sec Dynamic Principal Component Analysis Energy 1-NN Z-S ZONF-S Accuracy: 99.9%-100% [17] 4 non-overlapping window sizes ranging from 0.25-1 sec Raw segmented data Weighted Permutation Entropy SVM Z-S O-S N-S F-S Average F1 values for each window size: 0.866 - 0.917 [18] 0.35 sec non- overlapping window Raw segmented data Weighted Permutation Entropy SVM Z-O-N-F-S NOV Accuracy: 90.63 - 92.88% 0.35 sec with 50% overlap window OV Accuracy: 91.63 - 93.88% This work 24 overlapping windows with 50% overlap ranging from 1- 24sec Raw segmented data Mean, Range, Variance, Median, Spectral amplitude in alpha, beta, delta, theta band Naïve Bayes, MLP, SVM, DT Z-O-N-F-S NB Accuracy: 59.11% - 81.23% MLP Accuracy: 72.41% - 83.53% SVM Accuracy: 63.63% - 70.46% DT Accuracy: 79.06% - 85.63% Fig ML stu mo me ext dec cha exp do Co com sig Fu sec win wa abo win wh [1] [2] [3] [4] [5] [6] [7] Engineerin www.etasr g. 3. Results i LP, yellow: NB, p Epilepsy is udies have foc ost of the stu ethodology, w traction and composition anges of the periments indi not have a sig onsequently, a mputational t gnificant par urthermore, the conds with 10 ndow length f as 23.6 second ove 24 secon ndow size ca hich are closer World Health O O. Devinsky, preventing ep Vol. 86, No. 8, I. E. Scheffer, Guilhoto, E. H Perucca, T. classification o for Classificati 521, 2017 S. Segkouli, I “Study of EEG Cognitive Dec Computing Par A. T. Tzallas, Astrakas, S. K detection meth I. Guler, E. D signals classifi Biomedicine, V Y. Kumar, M. DWT based fu Neurocomputin ng, Technology r.com T in terms of accur purple: SVM) IV. CO constantly un cused on seiz udies, the auth which contain classification is very imp e EEG recor icated that sm gnificant impa a smaller nu time of the c rameter in e best window 0.5 seconds ov for seizure pre ds, little do we nds. In the f an be perform r to the clinical REFE Organization, “Ep T. Spruill, D. Th ilepsy-related mo , pp. 779-786, 20 S. Berkovic, G. Hirsch, S. Jain, Tomson, S. W of the epilepsies: ion and Terminol I. Paliokas, D. T G Power Fluctuati cline Screening”, radigms for Ment M. G. Tsipouras Konitsiotis, M. T hods: a review stu D. Ubeyli, “Multic ication”, IEEE Tr Vol. 11, No. 2, pp L. Dewal, R. S. A fuzzy approximat ng, Vol. 133, pp. y & Applied Sci Tzimourta et al. racy for the 4 cla ONCLUSIONS nder the mic zure detection hors follow a n signal deco n. The windo ortant on de rding. Howev mall difference act on the clas umber of epo lassification p real-time w size seemed verlap, which m ediction. Since e know about t future, the in med in long-t l EEG recordin ERENCES pilepsy”, Fact she hurman, D. Fried ortality: A call 16 Capovilla, M. B. G. Mathern, S. Wiebe, Y. Zhan Position paper o logy”, Epilepsia, Tzovaras, M. Tso ions Enhanced by , International Sy tal Health, pp. 165 s, D. G. Tsalikak zaphlidou. “Auto udy”, in :Epilepsy, class support vec ransactions on Inf p. 117-126, 2007 Anand, “Epileptic e entropy and su 271-279, 2014 ience Research : Evaluation of assifiers (blue: D roscope and and predictio a certain thre omposition, f ow size in etecting the ver, results o es in epoch du ssification accu ochs decrease process which applications to be the one may be a prom e the signal du the window si nvestigation o erm EEG dat ngs. eet N999, 2012 dman, “Recognizi for action”, Neu Connolly, J. Fre Moshé, D. Nord ng, S. Zuberi, f the ILAE Comm Vol. 58, No. 4, p olaki, C. Karagia y Linguistic Stimu ymposium on Pe 5-175, 2015 kis, E. C. Karvou omated epileptic , Intechopen, 201 ctor machines for formation Techno c seizure detectio upport vector ma h V f Window Size i DT, red: many on. In ee-step feature signal subtle of our uration uracy. es the h is a [19] of 21 mising uration ize for of the tasets, ing and urology, ench, L. dli, EE. “ILAE mission pp. 512- annidis, ulus for ervasive unis, L. seizure 2. r EEG- ology in on using achine”, [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] [20] [21] [22] Vol. 8, No. 4, 20 in Classificatio K. D. Tzimourt Tzallas, S. Ko seizures in EEG Computer-Base A. T. Tzallas, detection base networks”, Com 13, 2007 ] S. S. Alam, M. higher order sta and Health Info ] A. Bhardwaj, programming Methods and Pr ] L. Huang, G. treatment”, in: Future Prospect ] Y. Renard, F. L Bertrand, A. Lé to Design, Test Environments1” Vol. 19, 2010 ] R. G. Andrzeja Elger. “Indicat structures in ti recording regio 2001 ] A.T. Tzallas, detection in EE on Information 2009 ] C. Kamath. “A Cepstrum and p Vol. 1, 2014 ] A. Bhardwaj, programming methods and pr ] R. O. Duda, P. & Sons, 2012 ] S. Xie, S. Kr nonoverlapping classification”, 419308, 2014 ] N. Seddik, S. long-term scalp Vector Machin 170-173, 2014 ] N. S. Tawfik, S of epileptic s Engineering, Vo ] V. Nagaraj, S. Irazoqui, T. Ne Closing the lo Publication of t No. 3, pp. 194, 018, 3093-3097 on of Epileptic S ta, L. G. Astrakas onitsiotis, “Wav G signals”, 2017 ed Medical System M. G. Tsipoura ed on time-freq mputational Intelli I. H. Bhuiyan, “ atistics in the EM ormatics, Vol. 17, A. Tiwari, R. K approach for ep rograms in Biome van Luijtelaar, “ Brain-Computer ts, InTech, 2013 Lotte, G. Gibert, M écuyer, “OpenViB t and Use Brain-C ”, Presence: Tel ak, K. Lehnertz, F tions of nonlinea ime series of br on and brain stat M. G. Tsipoura EGs using time–f Technology in B Automatic seizur pattern recognitio A. Tiwari, R. K approach for ep rograms in biomed E. Hart, D. G. S rishnan, “Dynam g moving window The Scientific W Youssef, M. Kh p EEG using Wei ne”, Biomedical E S. M. Youssef, M. seizures in EE ol. 53, pp. 177-19 . Lee, E. Krook toff, “The Future oop”, Journal o the American Ele 2015 Short-Term EE s, M. G. Tsipoura velet based clas IEEE 30th Intern ms (CBMS), pp. 3 as, D. I. Fotiadi quency analysis igence and Neuro “Detection of seiz D domain”, IEEE No. 2, pp. 312-3 Krishna, V. Var pileptic seizure edicine, Vol. 124, “Brain computer Interface System M. Congedo, E. M BE: An Open-So Computer Interfa leoperators and F. Mormann, C. ar deterministic rain electrical ac te”, Physical Rev as, D. I. Fotiad frequency analys iomedicine, Vol. re detection bas on neural network Krishna, V. Var pileptic seizure dicine, Vol. 124, Stork, Pattern Cla mic principal com w and its applica World Journal, Vo holief, “Automati ighted Permutatio Engineering Con . Kholief, “A hyb EG records”, Com 90, 2014 -Magnuson, I. S e of Seizure Pre-d of Clinical Neu ectroencephalogr 3097 EG Signals Usin as, N. Giannakeas sification of ep national Symposi 35-39, 2017 is, “Automatic s and artificial oscience, Vol. 18, zure and epilepsy E Journal of Biom 18, 2013 rma, “A novel g detection”, Com , pp. 2-18, 2016 interface for ep ms-Recent Progre Maby, V. Delann urce Software Pl ces in Real and V Virtual Environ Rieke, P. David, and finite-dimen ctivity: Dependen iew E, Vol. 64, N dis, “Epileptic s is”, IEEE Transa 13, No. 5, pp. 70 sed on Teager E ks”, QScience Co rma, “A novel g detection”, Com pp. 2-18, 2016 assification, John mponent analysis ations to epileptic olume 2014, Arti ic seizure detect on Entropy and S nference (CIBEC rid automated det mputers & Ele Soltesz, P. Benqu diction and Interve urophysiology: O aphic Society, V ng … , A. T. pileptic ium on seizure neural , pp. 1- y using medical genetic mputer pilepsy ess and noy, O. atform Virtual nments, , C. E. nsional nce on No. 6, seizure actions 03-710, Energy onnect, genetic mputer Wiley s with c EEG icle ID tion in upport C), pp. tection ectrical uet, P. ention: Official Vol. 32,