Research Article Adaptive Neuro-Fuzzy Inference System for ...Research Article Adaptive Neuro-Fuzzy...

Research ArticleAdaptive Neuro-Fuzzy Inference System for Classification ofBackground EEG Signals from ESES Patients and Controls

Zhixian Yang1 Yinghua Wang23 and Gaoxiang Ouyang23

1 Department of Pediatrics Peking University First Hospital No 1 of Xian Men Street Xicheng District Beijing 100034 China2 State Key Laboratory of Cognitive Neuroscience and Learning and IDGMcGovern Institute for Brain ResearchBeijing Normal University Beijing 100875 China

3 Center for Collaboration and Innovation in Brain and Learning Sciences Beijing Normal University Beijing 100875 China

Correspondence should be addressed to Gaoxiang Ouyang ouyangbnueducn

Received 13 December 2013 Accepted 18 March 2014 Published 25 March 2014

Academic Editors J Tang and H Zhou

Copyright copy 2014 Zhixian Yang et al This is an open access article distributed under the Creative Commons Attribution Licensewhich permits unrestricted use distribution and reproduction in any medium provided the original work is properly cited

Background electroencephalography (EEG) recorded with scalp electrodes in children with electrical status epilepticus duringslow-wave sleep (ESES) syndrome and control subjects has been analyzed We considered 10 ESES patients all right-handed andaged 3ndash9 years The 10 control individuals had the same characteristics of the ESES ones but presented a normal EEG Recordingswere undertaken in the awake and relaxed states with their eyes open The complexity of background EEG was evaluated usingthe permutation entropy (PE) and sample entropy (SampEn) in combination with the ANOVA test It can be seen that the entropymeasures of EEG are significantly different between the ESES patients and normal control subjectsThen a classification frameworkbased on entropy measures and adaptive neuro-fuzzy inference system (ANFIS) classifier is proposed to distinguish ESES andnormal EEG signals The results are promising and a classification accuracy of about 89 is achieved

1 Introduction

Encephalopathy with electrical status epilepticus duringslow-wave sleep (ESES) syndrome is a condition charac-terized by continuous spikes and waves occurring duringsleep [1] The recent literature refers to it as ldquoESES syn-dromerdquo which is an age-related reversible disorder withonset at around 4-5 years of age and a generally favorablecourse with disappearance at around 10ndash15 years of age [2]usually associated with variable cognitive and behavioralimpairments [3 4] The pathophysiological mechanisms andneuropsychological deficits associated with this conditionare still poorly understood [5] Therefore it is important toidentify the ESES patients as early as possible such that theclinician can prescribe the necessary medication to stop itsprogression

The electroencephalograph (EEG) signal is a measureof the summed activities of approximately 1ndash100 millionneurons lying in the vicinity of the recording electrode Sinceit may provide insight into the functional structure and

dynamics of the brain [6] exploration of hidden dynamicalstructures within EEG signals is of both basic and clinicalinterest andhas attractedmore andmore attention [7ndash9]Onecan assume that the EEG is a signal containing informationabout the condition of the brain We can also accept asworking hypothesis that the EEG recorded under restingconditions is representative of the global state of the brain [1011] Then a plausible working hypothesis is that backgroundEEG corresponding to healthy controls is different fromthat corresponding to patients with pathologies (eg ESES)However it is currently accepted that a human observerhardly discriminates EEG traces of healthy controls fromthose of ESES subjects Quantitative EEG analysis usingcomputational methods can therefore assist in the back-ground EEG characterizationThe EEG pattern classificationscheme usually includes two major parts feature extractionand classification

Various methods have been widely used for featureextraction ranging from traditional linear methods such

Hindawi Publishing Corporatione Scientific World JournalVolume 2014 Article ID 140863 8 pageshttpdxdoiorg1011552014140863

2 The Scientific World Journal

as Fourier transforms and spectral analysis [12] to nonlin-ear methods such as Lyapunov exponents [13] correlationdimension [14] and similarity [15 16] Due to the complexinterconnections between billions of neurons the recordedEEG signals are complex nonlinear nonstationary andrandom in natureTherefore the classification of EEG signalsusing nonlinear methods that detect and quantify nonlinearmechanisms and thereby better reflect the characteristics ofthe EEG signals Nonlinear features may be able to unearththe hidden complexities existing in the EEG time series Ferriet al applied the nonlinear cross-prediction test to assessthe dynamic properties of the EEG and showed that ESESlike other types of epileptic EEG activities seems to reflecthighly nonlinear and possibly low-dimensional dynamicswhereas non-ESES waking EEG seems to correspond withlinear stochastic dynamics [17] In the recent years a seriesof entropy-based approaches have been widely used sincethey can quantify the complexity (regularity) of an EEGsignal [18 19] The entropy of the EEG may act as a reliableindicator of changes in cortical neuronal interactions andtruly reflect the intracortical information flow [20] and thusthe term ldquoentropyrdquo may be more than merely a statisticalmeasure of EEG patterns which are well exploited usingentropies and it helps in providing distinguishable variationfor normal and abnormal biomedical signals [21 22] Abasoloet al applied the approximate entropy (ApEn) to analysethe EEG background activity of Alzheimerrsquos disease (AD)patients and age-matched controls They found that ApEn issignificantly lower in the AD patients at electrodes P3 and P4[23]Then the spectral entropy (SpecEn) and sample entropy(SampEn) were used to analyse the EEG background activityofADpatients and showed thatADpatients have significantlylower SampEn values than control subjects at electrodes P3P4 O1 and O2 but no differences between AD patientsand control subjectsrsquo EEGs with SpecEn [24] Buriokaet al found that the ApEn values of EEG signals in absenceepilepsy during seizure-free intervals are very similar to thoseof healthy subjects but the EEG signals in absence epilepsyduring seizure intervals produce significant lower ApEnvalues than healthy subjects [25] In the study by Kannathalet al ApEn was used to investigate the epileptic seizuredetection where three other entropy-based features wereextracted and combined with ApEn for studying normal andepileptic EEG signals [26] A novel feature extractionmethodbased on ApEn SampEn and phase entropy was proposedfor diagnosing the epileptic EEG signals and showed that theextracted features with fuzzy classifier are able to differentiatethe EEGs with a high accuracy [27] The high identificationaccuracy was also reported in the study by Song et al [28]in which they developed a new scheme of automatic epilepticseizure detection on the basis of SampEn feature extraction

Recently Bandt and Pompe proposed the permutationentropy (PE) method to measure the irregularity (complex-ity) of nonstationary time series [29] The basic idea is toconsider order relations between the values of a time seriesrather than the values themselves Compared with ApEnand SampEn [21 22] the advantages of the PE methodare its simplicity low complexity in computation withoutfurther model assumptions and robustness in the presence

of observational and dynamical noise [29ndash31] Cao et alused PE to identify various phases of epileptic activity inthe intracranial EEG signals recorded from three patientssuffering from intractable epilepsy [32] Li et al used PE asa feature to predict the absence seizures in genetic absenceepilepsy rats and showed a sharp PE drop after the seizures[33] It was also found that the PE can better extract thepattern of EEG data for the prediction of absence seizure thanthe SampEn measure Nicolaou and Georgiou investigatedthe use of PE as a feature for automated epileptic seizuredetection [34] Bruzzo et al applied PE to detect vigilancechanges and the preictal phase from scalp EEG in threeepileptic patients [35] These results showed that the EEGduring epileptic seizures is characterized by a lower value ofPE than the normal EEG It was found that there is a goodseparability between the seizure-free phase and the preseizurephase and the changes of PE values during the preseizurephase and seizure onset coincide with changes in vigilancestate [35]

In terms of classifiers lots of methodologies have beenproposed and applied to process and discriminate biomedicalsignals [36ndash38] such as electromyography [39 40] and EEGsignals [41 42] In particular artificial neural networks havebeen utilized as the most common method for classifyingthe EEGs Moreover fuzzy set theory plays an importantrole in dealing with uncertainty when making decisions inmedical applications Therefore fuzzy sets have attractedthe growing attention and interest in data analysis decisionmaking pattern recognition diagnostics and so forth [4344] Neuro-fuzzy systems are fuzzy systems which use ANNstheory in order to determine their properties (fuzzy setsand fuzzy rules) by processing data samples [45] A specificapproach in neuro-fuzzy development is the adaptive neuro-fuzzy inference system (ANFIS) which has shown significantresults in classification of EEG signals Kannathal et alproposed a novel classification framework based on entropymeasures and ANFIS classifier to distinguish normal andepileptic EEG signals [26] Guler and Ubeyli proposed anew scheme using ANFIS and wavelet transform as theclassifier which can identify five types of EEG signals witha recognition rate greater than 98 [45] Ubeyli proposeda system using Lyapunov exponents of EEG signals andANFIS as the classifier which can identify these five types ofEEG signals with a recognition rate greater than 99 [46]In the study by Yildiz et al [47] a wavelet entropy-ANFISframework is proposed for classifying a state of vigilance asalert drowsy or sleep state on an ongoing EEG recording Aclassification accuracy of more than 98 is achieved Theseresults show that ANFIS has potential in classifying the EEGsignals

In this study a new approach based on ANFIS employingPE and SampEn measures was presented for classification ofbackground EEG signals from ESES patients and controlsThe proposed technique involved training the two ANFISclassifiers to classify the two classes of the EEG signals whenPE and SampEn of the EEG signals were used as inputs Thegoal was to find a clear differentiation between backgroundEEG corresponding to a sample set of ESES patients and thatcorresponding to healthy control individuals The paper is

The Scientific World Journal 3

Table 1 Summary of the EEG data

Set 1 Set 2Subjects 10 healthy subjects 10 ESES patients

Age 3ndash9 years4 males and 6 females

3ndash9 years4 males and 6 females

Patientrsquos state Awake and eyes open(normal)

Awake and eyes open(no spikes)

Number of epochs 100 100Epoch duration (s) 8 8

organized as follows Section 2 presents a description of thedata used in this work and briefly describes the extractedfeatures and classifiers that were used Section 3 presents theresults obtained Finally conclusions are given in Section 4

2 Materials and Methods

21 EEG Data The EEG data used in this study consists oftwo different sets The first set includes EEG recordings thatwere collected from 10 right-handed healthy subjects Thesubjects were awake and relaxed with their eyes open 10016-channel EEG epochs of 8 s duration were selected andcut out from each continuous EEG recording after visualinspection for artifacts for example due to muscle activityor eye movements The second set was obtained from 10patients with ESES all right-handed The data set consists ofEEG recordings during wakeful state Similar to healthy datanoise-free segments are selected from the EEG recordingswith ESES patients and used for the analysis All EEG datawere recorded by theNihonKohdendigital videoEEG systemfrom a standard international 10ndash20-electrode placement(Fp1 Fp2 F3 F4 C3 C4 P3 P4 O1 O2 F7 F8 T3 T4 T5and T6) They were sampled at a frequency of 500Hz usinga 16-bit analogue-to-digital converter and filtered within afrequency band from 05 to 35Hz

The study protocol had previously been approved by theEthics Committee of Peking University First Hospital andthe patients had signed informed consent that their clinicaldata might be used and published for research purposes Asummary of the data set is given in Table 1 A sample of EEGepochs from each of the two data sets is plotted in Figure 1

22 Sample Entropy Sample entropy (SampEn) is an algo-rithm derived from approximate entropy (ApEn) [21] Intro-duced by Pincus et al [48] ApEn is a technique that isuseful in determining changing system complexity and itfinds application in biomedical research [49]The first step incomputing ApEn of an EEG series 119909

1 1199092 119909

119873 is to form

a sequence of vectors 1198831 1198832 119883

119873minus119898+1in 119877119898 defined by

119883119894= [119909119894 119909119894+1 119909

119905+119898minus1] and 119883

119895= [119909119895 119909119895+1 119909

119895+119898minus1]

Next we define for each 119894

119862119898

119894(119903) =

119894

119873 minus 119898 + 1sum119895=1

120579 (119903 minus 119889 (119883119894 119883119895)) (1)

where 120579 is the standard Heaviside function and 120579(119909) = 1 for119909 gt 0 120579(119909) = 0 otherwise 119903 is a tolerance threshold and119889(119883119894 119883119895) is a distance measure defined by

119889 (119883119894 119883119895) = max (10038161003816100381610038161003816119909119894+119896minus1 minus 119909119895+119896minus1

10038161003816100381610038161003816) 119896 = 1 2 119898

(2)

Then we define 120601119898(119903) as

120601119898(119903) =

1

119873 minus 119898 + 1

119873minus119898+1

sum119894=1

log119862119898119894(119903) (3)

For fixed119898 and 119903 ApEn is given by the following formula

ApEn (119898 119903119873) = 120601119898 (119903) minus 120601119898+1 (119903) (4)

which is basically the logarithmic likelihood that runsof patterns of length 119898 that are close (within 119903) will remainclose on next incremental comparisonsThemore regular theEEG is the smaller the ApEn will be The exact value of theApEn(119898 119903119873) will depend on three parameters119873 (length ofthe time series)119898 (length of sequences to be compared) and119903 (tolerance threshold for accepting matches)

The ApEn specifies a tolerance threshold and so maybe better than spectral entropy in the quantification ofcomplexity of EEG recording [50]The disadvantage of ApEnis that it is heavily dependent on the record length and is oftenlower than expected for short records Another disadvantageis that ApEn lacks relative consistency [21] To overcomethe disadvantages of ApEn a sample entropy (SampEn)was proposed to replace ApEn By excluding self-matches[21] SampEn reduces the computing time by one-half incomparisonwithApEn Another advantage of SampEn is thatit is largely independent of record length and displays relativeconsistency [51] The key idea that differentiates SampEnfrom ApEn is using the correlation sum 119862119898(119903) in the entropydefinition instead of the 120601119898(119903) functions defined in (3)mdashpractically the position of the log function changes ThusRichman and Moorman defined sample entropy as

SampEn (119898 119903119873) = log 119862119898(119903)

119862119898+1 (119903) (5)

The choice of input parameters has been discussed byPincus and Goldberger in [52] They concluded that for119898 =2 values of 119903 from 01 to 025 SD (the standard deviation ofthe signal) produce good statistical validity of SampEn In thisstudy SampEn was estimated with119898 = 2 and 119903 = 02 times SD ofthe EEG epoch

23 Permutation Entropy Bandt and Pompe proposed a newpermutation method to map a continuous time series onto asymbolic sequence [29] where the statistics of the symbolicsequences was called permutation entropy (PE) PE refersto the local order structure of the time series which cangive a quantitative complexity measure for a dynamical timeseries [53] Given a time series 119909

1 1199092 119909

119873 an embedding

procedure was used to generate 119873 minus (119898 minus 1)119897 vectors1198831 1198832 119883

119873minus(119898minus1)119897defined by 119883

119905= [119909119905 119909119905+119897 119909

119905+(119898minus1)119897]


Fp1Fp2F3F4C3C4P3P4O1

O2

F7F8T3T4T5T6

(a)

1 s 40uv

(b)

Figure 1 Sample EEG epochs from both ESES patient (a) and control subject (b)

with the embedding dimension 119898 and the lag 119897 The vector119883119905can be rearranged in an ascending order as [119909

119905+(1198951minus1)119897

le

119909119905+(1198952minus1)119897 le 119909

119905+(119895119898minus1)119897] For 119898 different numbers there

will be 119898 possible order patterns 120587 which are also calledpermutations Then we can count the occurrences of theorder pattern 120587

119894 which is denoted as 119862(120587

119894) 119894 = 1 2 119898 Its

relative frequency is calculated by119901(120587) = 119862(120587)(119873minus(119898minus1)119897)The PE is defined as

PE = minus119898

sum119898=1

119901 (120587) ln119901 (120587) (6)

The largest value of PE is log(119898) which means that thetime series is completely random the smallest value of PEis zero indicating that the time series is very regular Moredetails can be found in [29]

PE calculation depends on the selection of dimension 119898and lag 119897 When 119898 is too small (less than 3) the schemewill not work well since there are only a few distinct statesfor EEG recordings On the other hand the length of EEGrecording should be larger than 119898 in order to achieve aproper differentiation between stochastic and deterministicdynamics [31] In order to allow every possible order patternof dimension 119898 to occur in a time series of length 119873 thecondition 119898 le 119873 minus (119898 minus 1)119897 must hold Moreover 119873 ≫

119898 + (119898 minus 1)119897 is required to avoid undersampling [54] Inthis study we therefore choose the dimension 119898 = 5 whencalculating PEThe lag 119897 is referred to as the number of samplepoints spanned by each section of the vectorThe importanceof the lag is that it gives the resultant fraction characteristicsof the vector In practice an autocorrelation function (ACF)of a signal can be employed to automated determination ofthe lag 119897 An optimal lag can be found at the point where theACF has firstly decayed to 119890minus1 of its peak value [55]

24 Adaptive Neuro-Fuzzy Inference System The ANFISdescribed by Jang [56] is adopted to evaluate the abilityand effectiveness of the above entropy measures in classify-ing the EEG from the ESES patients and control subjectsThe ANFIS learns features in the data set and adjusts thesystem parameters according to a given error criterion Ithas been widely used in analysing the biological signals Inorder to improve the generalization ANFIS classifiers aretrained with the backpropagation gradient descent methodin combination with the least squares method In this studytwo ANFIS classifiers are trained with the backpropagationgradient descent method in combination with the leastsquares method when 16 features (dimension of the extractedfeature vectors entropy measures from 16-channel EEG) areused as inputsThe samples with target outputs ESES patientsand control subjects are given the binary target values of (10) and (0 1) respectively The fuzzy rule architecture of theANFIS classifiers was designed by using a generalized bell-shaped membership function defined as follows

120583119895119894(119909119894) = (1 + [

119909119894minus 119888119895119894

119886119895119894

]

2119887119895119894

)

minus1

(7)

where (119886119895119894 119887119895119894 119888119895119894) are adaptable parameters

Next two first-order Sugeno-type ANFIS models with16 inputs and one output are implemented The first-orderSugeno fuzzy models have rules of the following form

119877119894 IF (119909

1is 1198801198941sdot sdot sdot 119909119898is 119880119894119898)

THEN 119910 is 119892119894(1199091 119909119898) = 1198870+ 11988711199091+ sdot sdot sdot + 119887

119898119909119898

(8)

where 119877119894is the 119894th rule of the fuzzy system 119909

119894(119894 = 1 119898)

are the inputs to the fuzzy system and 119910 is the output of the


38

4

42

44

46

48

ESES patientsNormal

Channel

Fp1

Fp2 F3 F4 C3 C4 P3 P4 O1

O2 F7 F8 T3 T4 T5 T6

Perm

utat

ion

entro

py

Figure 2 The averaged PE on channel of all EEG recordingsgrouped by ESES patients and normal control subjects Symbolsrepresent the mean values of PE for each group and bars representthe standard error

fuzzy system 119887119894(119894 = 0 1 119898) are adaptable parameters

The ANFIS output is given by

119865 =sum119895119892119895(1198861 1198862 119886

119898)Π119894120583119880119895119894(119886119894)

sum119895Π119894120583119880119895119894(119886119894)

(9)

where 120583119880119895119894(119886119894) is the degree of membership of 119886

119894(119894 =

1 2 119898) to the antecedent linguistic term 119880119895119894for the 119894th

rule of the fuzzy system Each ANFIS classifier is imple-mented by using the MATLAB software package (MATLABversion 70 with fuzzy logic toolbox)

3 Results

31 Entropy Measures of EEG EEG epochs from both ESESpatients and normal control subjects are investigated in thisstudy First PE is applied to analyse the EEG recordings with119898 = 5 for channels Fp1 Fp2 F3 F4 C3 C4 P3 P4 O1 O2F7 F8 T3 T4 T5 and T6 The results have been averagedbased on all the artefact-free 8 s epochs for each channelTheaveraged PEs of all channels are shown in Figure 2 Symbolsrepresent the mean values of PE for each group and barsrepresent the standard error It can be found that the PEvalues of EEG epochs in normal control subjects are muchlarger than those in ESES patients The PE values (mean plusmnSD) for the control subjects and ESES patients and the 119875values of the one-way ANOVA test performed to examine thedifferences between both groups are summarized in Table 2It can be seen that ESES patients have significant lower PEvalues at all 16 electrodes These results suggest that EEGactivity of ESES patients is less complex (more regular) thanin a normal control subject This result supports the viewthat ESES like other types of epileptic EEG activity wouldreflect low complex and high nonlinear dynamics whereasnon-ESES waking EEGwould correspond with high complexdynamics

To compare the extracted entropy information of EEGbetween PE and SampEn methods SampEn is estimated for

Table 2 The average PE values (mean plusmn SD) of the EEGs for thenormal control subjects and ESES patients for all channels

Electrode Normal subjects ESES patients 119875 valueFp1 4563 plusmn 0088 4304 plusmn 0224 119875 lt 005

Fp2 4555 plusmn 0089 4349 plusmn 0184 119875 lt 005

F3 4606 plusmn 0055 4292 plusmn 0240 119875 lt 005


C3 4550 plusmn 0116 4341 plusmn 0192 119875 lt 005


P3 4556 plusmn 0117 4311 plusmn 0212 119875 lt 005


O1 4454 plusmn 0204 4272 plusmn 0273 119875 lt 005




T3 4598 plusmn 0072 4331 plusmn 0220 119875 lt 005




19

2

21

22

23

ESES patientsNormal

Sam

ple e

ntro

py

Channel

Fp1

Fp2 F3 F4 C3 C4 P3 P4 O1

O2 F7 F8 T3 T4 T5 T6

Figure 3 The averaged SampEn on channel of all EEG recordingsgrouped by ESES patients and normal control subjects Symbolsrepresent the mean values of SampEn for each group and barsrepresent the standard error

channels Fp1 Fp2 F3 F4 C3 C4 P3 P4 O1 O2 F7 F8T3 T4 T5 and T6 with 119898 = 2 and 119903 = 02 times SD of theoriginal data sequenceThe averaged SampEns of all channelsare shown in Figure 3 It can be found that the SampEn valuesof EEG epochs in normal control subjects are also largerthan those in ESES patients However the SampEn valuesin control subjects and ESES patients are more overlappedthan those of the PE values Then the SampEn values (meanplusmn SD) for the control subjects and ESES patients and the 119875values of the one-way ANOVA test performed to examine thedifferences between both groups are summarized in Table 3It can be seen that ESES patients have significant lowerSampEn values at all 16 electrodes


Table 3 The average SampEn values (mean plusmn SD) of the EEGs forthe normal control subjects and ESES patients for all channels

















Table 4 Classification results with PE measure

Desired result Output resultESES patients Normal subjects

ESES patients 96 4Normal subjects 18 82

32 Classification As shown above both PE and SampEnvalues of EEG were significantly different between ESESpatients and normal control subjects The performance ofthe above measures to discriminate among groups is alsoevaluated by means of ANFIS classifier and 10-fold cross-validations are employed to demonstrate the accuracy ofclassification First the ability of the PE in classifying differentEEG epochs is evaluated using the ANFIS Two ANFIS clas-sifiers are trained with the backpropagation gradient descentmethod in combination with the least squares method whenthe calculated PE values are used as input Each of the ANFISclassifiers is trained so that they are likely to bemore accuratefor one state of EEG signals than the other state Samples withtarget outputs sets are given the binary target values of (1 0)and (0 1) respectively Each ANFIS classifier is implementedby using the MATLAB software package (MATLAB version70 with fuzzy logic toolbox) The classification results areillustrated in Table 4 Of 200 EEG epochs in two groups178 are classified correctly Only 4 normal EEG epochs areclassified incorrectly by ANFIS as ESES EEG epochs and18 ESES EEG epochs are classified as normal EEG epochsThe classification accuracy was 890 which is defined as thepercentage ratio of the number of epochs correctly classifiedto the total number of epochs considered for classification

Then in order to compare the classification accuracy ofPE method with that of the SampEn method the calculatedSampEn values were used as the input data in the ANFISclassifiers and 10-fold cross-validations were employed to

Table 5 Classification results with SampEn measure



demonstrate the performance of classificationThe classifica-tion results are listed in Table 5 Of 200 EEG epochs in twogroups 164 were classified correctly The total classificationaccuracy was 820 Therefore it is found that the PE mea-sures can provide a better separability between ESES patientsand normal control subjects than the SampEn measures

4 Conclusions

In this study we have analysed the complexity characteristicsin background EEG signals from ESES patients and controlsusing the entropy measures Although the background EEGmarked was indeed ldquonormalrdquo to standard visual inspectionthe proposed methodology based on entropy measures plusANOVA statistical test demonstrates that the backgroundEEG in ESES patients does differ from that in controls It canbe seen that there is a significant increase of the calculated PEand SampEn values of the EEG epochs from ESES patientsto control subjects Then a new approach based on ANFISemploying entropy measures was presented for classificationof background EEG signals from ESES patients and controlsThe two ANFIS classifiers were used to classify two classes ofEEG epochs when the PE and SampEn of the EEG epochswere used as inputs The experimental results showed thatthe classification accuracy 89 based on the PE measuresis much higher than that with the SampEn measures 82These results suggest that the proposed ANFIS combinedwith PE measures might be a potential tool to classify thebackground EEG from ESES patients and normal controlsubjects Our next goal is to confirm the results presentedhere in a much larger clinical cohort of ESES patients

Conflict of Interests

The authors declare that there is no conflict of interestsregarding the publication of this paper

Acknowledgments

This research was supported by National Natural ScienceFoundation of China (61025019 61105027) Beijing NaturalScience Foundation (4143063) the Fundamental ResearchFunds for the Central Universities and the Peking-TsinghuaCenter for Life Sciences

References

[1] G Patry S Lyagoubi andCA Tassinari ldquoSubclinical ldquoelectricalstatus epilepticusrdquo induced by sleep in children A clinicaland electroencephalographic study of six casesrdquo Archives ofNeurology vol 24 no 3 pp 242ndash252 1971


[2] C A Tassinari G Rubboli L Volpi et al ldquoEncephalopathywithelectrical status epilepticus during slow sleep or ESES syndromeincluding the acquired aphasiardquo Clinical Neurophysiology vol111 supplement 2 pp S94ndashS102 2000

[3] F B J Scholtes M P H Hendriks andWO Renier ldquoCognitivedeterioration and electrical status epilepticus during slow sleeprdquoEpilepsy and Behavior vol 6 no 2 pp 167ndash173 2005

[4] S Raha U Shah and V Udani ldquoNeurocognitive and neurobe-havioral disabilities in epilepsy with electrical status epilepticusin slow sleep (ESES) and related syndromesrdquo Epilepsy andBehavior vol 25 pp 381ndash385 2012

[5] M Siniatchkin K Groening J Moehring et al ldquoNeuronalnetworks in children with continuous spikes and waves duringslow sleeprdquo Brain vol 133 no 9 pp 2798ndash2813 2010

[6] G Buzsaki Rhythms of the Brain Oxford University PressOxford UK 2006

[7] R A Sarkis and J W Lee ldquoQuantitative EEG in hospitalencephalopathy review and microstate analysisrdquo Journal ofClinical Neurophysiology vol 30 pp 526ndash530 2013

[8] W Klimesch P Sauseng and S Hanslmayr ldquoEEG alphaoscillations the inhibition-timing hypothesisrdquo Brain ResearchReviews vol 53 no 1 pp 63ndash88 2007

[9] M Scheltens-de Boer ldquoGuidelines for EEG in encephalopathyrelated to ESESCSWS in childrenrdquo Epilepsia vol 50 no 7 pp13ndash17 2009

[10] O A Rosso A Mendes J A Rostas M Hunter and PMoscato ldquoDistinguishing childhood absence epilepsy patientsfrom controls by the analysis of their background brain electri-cal activityrdquo Journal of Neuroscience Methods vol 177 no 2 pp461ndash468 2009

[11] O A Rosso A Mendes R Berretta J A Rostas M Hunterand P Moscato ldquoDistinguishing childhood absence epilepsypatients from controls by the analysis of their background brainelectrical activity (II) a combinatorial optimization approachfor electrode selectionrdquo Journal of Neuroscience Methods vol181 no 2 pp 257ndash267 2009

[12] Z Rogowski I Gath and E Bental ldquoOn the prediction ofepileptic seizuresrdquo Biological Cybernetics vol 42 no 1 pp 9ndash15 1981

[13] L D Iasemidis J C Sackellares H P Zaveri and W JWilliams ldquoPhase space topography and the lyapunov exponentof electrocorticograms in partial seizuresrdquo Brain Topographyvol 2 no 3 pp 187ndash201 1990

[14] R G Andrzejak K Lehnertz F Mormann C Rieke P Davidand C E Elger ldquoIndications of nonlinear deterministic andfinite-dimensional structures in time series of brain electricalactivity dependence on recording region and brain staterdquoPhysical Review E Statistical Nonlinear and SoftMatter Physicsvol 64 no 6 Article ID 061907 2001

[15] V Navarro J Martinerie M Le van Quyen et al ldquoSeizureanticipation in human neocortical partial epilepsyrdquo Brain vol125 no 3 pp 640ndash655 2002

[16] M Niknazar S R Mousavi S Motaghi A Dehghani BV Vahdat and M B Shamsollahi ldquoA unified approach fordetection of induced epileptic seizures in rats using ECoGsignalsrdquo Epilepsy and Behavior vol 27 pp 355ndash364 2013

[17] R Ferri R ChiaramontiM Elia S AMusumeci A Ragazzoniand C J Stam ldquoNonlinear EEG analysis during sleep inpremature and full-term newbornsrdquo Clinical Neurophysiologyvol 114 no 7 pp 1176ndash1180 2003

[18] J Gao J Hu andW-W Tung ldquoEntropy measures for biologicalsignal analysesrdquoNonlinear Dynamics vol 68 pp 431ndash444 2011

[19] C Paisansathan M D Ozcan Q S Khan V L Baughmanand M S Ozcan ldquoSignal persistence of bispectral index andstate entropy during surgical procedure under sedationrdquo TheScientific World Journal vol 2012 Article ID 272815 5 pages2012

[20] JW SleighDA Steyn-RossM L Steyn-Ross CGrant andGLudbrook ldquoCortical entropy changes with general anaesthesiatheory and experimentrdquo Physiological Measurement vol 25 no4 pp 921ndash934 2004

[21] J S Richman and J R Moorman ldquoPhysiological time-seriesanalysis using approximate and sample entropyrdquoThe AmericanJournal of PhysiologymdashHeart and Circulatory Physiology vol278 no 6 pp H2039ndashH2049 2000

[22] S M Pincus ldquoApproximate entropy as a measure of systemcomplexityrdquo Proceedings of the National Academy of Sciences ofthe United States of America vol 88 pp 2297ndash2301 1991

[23] D Abasolo R Hornero P Espino J Poza C I Sanchez and Rde la Rosa ldquoAnalysis of regularity in the EEG background activ-ity of Alzheimerrsquos disease patients with Approximate EntropyrdquoClinical Neurophysiology vol 116 no 8 pp 1826ndash1834 2005

[24] D Abasolo R Hornero P Espino D Alvarez and JPoza ldquoEntropy analysis of the EEG background activity inAlzheimerrsquos disease patientsrdquo Physiological Measurement vol27 no 3 pp 241ndash253 2006

[25] N Burioka G Cornelissen Y Maegaki et al ldquoApproximateentropy of the electroencephalogram in healthy awake subjectsand absence epilepsy patientsrdquo Clinical EEG and Neurosciencevol 36 no 3 pp 188ndash193 2005

[26] N Kannathal M L Choo U R Acharya and P K SadasivanldquoEntropies for detection of epilepsy in EEGrdquoComputerMethodsand Programs in Biomedicine vol 80 no 3 pp 187ndash194 2005

[27] U R Acharya F Molinari S V Sree S Chattopadhyay K-HNg and J S Suri ldquoAutomated diagnosis of epileptic EEG usingentropiesrdquo Biomedical Signal Processing and Control vol 7 pp401ndash408 2012

[28] Y Song J Crowcroft and J Zhang ldquoAutomatic epileptic seizuredetection in EEGs based on optimized sample entropy andextreme learning machinerdquo Journal of Neuroscience Methodsvol 210 pp 132ndash146 2012

[29] C Bandt and B Pompe ldquoPermutation entropy a naturalcomplexitymeasure for time seriesrdquoPhysical ReviewLetters vol88 no 17 Article ID 174102 2002

[30] M Zanin L Zunino O A Rosso and D Papo ldquoPermutationentropy and itsmain biomedical and econophysics applicationsa reviewrdquo Entropy vol 14 pp 1553ndash1577 2012

[31] G Ouyang C Dang D A Richards and X Li ldquoOrdinalpattern based similarity analysis for EEG recordingsrdquo ClinicalNeurophysiology vol 121 no 5 pp 694ndash703 2010

[32] Y Cao W W Tung J B Gao V A Protopopescu and L MHively ldquoDetecting dynamical changes in time series using thepermutation entropyrdquo Physical Review E Statistical Nonlinearand Soft Matter Physics vol 70 no 4 Article ID 046217 2004

[33] X Li G Ouyang and D A Richards ldquoPredictability analysis ofabsence seizures with permutation entropyrdquo Epilepsy Researchvol 77 no 1 pp 70ndash74 2007

[34] N Nicolaou and J Georgiou ldquoDetection of epileptic electroen-cephalogram based on permutation entropy and support vectormachinesrdquo Expert Systems with Applications vol 39 no 1 pp202ndash209 2012

[35] A A Bruzzo B Gesierich M Santi C A Tassinari NBirbaumer and G Rubboli ldquoPermutation entropy to detect


vigilance changes and preictal states from scalp EEG in epilepticpatients A preliminary studyrdquoNeurological Sciences vol 29 no1 pp 3ndash9 2008

[36] S Baek C A Tsai and J J Chen ldquoDevelopment of biomarkerclassifiers from high-dimensional datardquo Briefings in Bioinfor-matics vol 10 no 5 pp 537ndash546 2009

[37] R Palaniappan K Sundaraj and N U Ahamed ldquoMachinelearning in lung sound analysis a systematic reviewrdquo Biocyber-netics and Biomedical Engineering vol 33 pp 129ndash135 2013

[38] H Zhou H Hu H Liu and J Tang ldquoClassification of upperlimb motion trajectories using shape featuresrdquo IEEE Trans-actions on Systems Man and Cybernetics C Applications andReviews vol 42 pp 970ndash982 2012

[39] Z J Ju G X Ouyang M Wilamowska-Korsak and H HLiu ldquoSurface EMG based hand manipulation identification vianonlinear feature extraction and classificationrdquo IEEE SensorsJournal vol 13 pp 3302ndash3311 2013

[40] A Phinyomark P Phukpattaranont and C Limsakul ldquoFeaturereduction and selection for EMG signal classificationrdquo ExpertSystems with Applications vol 39 no 8 pp 7420ndash7431 2012

[41] U R Acharya S V Sree G Swapna R J Martis and J S SurildquoAutomated EEG analysis of epilepsy a reviewrdquo Knowledge-Based Systems vol 45 pp 147ndash165 2013

[42] M Besserve K Jerbi F Laurent S Baillet J Martinerie and LGarnero ldquoClassification methods for ongoing EEG and MEGsignalsrdquo Biological Research vol 40 no 4 pp 415ndash437 2007

[43] D Nauck and R Kruse ldquoObtaining interpretable fuzzy clas-sification rules from medical datardquo Artificial Intelligence inMedicine vol 16 no 2 pp 149ndash169 1999

[44] L I Kuncheva and F Steimann ldquoFuzzy diagnosis editorialrdquoArtificial Intelligence inMedicine vol 16 no 2 pp 121ndash128 1999

[45] I Guler and E D Ubeyli ldquoAdaptive neuro-fuzzy inference sys-tem for classification of EEG signals using wavelet coefficientsrdquoJournal of Neuroscience Methods vol 148 no 2 pp 113ndash1212005

[46] E D Ubeyli ldquoAutomatic detection of electroencephalographicchanges using adaptive neuro-fuzzy inference system employ-ing Lyapunov exponentsrdquo Expert Systems with Applications vol36 no 5 pp 9031ndash9038 2009

[47] A Yildiz M Akin M Poyraz and G Kirbas ldquoApplicationof adaptive neuro-fuzzy inference system for vigilance levelestimation by using wavelet-entropy feature extractionrdquo ExpertSystems with Applications vol 36 no 4 pp 7390ndash7399 2009

[48] S M Pincus I M Gladstone and R A Ehrenkranz ldquoAregularity statistic for medical data analysisrdquo Journal of ClinicalMonitoring vol 7 no 4 pp 335ndash345 1991

[49] S M Pincus ldquoApproximate entropy as a measure of irregularityfor psychiatric serial metricsrdquo Bipolar Disorders I vol 8 no 5pp 430ndash440 2006

[50] J Bruhn H Ropcke B Rehberg T Bouillon and A HoeftldquoElectroencephalogram approximate entropy correctly classi-fies the occurrence of burst suppression pattern as increasinganesthetic drug effectrdquo Anesthesiology vol 93 no 4 pp 981ndash985 2000

[51] D E Lake J S Richman M Pamela Griffin and J RandallMoorman ldquoSample entropy analysis of neonatal heart ratevariabilityrdquo The American Journal of PhysiologymdashRegulatoryIntegrative and Comparative Physiology vol 283 no 3 ppR789ndashR797 2002

[52] S M Pincus and A L Goldberger ldquoPhysiological time-seriesanalysis what does regularity quantifyrdquoThe American Journal

of PhysiologymdashHeart and Circulatory Physiology vol 266 no 4pp H1643ndashH1656 1994

[53] G Ouyang X Li C Dang and D A Richards ldquoDeterministicdynamics of neural activity during absence seizures in ratsrdquoPhysical Review E Statistical Nonlinear and SoftMatter Physicsvol 79 Article ID 041146 2009

[54] M Staniek and K Lehnertz ldquoParameter selection for permu-tation entropy measurementsrdquo International Journal of Bifurca-tion and Chaos vol 17 no 10 pp 3729ndash3733 2007

[55] X Li and G Ouyang ldquoEstimating coupling direction betweenneuronal populations with permutation conditional mutualinformationrdquo NeuroImage vol 52 no 2 pp 497ndash507 2010

[56] J S R Jang ldquoANFIS adaptive-network-based fuzzy inferencesystemrdquo IEEE Transactions on Systems Man and Cyberneticsvol 23 no 3 pp 665ndash685 1993

International Journal of

AerospaceEngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

RoboticsJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014


Active and Passive Electronic Components

Control Scienceand Engineering

Journal of



RotatingMachinery


Hindawi Publishing Corporation httpwwwhindawicom

Journal ofEngineeringVolume 2014

Submit your manuscripts athttpwwwhindawicom

VLSI Design



Shock and Vibration


Civil EngineeringAdvances in

Acoustics and VibrationAdvances in



Electrical and Computer Engineering

Journal of

Advances inOptoElectronics


Volume 2014

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

SensorsJournal of


Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014


Chemical EngineeringInternational Journal of Antennas and

Propagation




Navigation and Observation



DistributedSensor Networks



as Fourier transforms and spectral analysis [12] to nonlin-ear methods such as Lyapunov exponents [13] correlationdimension [14] and similarity [15 16] Due to the complexinterconnections between billions of neurons the recordedEEG signals are complex nonlinear nonstationary andrandom in natureTherefore the classification of EEG signalsusing nonlinear methods that detect and quantify nonlinearmechanisms and thereby better reflect the characteristics ofthe EEG signals Nonlinear features may be able to unearththe hidden complexities existing in the EEG time series Ferriet al applied the nonlinear cross-prediction test to assessthe dynamic properties of the EEG and showed that ESESlike other types of epileptic EEG activities seems to reflecthighly nonlinear and possibly low-dimensional dynamicswhereas non-ESES waking EEG seems to correspond withlinear stochastic dynamics [17] In the recent years a seriesof entropy-based approaches have been widely used sincethey can quantify the complexity (regularity) of an EEGsignal [18 19] The entropy of the EEG may act as a reliableindicator of changes in cortical neuronal interactions andtruly reflect the intracortical information flow [20] and thusthe term ldquoentropyrdquo may be more than merely a statisticalmeasure of EEG patterns which are well exploited usingentropies and it helps in providing distinguishable variationfor normal and abnormal biomedical signals [21 22] Abasoloet al applied the approximate entropy (ApEn) to analysethe EEG background activity of Alzheimerrsquos disease (AD)patients and age-matched controls They found that ApEn issignificantly lower in the AD patients at electrodes P3 and P4[23]Then the spectral entropy (SpecEn) and sample entropy(SampEn) were used to analyse the EEG background activityofADpatients and showed thatADpatients have significantlylower SampEn values than control subjects at electrodes P3P4 O1 and O2 but no differences between AD patientsand control subjectsrsquo EEGs with SpecEn [24] Buriokaet al found that the ApEn values of EEG signals in absenceepilepsy during seizure-free intervals are very similar to thoseof healthy subjects but the EEG signals in absence epilepsyduring seizure intervals produce significant lower ApEnvalues than healthy subjects [25] In the study by Kannathalet al ApEn was used to investigate the epileptic seizuredetection where three other entropy-based features wereextracted and combined with ApEn for studying normal andepileptic EEG signals [26] A novel feature extractionmethodbased on ApEn SampEn and phase entropy was proposedfor diagnosing the epileptic EEG signals and showed that theextracted features with fuzzy classifier are able to differentiatethe EEGs with a high accuracy [27] The high identificationaccuracy was also reported in the study by Song et al [28]in which they developed a new scheme of automatic epilepticseizure detection on the basis of SampEn feature extraction

Recently Bandt and Pompe proposed the permutationentropy (PE) method to measure the irregularity (complex-ity) of nonstationary time series [29] The basic idea is toconsider order relations between the values of a time seriesrather than the values themselves Compared with ApEnand SampEn [21 22] the advantages of the PE methodare its simplicity low complexity in computation withoutfurther model assumptions and robustness in the presence

of observational and dynamical noise [29ndash31] Cao et alused PE to identify various phases of epileptic activity inthe intracranial EEG signals recorded from three patientssuffering from intractable epilepsy [32] Li et al used PE asa feature to predict the absence seizures in genetic absenceepilepsy rats and showed a sharp PE drop after the seizures[33] It was also found that the PE can better extract thepattern of EEG data for the prediction of absence seizure thanthe SampEn measure Nicolaou and Georgiou investigatedthe use of PE as a feature for automated epileptic seizuredetection [34] Bruzzo et al applied PE to detect vigilancechanges and the preictal phase from scalp EEG in threeepileptic patients [35] These results showed that the EEGduring epileptic seizures is characterized by a lower value ofPE than the normal EEG It was found that there is a goodseparability between the seizure-free phase and the preseizurephase and the changes of PE values during the preseizurephase and seizure onset coincide with changes in vigilancestate [35]

In terms of classifiers lots of methodologies have beenproposed and applied to process and discriminate biomedicalsignals [36ndash38] such as electromyography [39 40] and EEGsignals [41 42] In particular artificial neural networks havebeen utilized as the most common method for classifyingthe EEGs Moreover fuzzy set theory plays an importantrole in dealing with uncertainty when making decisions inmedical applications Therefore fuzzy sets have attractedthe growing attention and interest in data analysis decisionmaking pattern recognition diagnostics and so forth [4344] Neuro-fuzzy systems are fuzzy systems which use ANNstheory in order to determine their properties (fuzzy setsand fuzzy rules) by processing data samples [45] A specificapproach in neuro-fuzzy development is the adaptive neuro-fuzzy inference system (ANFIS) which has shown significantresults in classification of EEG signals Kannathal et alproposed a novel classification framework based on entropymeasures and ANFIS classifier to distinguish normal andepileptic EEG signals [26] Guler and Ubeyli proposed anew scheme using ANFIS and wavelet transform as theclassifier which can identify five types of EEG signals witha recognition rate greater than 98 [45] Ubeyli proposeda system using Lyapunov exponents of EEG signals andANFIS as the classifier which can identify these five types ofEEG signals with a recognition rate greater than 99 [46]In the study by Yildiz et al [47] a wavelet entropy-ANFISframework is proposed for classifying a state of vigilance asalert drowsy or sleep state on an ongoing EEG recording Aclassification accuracy of more than 98 is achieved Theseresults show that ANFIS has potential in classifying the EEGsignals

In this study a new approach based on ANFIS employingPE and SampEn measures was presented for classification ofbackground EEG signals from ESES patients and controlsThe proposed technique involved training the two ANFISclassifiers to classify the two classes of the EEG signals whenPE and SampEn of the EEG signals were used as inputs Thegoal was to find a clear differentiation between backgroundEEG corresponding to a sample set of ESES patients and thatcorresponding to healthy control individuals The paper is














1 1199092 119909

119873 is to form



119883119894= [119909119894 119909119894+1 119909

119905+119898minus1] and 119883

119895= [119909119895 119909119895+1 119909

119895+119898minus1]


119862119898

119894(119903) =

119894

119873 minus 119898 + 1sum119895=1

120579 (119903 minus 119889 (119883119894 119883119895)) (1)



10038161003816100381610038161003816) 119896 = 1 2 119898

(2)

Then we define 120601119898(119903) as

120601119898(119903) =

1

119873 minus 119898 + 1

119873minus119898+1

sum119894=1

log119862119898119894(119903) (3)


ApEn (119898 119903119873) = 120601119898 (119903) minus 120601119898+1 (119903) (4)



SampEn (119898 119903119873) = log 119862119898(119903)

119862119898+1 (119903) (5)



1 1199092 119909

119873 an embedding



119905= [119909119905 119909119905+119897 119909

119905+(119898minus1)119897]



O2

F7F8T3T4T5T6

(a)

1 s 40uv

(b)



119905+(1198951minus1)119897

le

119909119905+(1198952minus1)119897 le 119909




119894) 119894 = 1 2 119898 Its


PE = minus119898

sum119898=1

119901 (120587) ln119901 (120587) (6)





120583119895119894(119909119894) = (1 + [

119909119894minus 119888119895119894

119886119895119894

]

2119887119895119894

)

minus1

(7)



119877119894 IF (119909



119898119909119898

(8)


119894(119894 = 1 119898)



38

4

42

44

46

48

ESES patientsNormal

Channel

Fp1

Fp2 F3 F4 C3 C4 P3 P4 O1

O2 F7 F8 T3 T4 T5 T6

Perm

utat

ion

entro

py




119865 =sum119895119892119895(1198861 1198862 119886

119898)Π119894120583119880119895119894(119886119894)

sum119895Π119894120583119880119895119894(119886119894)

(9)


119894(119894 =



3 Results




















19

2

21

22

23

ESES patientsNormal

Sam

ple e

ntro

py

Channel

Fp1

Fp2 F3 F4 C3 C4 P3 P4 O1

O2 F7 F8 T3 T4 T5 T6






























4 Conclusions




Acknowledgments


References































































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation






















1 1199092 119909

119873 is to form



119883119894= [119909119894 119909119894+1 119909

119905+119898minus1] and 119883

119895= [119909119895 119909119895+1 119909

119895+119898minus1]


119862119898

119894(119903) =

119894

119873 minus 119898 + 1sum119895=1

120579 (119903 minus 119889 (119883119894 119883119895)) (1)



10038161003816100381610038161003816) 119896 = 1 2 119898

(2)

Then we define 120601119898(119903) as

120601119898(119903) =

1

119873 minus 119898 + 1

119873minus119898+1

sum119894=1

log119862119898119894(119903) (3)


ApEn (119898 119903119873) = 120601119898 (119903) minus 120601119898+1 (119903) (4)



SampEn (119898 119903119873) = log 119862119898(119903)

119862119898+1 (119903) (5)



1 1199092 119909

119873 an embedding



119905= [119909119905 119909119905+119897 119909

119905+(119898minus1)119897]



O2

F7F8T3T4T5T6

(a)

1 s 40uv

(b)



119905+(1198951minus1)119897

le

119909119905+(1198952minus1)119897 le 119909




119894) 119894 = 1 2 119898 Its


PE = minus119898

sum119898=1

119901 (120587) ln119901 (120587) (6)





120583119895119894(119909119894) = (1 + [

119909119894minus 119888119895119894

119886119895119894

]

2119887119895119894

)

minus1

(7)



119877119894 IF (119909



119898119909119898

(8)


119894(119894 = 1 119898)



38

4

42

44

46

48

ESES patientsNormal

Channel

Fp1

Fp2 F3 F4 C3 C4 P3 P4 O1

O2 F7 F8 T3 T4 T5 T6

Perm

utat

ion

entro

py




119865 =sum119895119892119895(1198861 1198862 119886

119898)Π119894120583119880119895119894(119886119894)

sum119895Π119894120583119880119895119894(119886119894)

(9)


119894(119894 =



3 Results




















19

2

21

22

23

ESES patientsNormal

Sam

ple e

ntro

py

Channel

Fp1

Fp2 F3 F4 C3 C4 P3 P4 O1

O2 F7 F8 T3 T4 T5 T6






























4 Conclusions




Acknowledgments


References































































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation











O2

F7F8T3T4T5T6

(a)

1 s 40uv

(b)



119905+(1198951minus1)119897

le

119909119905+(1198952minus1)119897 le 119909




119894) 119894 = 1 2 119898 Its


PE = minus119898

sum119898=1

119901 (120587) ln119901 (120587) (6)





120583119895119894(119909119894) = (1 + [

119909119894minus 119888119895119894

119886119895119894

]

2119887119895119894

)

minus1

(7)



119877119894 IF (119909



119898119909119898

(8)


119894(119894 = 1 119898)



38

4

42

44

46

48

ESES patientsNormal

Channel

Fp1

Fp2 F3 F4 C3 C4 P3 P4 O1

O2 F7 F8 T3 T4 T5 T6

Perm

utat

ion

entro

py




119865 =sum119895119892119895(1198861 1198862 119886

119898)Π119894120583119880119895119894(119886119894)

sum119895Π119894120583119880119895119894(119886119894)

(9)


119894(119894 =



3 Results




















19

2

21

22

23

ESES patientsNormal

Sam

ple e

ntro

py

Channel

Fp1

Fp2 F3 F4 C3 C4 P3 P4 O1

O2 F7 F8 T3 T4 T5 T6






























4 Conclusions




Acknowledgments


References































































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation










38

4

42

44

46

48

ESES patientsNormal

Channel

Fp1

Fp2 F3 F4 C3 C4 P3 P4 O1

O2 F7 F8 T3 T4 T5 T6

Perm

utat

ion

entro

py




119865 =sum119895119892119895(1198861 1198862 119886

119898)Π119894120583119880119895119894(119886119894)

sum119895Π119894120583119880119895119894(119886119894)

(9)


119894(119894 =



3 Results




















19

2

21

22

23

ESES patientsNormal

Sam

ple e

ntro

py

Channel

Fp1

Fp2 F3 F4 C3 C4 P3 P4 O1

O2 F7 F8 T3 T4 T5 T6






























4 Conclusions




Acknowledgments


References































































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation




































4 Conclusions




Acknowledgments


References































































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation






































































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation









Research Article Adaptive Neuro-Fuzzy Inference System for ...Research Article Adaptive Neuro-Fuzzy...

Documents

Transcript of Research Article Adaptive Neuro-Fuzzy Inference System for ...Research Article Adaptive Neuro-Fuzzy...