Document 2 bjil 05042017 by IJARMATE

ISSN (ONLINE): 2454-9762 ISSN (PRINT): 2454-9762 Available online at www.ijarmate.com International Journal of Advanced Research in Management, Architecture, Technology and Engineering (IJARMATE) Vol. 3, Special Issue 10, March 2017

PERFORMANCE OF K-NEAREST NEIGHBOR BASED CLASSIFIER FOR CLASSIFICATION OF CARDIOTOCOGRAMS JagannathanD1 1

M.Phil. (PGScholar)Dept of Computer Science ,Dr.C.V.RamanUniversity, Chhattisgarh.

Abstract:Fetalheartrate(FHR)anduterinecontractions(U C)aresimultaneouslyrecordedbyCardiotocography(CT G).TheCTG,whichisoneofthemostcommondiagnosticte chniquesusedtoevaluatematernalandetalwellbeingduringpregnancyandbeforedelivery.Byobservingt heCardiotocographytracepatternsdoctorscanunderstan dthestateofthefetus.Thereareseveralsignalprocessingan dcomputerprogrammingbasedtechniquesforinterpretin ga typicalCardiotocographydata.Evenfewdecadesafterthei ntroductionofcardiotocographyintoclinicalpractice,the predictivecapacityofthesemethodsremainscontroversial andstillinaccurate.InthisresearchworkIproposeanintegr atedmethodologyforCTGanalysisandclassification.Ano velsetoffeatures,derivedfromthetimeandfrequencydom ains,is usedto feedthenewtoolforpatternclassification,namedKNearestneighbor.WeusedAccuracy,Specificity,NPV,Pr ecision,RecallandROCasthemetrictoevaluatetheperfor mance.Thearrivedresultsprovethat,eventhoughthetradit ionalclusteringmethodscanidentifytheNormalCTGpatt erns,theywereincapableofSuspiciousandPathologicpatt erns.Itwasfoundthat,theKNearestneighborbasedclassifierwascapableofidentifyin gNormal,SuspiciousandPathologiccondition,fromthen atureof CTGdata. Keywords— CTG,Datamining,Classification,fetalheartrate,uterine contractionsandK-Nearestneighbor

beconsideredadditionalknowledgeaboutthedata.Thekn owledgeobtainedthroughtheprocessmaybecomeadditio naldatathatcanbeusedforfurthermanipulationanddiscov ery.Applicationofdataminingconceptstothemedicalaren ahasundeniablymaderemarkablestridesinthesphereofm edicalresearchandclinicalpracticesavingtime,moneyan dlife.Clinicaldataminingistheapplicationofdataminingt echniquesusingclinicaldata.ClinicalDataMining(CDM)involvestheconceptualization,extraction ,analysis,andinterpretationof availableclinicaldataforpracticalknowledgebuilding,clinicaldecisionmakingandpractitionerreflection.Themainobjectiveofcl inicaldataminingistohaulnewandpreviouslyunknowncli nicalsolutionsandpatternstoaidthecliniciansindiagnosis, prognosisandtherapy.Moreoverapplicationofsoftwares olutionstostore patient recordsinanelectronic formisexpectedtomakeminingknowledgefromclinicald ata lessstressful. A.Cardiotocography(CTG) Sincethe1960’s,obstetriciansareusingtheCardi otocography,anelectronic methodforrecording(graphy)thefetalheartbeat(cardio)a nduterinecontractions(toco)duringpregnancy,bymeans ofaCardiotocographoranelectronicfetalmonitor(EFM). Fig.1illustratesatypicalCardiotocogram(CTG).

I. INTRODUCTION Dataminingreferstoacollection oftechniquesthatprovidethenecessaryactionstoretrievea ndgatherknowledgefromanexhaustivecollectionofdataa ndfacts.Dataisavailableinenormousmagnitude,butthekn owledgethatcanbeinferredfromthe dataisstillnegligible.Dataminingconceptsarefocusedon discoveringknowledge,predictingtrendsanderadicating superfluousdata.Discoveringknowledge inmedicalsystemsandhealthcarescenariosisaherculeany etcriticaltask.Knowledgediscoverydescribestheprocess ofautomaticallysearchinglargevolumesofdataforpattern sthatcan

Figure1.AtypicalCTG[4]

Thecontinuousmonitoringbyusing CTGrequiresqualitativeandquantitativeinterpretationso fseveral parametersdescribedasfollows[3]: Uterine activity(contractions): • Frequency:Numberofcontractioninastandard interval. • Duration:Theamountoftimefromthestartofaco ntractiontotheendofthesamecontraction. • Intensity:Ameasureofhowstrongacontractioni s. • Restingtone:Ameasureofhowrelaxedtheuteru sis betweencontractions. • Interval:Theamountoftimebetweentheendofon econtractiontothebeginningofthenextcontracti on. Uterineactivitymaybe definedas: • Normallessthanorequalto5contractionsin10minutes,a veragedover a30-minutewindow. • Tachysystolemorethan5contractionsin10minutes,averaged overa30-minutewindow. Baselinefetalheartrate(FHR),whichisdeterminedb yapproximatingthemeanFHRroundedtoincrementsoffi vebeatsperminuteduringa10minutewindow,excludingaccelerationsanddeceleration sandperiodsof marked FHRvariability. •

•

Baseline FHRlessthan110beatsperminuteandsymptoms aretermedasBradycardia. BaselineFHRgreaterthan160beatsperminutea ndsymptomsaretermedastachycardia.

BaselineFHRvariability,whichisdeterminedina10minutewindow,excludingaccelerationsanddeceleration s.BaselineFHRvariabilityisdefinedasfluctuationsintheb aselineFHRthatareirregularinamplitudeandfrequency.T hefluctuationsarevisuallyquantifiedastheamplitudeofth epeak-to-troughinbpm(beatperminute). • • • •

Absent Minimal Moderate Marked

Presenceofaccelerations:Visuallyapparentabruptin creaseinFHR.Anabruptincreaseisanincreasefromanons etofaccelerationtothepeakinlessthanorequalto30second s(tobeconsideredasacceleration, the peakmustbe greaterthanorequal to15bpm). Periodic orepisodic decelerations • Periodic:Referstodecelerationsthatareass ociatedwithcontractions.

Episodic:Referstothosenotassociatedwithcont ractions Therearefourtypesofdecelerations: • Earlydeceleration:Itisrelatedtoagradualdecrea seintheFHRwithanonsetofdecelerationtoanadi r(morethan30seconds)wherethenadiroccurswi ththepeakof acontraction. • Latedeceleration:Itisrelatedtoagradualdecreas eintheFHRwithanonsetofdecelerationtoanadir (morethan30seconds). • Variabledeceleration:Itisrelatedtoanabruptdec reaseintheFHR(morethan15bpm)thatwasmea suredfromthemostrecentlybaselinewherefrom thedeceleration’sonsettonadirislessthan30sec ondsandthedecelerationlasts(morethan15seco nds). • Prolongeddeceleration:Itispresentwhenthereis avisuallyapparentdecreaseinFHRfromthebase linethatisgreaterthanorequalto15bpm,lastingg reaterthanorequalto2minutes,butlessthan10mi nutes.Adecelerationthatlastsgreaterthanorequ al to10minutesisa baseline change.Changesortrendsof FHRpatterns overtime. • CategoryI(Normal):Baselinerate110160bpm,Moderatevariability,Absenceoflate,o rvariabledecelerations,andearlydecelerationsa ndaccelerationsmayormaynotbepresent. • CategoryII(Indeterminate):Tracingisnotpredi ctiveofabnormalfetalacidbasestatus,butevaluationandcontinuedsurveill anceandrevaluationsareindicated. • CategoryIII(Abnormal):Absenceofbaselineva riabilitywithrecurrentlateorvariabledecelerati onsorBradycardia;orsinusoidal fetal heartrate •

II. RELATEDWORK TheInternationalFederationofObstetricsandG ynaecology(FIGO)guidelines[5]wereintroducedasan attempttostandardizetheuseofelectronicmonitoringofF HR.ThefirstworkofautomaticCTGanalysisfollowingFI GOguidelinesconsistsindescribingandextractingtheCT Gmorphologicalfeatures[6].Bernades[7]developedSisP orto,asystemforautomaticanalysisof CTGtracings,basedonanimprovementofthemorphologi cal featureextractionintroducedin[6]. ArtificialNeuralNetworks(ANNs)wereusedas aclassifier[8]todetectFHRaccelerationanddecelerationp atternsandtoestimatethe FHRbaselineandvariability.ANNswereusedtoclassifyd ecelerationpatternsintoepisodicandperiodicdeceleratio ns[9]accordingtoFIGOguidelinesandbasedontherelatio nshipbetweentheparametersof

thedecelerationandtheassociateduterinecontraction.AN NswithRadialBasisFunctions(RBF)andMultiLayerPerc eptrons(MLP)werethebestperformingclassifiers. SupportVectorMachines(SVM)havebeenused forFHRsignalanalysis.Georgoulasetal.[10]useddiscrete wavelettransformationtoextractscaledependentfeaturesoftheFHRsignalandSVMfortheirclas sification.Georgoulasetal.[11]usedSVMwithRBFand polynomial kernelstoidentifyfetal andneonatalcompromise,namelymetabolicacidosis[12] .TheRBFkernelmachinesoutperformedthepolynomial machinesandbothofthemoutperformedtheconventional methodsofk-nearestneighbor(kNN),linearandquadraticdiscriminantclassifiers. Chudáčeketal.[13]usedSVM,naïveBayes,and adecisiontree(C4.5algorithm)withapolynomialkernelto analyzeFHRsignalsbasedonlinearfeatures(e.g.Descripti onoftheFHRbaselineusingmean)andnonlinearfeatures( e.g.Fractaldimensionofwaveform).TheyusedthreeFSm ethods:PrincipalComponentAnalysis,InformationGain, andGroupofAdaptiveModelsEvolution(GAME). Krupaetal.[14]proposedanewmethodforFHRs ignal analysisbasedonEmpiricalModeDecomposition(EMD)f orfeatureextractionandSVMwithRBFforclassificationo f FHRrecordings. Georgoulasetal.[15]proposedaFSmethodbase donbPSO(binaryParticleSwarmOptimization)forFHR signalanalysisusingSVMandk-NN. Hybridmethodshavebeenalsoconsideredforaut omatedFHRsignalanalysis.FontenlaRomeroetal.[16]proposedseveralapproachesforthereco gnitionofaccelerationanddecelerationpatternsinFHRsig nals,includingrule-basedapproach,ANNs,andaneurofuzzyapproach. III. MOTIVATIONANDJUSTIFICATIONCar diotocography(CTG),consistingoffetalheart rate(FHR)andtocographic(TOCO)measurements,isuse dtoevaluatefetalwellbeingduringthedelivery.Since1970,manyresearchersha veworkeddifferentminingmethodsto helpthedoctorsthatinterprettheCTGtracepatternfromthe fieldofsignalprocessingandcomputerprogramming[18]. WiththehelpofCTGtracepatternanalysisthedoctorswithi nterpretationsinordertoreachasatisfactorylevelofreliabil ity.So,theyactasadecisionsupportsysteminobstetrics.Fo reverydaypractice,noneofthemhasbeenadaptedworldwi de.Baselineestimationincomputeranalysisofcardiotoco graphs,whichiscurrentlynoconsensusonthebestmethod ology.Morethan30yearsaftertheintroductionofantepartu mcardiotocographyintoclinicalpractice,thepredictiveca pacityof themethodremainscontroversial.Inareviewoflotofarticl espublishedonthissubject,itwasfoundthatitsreportedsen sitivityvariesbetween2and100%,anditsspecificitybetwe en37and100%[19].So,inthis

work,wearegoingtoevaluateKNearestneighboralgorithmsforclusteringCTGdata. IV. MATERIALANDMETHODOLOGY A.DATASETDESCRIPTION TheCardiotocographydatasetusedinthisstudyi spubliclyavailableatTheDataMiningRepositoryofUniv ersityofCaliforniaIrvine(UCI).Byusing21givenattribut esdatacanbeclassifiedaccordingtoFHRpatternclassorfet alstateclasscode.Inthisstudy,fetalstateclasscodeisuseda stargetattributeinsteadofFHRpatternclasscodeandeachs ampleisclassified intooneofthreegroupsNormal,SuspiciousorPathologic. Thedatasetincludesatotalof2126samplesofwhichis1655 normal,295suspiciousand176pathologicsampleswhichi ndicatethe existing of fetaldistress. Attributeinformationisgivenas:LB— FHRbaseline (beatsperminute)AC—#of accelerationspersecondFM—#of fetal movementspersecondUC—#of uterinecontractionspersecondDL—#of light decelerationspersecondDS—#of severe decelerationspersecond DP—#of prolongueddecelerationspersecondASTV— percentage of timewithabnormalshort termvariability MSTV—meanvalue of short term variabilityALTV— percentage of time withabnormal longtermvariability MLTV—meanvalue of longterm variabilityWidth—widthof FHRhistogramMin—minimumof FHRhistogramMax—Maximum of FHRhistogramNmax—#of histogrampeaks Nzeros—#of histogram zerosMode— histogram modeMean—histogram meanMedian— histogrammedianVariance—histogram varianceTendency—histogram tendencyCLASS— FHRpatternclasscode(1to10) NSP—fetalstate classcode(N=normal; S=suspect;P=pathologic) B. CLASSIFICATION Classificationprocessmaybeappliedindifferentareasofre searchandpractice,e.g.,farms,military,medicine,Earthre motesensing.Theclassicalclassificationtechniquesusest atisticalapproach,whichtypicallyassumesthenormal multidimensionaldistributionofprobabilityintheexperi mentaldataset.Dataclassificationmaybesupervisedand unsupervised. Thesupervisedclassificationmethodrequiresthepresenc eoftrainingdatasettypicallydefinedbytheexperttheteacher.Eachclassofobjectsischaracterisedby thebasicstatisticalparameters

(meanvaluesvector,covariancematrix),whicharevaluesv ector,covariancematrix),whicharecomputedfromthetrai ningset.Theseparametersguidethediscriminationproces s.TheBayesianclassifiersaretypicalrepresentatives(Bay esclassifier,Fisher,Waldsequential). Theunsupervisedclassificationisalsoknownasclassificat ionwithouttheteacher.Thisclassificationuses,inmostcas es,themethodsofclusteranalysis.Thedevicethat performsthefunctionofclassificationiscalledclassifier.T heclassifieristhesystemcontainingseveralinputsthataret ransportedwithsignalscarryinginformationabouttheobje cts.Thesystemgeneratesinformationaboutthecompetenc eofobjects intoaparticularclassontheoutput.

toaheartratethatexceedsthenormalrangeforarestinghear trate(heartrateinaninactiveorsleepingindividual).Depen dingonthespeedandtypeofrhythm,itcanbe dangerous.

C. THEMEDICALBACKGROUNDOFCARDIOTO COGRAPHY(CTG)

Type 2(late)

Cardiotocographyisamedicaltestconducteddu ringpregnancythatrecordsfetalheartrate(FHR)anduterin e contractions.Eitherinternalorexternalmethodsthetestsm aybeconducted.Duringtheinternaltesting,theuterusplac edbyacatheterafteraspecificamountofdilationhastakenp lace.Theexternaltests,apairofsensorynodesareaffixedtot hemother'sstomach.TheCTGtracegenerallyshowstwoli nes.Thefetalheartrateisrecordedbytheupperlineinbeats perminuteandtheuterinecontractionsarerecordingbythe lowerlinefromtheTOCO. BaselineHeart Rate Thebaselineheartratehelpstoevaluatethehealth yfunctioningofthe cardiovascularsystem.Thebaselinefetalheartrateisdeter minedbyapproximatingthemeanFHRroundedtoincrem entsof5beatsperminute(bpm)duringa10minutewindow,excludingaccelerationsanddeceleration sandperiodsofmarkedFHRvariability(greaterthan 25bpm.Abnormalbaselineistermedbradycardiaandtach ycardia.Thefluctuationsarevisuallyquantitiesastheampl itudeofthepeak-totroughinbpm.Usingthisdefinition,thebaselineFHRvaria bilityiscategorizedbythequantitiesamplitudeas: Absent-undetectable Minimalgreaterthanundetectable,butlessthanorequalto 5bpm Moderate-6bpm-25bpmMarkedgreaterthan25bpm

Type 1(early) Thisoccursduringthepeakoftheuterinecontract ion.TheFHRwithonsetearlyinthecontractionandreturnt obaselineattheendofthecontractionwillbeuniform,repeti tiveandperiodicslowing.Thereasonsbehindthismaybefe talheadcompression,cordcompressionorearlyhypoxia.T hisoccursinfirstandsecondstagelaborwithdecentofthehe ad[21].Thisissynchronouswithuterine contraction.

Thisoccursafterthepeakoftheuterinecontractio n.TheFHRwithonsetmidtoendofthecontractionandnadir morethan20secondsafterthepeakofthecontractionanden dingafterthecontractionwillalsobeuniform,repetitivean dslowing.Ifthelagtimeishighseriousnessisalsohigh.This isalsosynchronouswithuterinecontraction.Mx:afetalpH measurementismandatory[21]. Type 3(variable) Thisisvariable,repetitive,andperiodicslowing ofFHRwithrapidonsetandrecovery.Variableandisolated timerelationshipswithcontractioncyclesmayoccur.Dece lerationpatternsintimingandshaperesemblesothertypesi nsomecases.Iftheyoccurconsistently,thereisachanceoff etalhypoxia.Thisisunrelatedtouterinecontractions.Mx:c heckfetalpHifthepatternpersistsafterturningthepatiento nherside(or ifotheradverse featuresare present)[21]. A. D.K-Nearest neighbor Thek-nearestneighborsareanexampleofaclassificationmethodwith theindependenceoftheparameters.Themethodcanbeclas sifiedasbythepointofimplicationassimple,butefficientin manydatasets[11].The kNNalgorithmisdefinedbythreeterms(S,k,T) whereSrepresentsaresemblancemeasurewhichlinkstoea chpairofdatainanappropriateNdimensionalspaceat(realorinteger)number,krepresentst henumberofnearestdatathataretrainedtocarryouttheclas sificationandTrepresentsthevectorofMtrainingdataappl iedbytheclassifiertoactuallycarryouttheclassification[1 2].k-NNusesdistancemetricsusuallyEuclideandistance toperformthe similaritymeasure formulatedby

Bradycardia: Itistherestingheartrateofunder60beatsperminu te,thoughitisseldomsymptomaticuntiltheratedropsbelo w50beats/min.Itmaycausecardiacarrestinsome patientsTachycardia:Ittypically refers

•

• Iftisadatasample,whichclassificationimplementeda nditsknearest neighborsturnup,thenthismakesaneighborhoodoft.Whil eusingthemethodtoclassifythedatasampleintheneighbo rhoodaboutt,thedistanceiswhetherconsideredornot.Ofc ourse,tousek-NNchoosingtherightvaluefor kiscritical;duetotherateofsuccessofclassificationdirectl yappertaintothisvalue.ThekNNmethodcanpassforbiasedbyk.Itispossibletochooseth ekvalueinsomanydifferentways.Ofcourse,thesmoothande fficientoneisapplying thealgorithmseveraltimeschanging thevalueofktogetthehighestaccuracy.Todecreasethedep endencylevelofchoice ofkforkNNmethod,Wang[13]recommendedcheckingmultiples etsofnearestneighborsinsteadofonesetofnearestneighbo rs.

• • •

•

Thefalsenegativerate(FN)is theproportionofpositivescasesthatwereincorre ctlyclassifiedasnegative Theaccuracy(AC)istheproportionofthetotalnu mberofpredictionsthatwerecorrect. TheSensitivityorRecalltheproportionofactualp ositivecaseswhicharecorrectlyidentified. TheSpecificitytheproportionofactualnegativec aseswhichare correctlyidentified. ThePositivePredictive ValueorPrecisiontheproportionof positivecasesthatwerecorrectlyidentified. TheNegativePredictiveValuetheproportionofn egativecasesthatwerecorrectlyidentified.

TABLE1:PERFORMANCEANALYSISOFKNEARESTNEIGHBORUSINGCROSSVALIDATIO N

Performance

KNearestN

V. EXPERIMENTATIONRESULT

Accuracy

97.4230

A. PERFORMANCE EVALUATION Thisisameasurementtooltocalculatetheperfor mance  TP+ TN  Accuracy=   TP+ TN+ FP+ FN

Sensitivity

96.3320

Specificity

96.6219

PPV

94.7793

NPV

97.5064

ROC

91.9874

Sensitivity=



Specificity=

TP  TP+ FN

 

TN     TN+ FP 

  TP+ FP

PositivePredictiveValue:PPV= 

NegativePredictiveValue:NPV=

ROC= Where

sensitivity+ speificity 2



TN    TN+ FN

VI. CONCLUSION ThisworkhasevaluatedtheperformanceoftheK -NearestneighbourAlgorithmwithrespectto confusionmatrixandaccuracy.AccordingtothatKNearestneighbourAlgorithmbasedclassificationapproac hprovidedsignificantlypoorperformance.Itwasfoundtha ttheK-NearestneighbourAlgorithm classifierwascapableofidentifyingNormal,Suspiciousa ndPathologiccondition,fromthenatureofCTGdatawithc omparativelypooraccuracy.If weconsideronlytheprecisionasametric,thenarrivedresul tsprovesthat,eventhoughthemachinelearningbasedmeth odscandistinguishthe NormalCTG

Therecallortruepositiverate(TP)isthe proportionofpositivecasesthat werecorrectlyidentified • Thefalsepositiverate(FP)istheproportionof negativescasesthatwereincorrectlyclassified aspositive • Thetruenegativerate(TN)isdefinedasthepro portionofnegativescasesthat wereclassifiedcorrectly B. patternsfromtheSuspiciousandPathologicpatter nswithrespecttoprecisionandpathologic,but,they wereincapableofdistinguishingSuspicious.Thati swhywearegettingcomparativelypooraveragepe rformancewhileclassifyingsuspiciousrecordswit hrespecttoprecision.Itisamajorweaknessoftheal gorithmswhichshouldbeovercomesinfuturedesi gn.Onemayaddressthewayto improvethesystemforgettingproperresultswithdi fferent C. classesofCTGpatterns.Onemayconsidermachinel earningbasedmethodtodesigntheCTGdataclassifi cationsystem.Futureworksmayaddresshybridmo delsusingstatisticalandmachinelearningtechnique sforimprovedclassificationaccuracy. D. E. Infuturework,weplantocollaboratewithobstetriccl iniciansandphysiciansinordertoassessthe computationalresults. •

omput.,25(4),pp.273-286,May1990. N. [7]J.Bernardes,C.Moura,J.P.deSa,andL.P.Leite, ―ThePortosystemforautomatedcardiotocographi csignalanalysis,‖J.Perinat.Med.,pp.61–65,1991. O. [8]P.A.Warrick,E.F.Hamilton,andM.Macieszcza k,―Neuralnetworkbaseddetectionoffetalheartrat epatterns,‖ in Proc.IEEEInt.JointConf.onNeuralNetw.IJCNN’0 5,Aug.2005,vol.5,pp.2400-2405. P. [9]M.Jezewski,P.Labaj,J.Wrobel,A.Matonia,J.Je zewski,andD.Cholewa,―Automatedclassificatio nofdecelerationpatternsinfetalheartratesignalusin gneuralnetworks,‖inProc.IVLatinAmericanCong .onBiomed.Eng.2007,Bioeng.Sol.forLatinAmeric aHealthIFMBEProc.,2008,vol.18,pp.5-8. Q. [10]G.Georgoulas,D.Stylios,andP.P.Groumpos, R. ―Classificationoffetalheartrateusingscaledepend entfeaturesandsupportvectormachines,‖inProcof1 6thIFACWorldCong.,July2005.

F. REFERENCES G. [1]Steer,P.J.(2008)HasElectronicFetalHeartRate MonitoringMadeaDifference?SeminarsinFetalan dNeonatalMedicine,13,WBSaunders,2-7. H. [2]YiPeng,GangKou―Adescriptiveframeworkfo rthefieldofdataminingandknowledgediscovery‖I nternationalJournalofInformationTechnology&D ecisionMakingVol.7,No.4(2008)pp.639–682. I. [3]D.Robertsa,B.Kumara,D.G.Tincellob,andS. J. A.Walkinshawa,―Computerisedantenatalfetalhe artraterecordingsbetween24,‖BritishJ.Obstet.Gy naecol.,pp.858862,Aug.2001.Sundar,C.,Chitradevi,M.andGeeth aramani,G.(2012)ClassificationofCardiotocogra mDataUsingNeuralNetworkBasedMachineLearn ingTechnique.InternationalJournalofComputerA pplications, 47,1925.http://dx.doi.org/10.5120/7256-0279 K. [4]G.Georgoulas,D.Chrysostomos,andP.P.Groum pos,―Predictingtheriskofmetabolicacidosisforne wbornsbasedonfetalheartratesignalclassificationu singsupportvectormachines,‖IEEETrans.Biomed. Eng.,pp.875-884,Apr.2006. L. [5]G.Rooth,A.Huch,andR.Huch,―Guidelinesfort heuseoffetalmonitoring,‖Int.J.Gynaecol.Obstet.,v ol.25,pp.159-167,1987. M. [6]R.Mantel,H.P.vanGeijn,F.J.Caron,J.M.Swartj es,E.E.vanWoerden,andH.W.Jongsma,―Compu teranalysisofantepartumfetalheartrate:2.Detectio nofaccelerationsanddecelerations,Int.J.Biomed.C

S. T. [11]G.Georgoulas,D.Chrysostomos,andP.P.Gro umpos,―Predictingtheriskofmetabolicacidosisf ornewbornsbasedonfetalheartratesignalclassific ationusingsupportvectormachines,‖IEEETrans. Biomed.Eng.,pp.875-884,Apr.2006. U. [12]H.P.vanGeijn,―DevelopmentsinCTGanaly sis,‖Bailliere’sClin.Obstet.Gynaecol.,vol.10,no. 2,pp.185–209,June1996. V. [13]V.Chudáček,J.Spilka,M.Huptych,G.Georgo ulas,P.Janků,M.Koucký,C.Stylios,and W. L.Lhotská,"Comparisonoflinearandnonlinearfeaturesforintrapartumcardiotocographyev aluation– Clinicalusabilityvs.Contributiontoclassification, ‖AnalysisBiomed.SignalsImages,BIOSIGNAL, pp.369-372,2010. X. [14]N.Krupa ,M.A.MA,E.Zahedi,S.Ahmed,andF.M.Hassan," Antepartumfetalheartratefeatureextractionandcla ssificationusingempiricalmodedecompositionand supportvectormachine,‖BioMed.Eng.OnLine,10: 6,2011. Y. [15]G.Georgoulas,C.Stylios,V.Chudáček,M.Mac as,J.Bernardes,and L.Lhotska,―Classificationoffetalheartratesignals basedonfeaturesselectedusingthebinaryparticles warmoptimizationalgorithm,‖WorldCong.Med.P hy.Biomed.lEng.2006,IMBEProc.,2007,pp.11561159. Z. [16]O.Fontenla-Romero,A.AlonsoBetanzos,andB.GuijarroBerdinas,―Adaptivepa tternrecognitionintheanalysisofcardiotocographi crecords,‖IEEETrans.NeuralNetw.,vol.12,no.5,

pp.1188–1195,Sep.2001. AA. [17]Sundar,C.,Chitradevi,M.a ndGeetharamani,G.(2013)AnOverviewofResea rchChallengesforClassificationofCardiotocogra mData.JournalofComputer Science, 9, 198206.http://dx.doi.org/10.3844/jcssp.2013.198.2 06 BB. [18]HakanSahinandAbdulha mitSubasi.ClassificationoftheCardiotocogramD ataforAnticipationofFetalRisksUsingMachineL earningTechniques.InternationalBurchUniversit y.FacultyofEngineeringandInformation Technologies.FrancuskeRevo lucijeb.b.,Ilidza,Sarajevo71000,BosniaandHerz egovina,AppliedSoftComputing33(2015)231– 238. CC. [19]ShahadNidhal,M.A.Mohd.Ali1andHind Najah,2010,―Anovelcardiotocographyfetalhear tratebaselineestimationalgorithm‖,ScientificRes earchandEssaysVol.5(24),p.4002-4010 DD. [20]DiogoAyres-deCamposa,CristinaCostaSantosb,Joa˜oBernardesa,2005,"Predictionofneo natalstatebycomputeranalysisoffetalheartratetrac ings:theantepartumarmoftheSisPorto1multicentr evalidationstudy‖,EuropeanJournalofObstetrics &Gynecology andReproductiveBiology 118,5260. EE. [21]Stirrat,MillsandDraycott,2003,"Noteson ObstetricsandGynaecologyfortheMRCOG,5thE dition",ISBN:9780443072239