基于小波包分解的苍白球的神经尖峰特征提取(IJEM-V1-N5-6)

I.J.EngineeringandManufacturing2011,5,46-51PublishedOnlineOctober2011inMECS()DOI:10.5815/ijem.2011.05.06Availableonlineat*aTheKeyLaboratoryofBiomedicalInformationEngineeringofMinistryofEducation,andInstituteofBiomedicalEngineering,SchoolofLifeScienceandTechnology,Xi’anJiaotongUniversity,Xi’an710049,ChinabDepartmentofNeurosurgery,TangduHospital,Xi'an710049,ChinaAbstractInthisstudy,featuresofneuralsignalrecordingswithmicroelectrodeinjectedintotheinternalglobuspallidusofnon-humanprimatesbeforeandafter1-methyl-4-phenyl-1,2,3,6-tetrahydropyridine(MPTP)treatmentwhichinducesParkinson’sdisease(PD)wereinvestigated.Neuraloscillationswereanalyzedusingthemeasureofwaveletspectrumandsingularvalueextraction.Differencesinthetime-frequencyanalysiswereobservedbetweengroups.ResultsshowedthatthePDstateshadsignificantlyhigheramplitudesofspectrum.Singularvaluesextractedfromwaveletcoefficientsafterwaveletpackagedecompositionwereusedtodifferentiatepathologicalchangesfromnormalstates,andthesingularvaluesofthereconstructedwaveletcoefficientsofthePDstateswerestatisticallysmallerthancontrols.WespeculatethattheirregularhighfrequencyoscillationsandreducedsingularvaluesobservedinthepathologicalfiringsareconsistentwiththepreviousviewthatalossofinformationtransmissionintheneuralcircuitryunderliesPDandthesemeasurementscouldbeofdiagnosticandmonitoringvaluesofthediseasestate.IndexTerms:Parkinson’sDisease;NeuronalOscillation;WaveletPacketDecomposition;SingularValue©2011PublishedbyMECSPublisher.Selectionand/orpeerreviewunderresponsibilityoftheResearchAssociationofModernEducationandComputerScience.1.IntroductionThepathophysiologyofthePDdiseaseisrelatedwiththelossofpigmenteddopaminergicneuronswhichleadstoalterationsintheactivityoftheneuralcircuitswithinthebasalgangliathatregulatemovement[1-2].Theroleofdopaminewithinthebasalgangliaappearstobecomplex[3]whilePDisanage-relatedneurologicaldisorderviewedasadysfunctionofinformationprocessingordisruptionofthedynamicsofneuralinformationprocessingintheneuralcircuitsinthebasalganglia.TheinformationprocessingintheParkinsonismbasalgangliainvolvesnotonlyactivitiesofindividualrhythms,butalsointeractionsbetweenrhythms.*Correspondingauthor:E-mailaddress:[e-mail]ractionofGlobusPallidusBasedonWaveletPackageDecomposition47Manypreviousstudieshavereportedchangesintheneuralactivityofglobuspallidus(GP)neuronsinMPTP-treatedmonkeyscomparedwithhealthyoneswhilefiringratesoftheinternalglobuspallidus(GPi)cellsareincreased,andtheexternalglobuspallidus(GPe)cellsdisplaydecreasedfiringratesinPDsubjects.AnotherexampleisthelowfrequencymodulationoffiringactivitythatarisesinPD[4].Thecross-correlationanalysisofsimultaneouslyrecordedpallidalcellsinMPTP-treatedmonkeysrevealedthatthespikingactivitiesofmanyoftheseneuronsaresynchronized[5].Meandischargerates,instantaneousdischargerates,interspikeinterval(ISI)histograms,andautocorrelationhaveallbeenusedtoassessspontaneousneuronalactivities.However,duetothecomplexityandlargeamountsofdata,thepreviousstudiesareunabletouncovermoredynamicalchangeshiddeninthespiketrain,moreeffectiveneuronalspiketraindecodingandfeatureextractionmethodsareneededfordelineationandcharacterizationofthesechangesindischargeneuralsignal.Inthispaper,wewillmainlyfocusontheresearchoffeatureextractionofneuronaloscillatoryactivityofGpibeforeandafterincidenceofthediseasedeterminedbytheMPTPinterventionintheanimalmodelandtrytocapturethedominantdynamicchangesoftheneuronalfiringpatternsunderPDstate.2.MaterialandMethodsWaveletdecompositioncanberealizedbytheMallatAlgorithm[6-7].Waveletpacketdecomposition(WPD)[8-9],sometimesknownasjustwaveletpacketsisawavelettransformwherethesignalispassedthroughmorefiltersthanthedigitalwavelettransform,andit’sanadvantageousandfavorablemethodindigitalsignalprocessingespeciallyonthenon-stationarysignalsuchasneuronalspiketrainswhichcomposedofdifferentfrequenciessimultaneously.Thecapabilityofthismathematicalmicroscopetoanalyzedifferentscalesofneuralrhythmsisprovedtobeapowerfultoolforinvestigatingmulti-scaleoscillationsofthebrainsignals.Waveletspectrumisadoptedforthespectralrepresentationoftheneuralsignalanalysisatfirstaftersignalde-noisingbywavelettransform.Thecoefficientsderivedfromthetemporalspikeprofilescouldbetakenasfeatureextractionparameters[10-11].FornlevelsofdecompositiontheWPDproduces2n(n=1,2,3…)differentsetsofcoefficientsasopposedto3n+1setforthediscretewavelettransform.InthetheoryofSingularvaluedecomposition,anym×nrealvaluedmatrixAofrankr,wherer≦min(m,n),canbedecomposedasA=UΣVT,whereUisanm×nmatrixwithUTU=I,Visanorthogonaln×nmatrix,andisann×ndiagonalmatrix.Thediagonalelementsα1,α2…αnofΣcanbearrangedinanon-increasingorderandarecalledthesingularvaluesofA(SV).Theenergyofspecialsub-bandsandcorrespondingcoefficientsofwaveletpacketdecompositionespeciallythelargestsingularvaluesareselectedasfeatur