脑电信号的特征分析与研究

广西师范大学硕士学位论文脑电信号的特征分析与研究姓名：周建芳申请学位级别：硕士专业：电路与系统指导教师：罗晓曙20080401I20051.5(1)(2)(3)16(4)1.51.51.5IITheStudyandAnalysisofEEGFeaturesGraduatestudent:ZhouJian-FangAdvisor:Prof.LuoXiao-ShuMajor:CircuitandSystemGrade:2005AbstractAplentifulofinformationaboutphysiologyandpathologyiscontainedinElectroencephalogram(EEG)signals.SomeevidencesofclinicaldiagnosisandaneffectivemeasureofadjuvanttherapyofcertainbraindiseasesareprovidedfordoctorsbyprocessingtheEEGsignals.ThestudiesabouttheextractionandanalysisofEEGfeatureshavemadesomeimportantprogressandachievementsathomeandabroad.Byusingmodernsignalprocessingmethods,suchassampleentropy,bispectrumand1.5-dimensionalspectrum,westudythewealthyinformationhiddeninEEGafterreadinganumberofliteraturesaboutEEGathomeandabroad.OurresultscanprovidetheoreticalreferenceworthinessforclinicaldiagnoseoftheEEG.Themaincontentsinthispapercanbesummarizedasfollows:(1)ThebasicknowledgeofEEGsignalsissummarized,andthedevelopment,datecollection,classificationandotherknowledgeofEEGarereviewed.Moreover,somemodernmethodsofEEG,suchastime-frequencyanalysis,Higher-Orderspectralanalysis,non-linearanalysisandartificialneuralnetworkanalysisareintroduced.Especially,someapplicationsofwaveletanalysis,bispectrumanalysis,complexanalysisandneuralnetworkanalysistotheEEGsignalsprocessingarereviewed.(2)Forshortcomingsofapproximateentropyalgorithm,sampleentropyisintroduced,whichisamodifiedalgorithmbasedonapproximateentropy,andisusedtoanalyzeEEGsignalsofepilepticpatientsandnormalpeople.Theresultsindicatethat,onthewhole,thevaluesofsampleentropyofepilepticpatientsarelowerthanthoseofnormalpeople;thevalueofepilepticpatientbeingintheattackperioddeclinesobviouslythanthatofepilepticpatientbeinginpre-attackperiod,andthevaluereturnstopreviouslevelafterseizure.Theseresultsarebasicallyconsistentwiththesymptomsofepilepticpatients,whichcanprovidereferencevaluefortheclinicaldiagnosisofepilepsy.(3)Thekurtosisandskewnessarecomputedtostudythecharacteristicsofnon-linearandnon-GaussianofEEGsignalsunderthedifferentstates.ByusingdirectmethodofbispectrumestimatetoanalyzetheEEGsignalsunderthreedifferentstates,therearesomedifferencesofbispectrumstructuresoftheEEGsignalsunderthethreestates.Thisverifiesthatthebispectrumanalysisisaneffectivewayofnonlinearanalysestoextractthewealthyhigh-orderinformation.AnditcontributestotheautomaticclassificationoftheEEGsignalsandprovidesthemoreusefulinformationforclinicalEEGstudies.III(4)Fordeficiencyoftraditionalbispectrumanalysis,anewmethod—1.5-dimensionalspectrumanalysisisadoptedtoanalyzeEEGandnumericallyverifythealgorithm.Ourresultsshowthat,theanalysisofthe1.5-dimensionalspectrumcansoeffectivelyinhibittheadditiveGaussiannoiseinthesignalsthatcaneasilyextractusefulnon-Gaussiansignal.Moreover,thismethodcanrevealsthequadraticphasecouplingcharacteristicoftheEEGandcangreatlyreducecomputationalcapacityandcomplexity,andalsocaneffectivelyextractusefulinformation,whichcannotbeacquiredbyusingconventionalspectralanalysis.Keywords:Electroencephalogram(EEG)signals;featureanalysis;sampleentropy;bispectrum;1.5-dimensionalspectrum11.12004201.2“”30201.31932Dietch2(1)[1-3]Lyapunov[4-6][7-8](2)[9-11](3)[12-13](4)[14](5)[15-16]1.410~2016(FP1FP2F3F4C3C4P3P4O1O2F7F8T3T4T5T6)(A1A2)3100HZ1-11-11.5(1)(2)matlab(3)()164(4)1.5Matlab1.5,1.51.61.552.1191875RichardCatonBeck1902HansBerger1929ONtheElectroencephalogramofMan”[17]1934AdrianMatthewsα1935GibbsDavisLennox3HzGibbs1936LommisHarveryHobart1939[18]2.25~1000µV[17]610~202-1[17]2-1FpZFzCzPzOTA100()()()10%20%520%10%2116167()()[19](50Hz)5000Ω2.3()()[20]25µV25µV~75µV75µV~150µV150µV()()()()(ms)8()()α0.5~30HZ1δδ4Hz20µVδδ2θθ4~7HZ10~40µVθθ20sθθθ3αα8~13HZα30~50µVααβαα4ββ14~30HZ5~30µV6%ββββββ80ms80200ms2.41932Dietch9[21]2.4.1EEG[22][23]Wingner-VillChoi-WilliamsCone1910Harr1981MorletMorlet1984GrossmanMorlet1987MallatDaubechiesEEG[2]SenhadjiEP[24]Thakor[25]AnnaCaterinaMerzagora[26]NanahoTanei[27]2.4.22060101989Vail1990IEEETrans.AutomaticControlIEEETrans.onASSP[20][28-30][31][32]AR6s[12][33][8]2.4.311206020[34],[35]1985BabloyantzEEGEEG[36]Lyapunov(1)()[37-40](2)()[41-43]1965KolmogorovKC1976LempelandZivKolmogorovLempel-Ziv1987KasperandSchuster1991Pincus(ApEn)2000RichmanC1C2C0[44]KCC0ApEnKCC1C2C0ApEn2.4.42040208012[45]1998L.Diambra[3]2002[46]2004NurettinAcır[47]2005[48]GMDH,84.5%(FNN),13Lorenz[44][15,49,50]3.13.1.11991PincusSM[51](1)u=(u(1),u(2),...,u(N))N(2)m()=[u(i),u(i+1),...,u(i+m-1)]Xi,()=[u(j),u(j+1),...,u(j+m-1)]Xj(3-1)1,1ijNm≤≤−+(3)()Xi()Xjd[X(i),X(j)]d[X(i),X(j)]=max[u(i+k-1)-u(j+k-1)],1km≤≤(3-2)(4)rrid[X(i),X(j)]r()miCr1(){d[X(i),X(j)]r}1miCrNm=≤−+jN-m+1≤(3-3)(5)()miCri()mrΦ111()ln()1NmmmiirCrNm−+=Φ=−+∑(3-4)14(6)1m+1(1)~(5)1()miCr+1()mr+Φ(7)1ApEn(m,r)=lim()()mmNrr+→∞Φ−Φ(3-5)(8)N1ApEn(m,r,N)=()()mmrr+Φ−Φ(3-6)m,rmrm=2r=0.1~0.25SD(SD)mm+1mm+1[52]3.1.2(1)100~5000(2)(3)1(,)()()mmApEnmrrrφφ+−=−{()()ujkuikr+−+≤(01)km≤≤−()()ujmuimr+−+≤}111111111(,)()()11ln()ln()11[ln()ln()]()1ln()()mmNmNmmmiiiiNmmmiiimNmimiiApEnmrrrCrCrNmNmCrCrNmCrNmCrφφ+−−++==−+=+−=−=−≅−−−+≅−−=−∑∑∑∑(3-7)15N3.1.3[53](1)NNNDDi