外文资料--Complexity Analysis of sleep EEG signal.PDF
ComplexityAnalysisofsleepEEGsignalLiLingWangRuiping*dept.ofbiomedicalengineeringBeijingJiaotongUniversityBeijing,china100044rpwangbjtu.edu.cnAbstractThecomplexityoftheEEGtimeseriesduringsleepingisinvestigated.TherelationshipsbetweenthesesleepstatesandthecomplexitiesoftheEEGareassessed.LempelZivcomplexityisusedasanovelindexforquantifyingthecomplexityoftheEEGtimeseriesduringdifferentsleepstates.ExperimentalresultsshowthattheLempelZiv(LZ)complexityoftheEEGtimeseriesduringactive(REM,rapideyemovement)sleeptendstobehigherthanduringquiet(NREM,nonrapideyemovement)sleep,andthecomplexityduringwakeishigherthanduringsleep.TheLempelZivcomplexitycaneffectivelydistinguishthesleepstatesofthebrain.Keywords-sleepEEG;sleepstates;LempelZivcomplexity(LZ)I.INTRODUCTIONTheelectroencephalogram(EEG)signalsreflecttheelectricalactivityofthebrain.Sleepstudieshavegrowntoencompassabroadrangeoftechnologiesemployedtostudyanddiagnoseavarietyofsleepdisorders.Thestudyofthebrainelectricalactivity,throughtheelectroencephalographicrecords,isoneofthemostimportanttoolsforthestudyofsleep.Duringsleep,advancedcentralandaseriesofplantsystemchange.In1968,theinstituteofthehumanbrainintheUniversityofCaliforniareleasedthedefinitionofsleepandtechnicalstandards.AccordingtothedifferentformsandfeaturesofEEG,EMGandEOG,sleepisdividedintowakeperiod(W),rapideyemovement(REM),nonrapideyemovement(NREM),including(S1,S2,S3andS4period)6.FollowingthenonlinearcharacteristicofsleepEEG,researchershavewitnessedagrowinguseofvariousnonlinearapproachesinfeatureextractionofEEGsignalsintherecentyears,suchasLyapunovexponents,complexity,spectrumentropyetc.Allthesemethodshavetheirrespectivemeritsanddemerits.TheEEGdataofdifferentsleepingstagesareusedtocalculatethecorrespondingcharacteristicparameters.Inthestudy,thesectiongivesthebriefintroductionsofcomplexity,thedataweuseandhowtoanalyzethedata.Thesectiongivesthecalculatedresultsanddiscussions.Finally,thesectionpresentssomeremarksbasedonthestudy1.II.METHODA.ComplexityLempleandZivdefinedthatalimitedlongseriesofcomplexityshouldbethespeedofnewpatternalongwiththesequenceslengthincreased2.Inrecentyears,LZcomplexityhasbeenappliedextensivelyinbiomedicalsignalsanalysisasametrictoestimatethecomplexityofdiscrete-timephysiologicsignals10.LZcomplexityhasalsobeenusedtostudybrainfunction,braininformationtransmission,EEGcomplexityinpatientswithdiseases,andsleepEEGsignals.ThecomplexityofEEGsequenceperformsrandomdegreeoftheEEGsequenceandreflectsthesizeoftheinformation2.LZcomplexityanalysisisbasedonacoarse-grainingofthemeasurements.Inthecontextofbiomedicalsignalanalysis,typicallythediscrete-timebiomedicalsignalisconvertedintoabinarysequence.Incomparisionwiththethreshold,thesignaldataareconvertedintoa0-1sequencePasfollows:()()()1,2,.,(1),SsssrQsr=+(1)Where()()0,1,dxiTsiotherwise<=(2)Usuallythemedianisusedasthethresholdbecauseofitsrobustnesstooutliers.Previousstudieshaveshownthat0-1conversionisadequatetoestimatetheLZcomplexityinbiomedicalsignals.InordertocomputeLZcomplexity,thesequencePisscannedformlefttorightandthecomplexitycounterisincreasedbyoneuniteverytimeanewsubsequenceofconsecutivecharactersisencountered.Thecomplexitymeasurecanbeestimatedusingthefollowingalgorithm.1)LetSandQdenotetwosubsequencesofPandSQbetheconcatenationofSandQ,whilesequenceSQvisderivedfromSQafteritslastcharacterisdeleted(vdenotesthe978-1-4244-4713-8/10/$25.00©2010Crownoperationofdeletingthelastcharacterinthesequence).Let()2sdenotethevocabularyofalldifferentsubsequencesofSQv.Atthebeginning,()cn=1,S=()1s,Q=()2s,therefore,SQv=()1s.2)Ingeneral,()()()1,2,.,(1),SsssrQsr=+then()()()1,2,.,;SQvsssr=ifQbelongsto()vSQv,thenQisasequenceofSQv,notanewsequence.3)RenewQtobe()1sr+,()2sr+andjudgeifQbelongsto()vSQvornot.4)RepeatthepreviousstepsuntilQdoesnotbelongto()vSQv.Now()()()1,2,.,Qsrsrsri=+isnotasubsequenceofSQv=()()()1,2,.,1sssri+,soincrease()cnbyone.5)Thereafter,Sisrenewedtobe(1),(2),.,()Ssssri=+,and(1)Qsri=+.TheaboveprocedureisrepeateduntilQisthelastcharacter.AtthistimethenumberofdifferentsubsequencesinPthemeasureofcomplexityis()cn.Inordertoobtainacomplexitymeasurewhichisindependentofthesequencelength,()cnmustbenormalized.Ifthelengthofthesequenceisnandthenumberofdifferentsymbolsinthesymbolsetis,ithasbeenprovedthattheupperboundof()cnisgivenby()(1)log()nancnn=(3)Wherenisasmallquantityand()0nn.Ingeneral,()lognnistheupperboundof()cn,wherethebaseofthelogarithmis,i.e.,lim()()log()nncnbnn=(4)Fora0-1sequence,=2,therefore()2log()nbnn=(5)And()cncanbenormalizedvia()bn.()()()cnCnbn=(6)Where()Cn,thenormalizedLZcomplexity,reflectsthearisingrateofnewpatternsinthesequence17810.ComplexitiesofEEGaredifferentcorrespondingtothedifferentsleepstages.Accordingtotheexperienceandanalysis,thecomplexityofEEGsequenceshowstheorderlydegreeofthebrainneuronsprocessinginformationactivities.B.ExperimentDataInthisstudy,theEEGdataisfromMIT/BIHPolysomnographicdatabase.Thisdatabaseisacollectionofrecordingsofmultiplephysiologicsignalsduringsleep.SubjectsweremonitoredinBostonsBethIsraelHospitallaboratory.Therecorddataevery30sisfollowedbyaannotationandthisannotaitoncontainssleepstages,heartconditionsandbreathing.Inthisstudy,wechoose“slp01a,slp01b,slp02a,slp02b,slp03,slp04,slp14,slp48”toanalyze.TheEEGchannelsareC4-A1、C4-A1、O2-A1、O2-A1、C3-O1、C3-O1、C3-O1、C3-O1.Thesedatalengthare2h,3h,3h,214h,6h,6h,6h,1016handthesamplingfrequencyis250HZ,markingthecorrespondingsleepingstagesevery30s.C.DATAAnalysisandResultsThestudygot2500pointsfromdifferentsleepingstages10sabouteveryobject,analyzedthesedataandcalculatedthecomplexities.OurprogramisinMATLABandtheresultsobtainedareshowedinTABLE1andFigure.1.TABLE1.Thecomplexityofeachsleepingstage(average)SubjectWakeperiodNREMperiodREMperiodperiodperiodperiodperiodSlp01a0.5012-0.46120.36510.22580.3206Slp01b0.79460.34540.3183-0.3564Slp02a0.62760.32460.27540.21670.20320.2122Slp02b0.77930.75630.2664-0.5508Slp030.39280.36800.26640.2099-0.2799Slp040.66210.58240.58020.2731-0.6073Slp140.41090.27990.24380.2032-0.5057Slp480.78560.51470.50570.1896-0.3251average0.61930.45300.36470.24290.21450.3947Figure.1.Theanalysisofthecomplexityofeachsleepingstage.Fromthetable1,thereistheconclusion:fromWakeperiodto、periodinNREMperiod,thecomplexitiesareallbythemaximumreducinggradually,then,backtoclosetoperiodandperiodwhenREMperiod.TheFig.1alsocanprovetheconclusion.Wefoundweaknonlinearsignaturesinallsleepstagesinthisstudy.Theresultsshowthatduringsleeptherearevarioustransitionsandthedegreeofchaoticityisdependentonthestageofsleep.ThecomplexityofEEGsequenceshowstheorderlydegreeofthebrainneuronsprocessinginformationactivities.Asaresult,fromshallowtodeepsleep,theoutcomemeansthediminutionoffreedomofbrainactivity.InthecaseofsleepEEGthesleepstagesareconsideredasdistinctpsychophysiologicalstates789.CONCLUDINGREMARKSInthispaper,thisstudycalculatedcomplexityofsleepingEEGsignalsofeighthealthyadults.Theresultsshowthatthenonlinearfeaturecanreflectsleepingstageadequately.Themethodisusefulinautomaticrecognitionofsleepstages.Butithassomelimitations.Complexityisalsosimplebutlosesinformationdetailsinitspreprocessingoforiginalmeasurementdata1.Duetothecoarseningpretreatmentalgorithmofcomplexityandanalysistimesequencefromone-dimensionalangle,thealgorithmofcomplexityiseasytoloseinformation.Theeffectsoftheotherfactorssuchasageandgenderontheperformanceofthenonlinearfeatureextractionmethodarestillunderactivetudy2.Inspiteofthesedifficultiesandshortcoming,complexityisusefulfortheanalysisofsleepEEG.REFERENCES1Wei-XingHe,Xiang-GuoYan,Xiao-PingChen,andHuiLiu,“NonlinearFeatureExtractionofSleepingEEGSignals”,Proceedingsofthe2005IEEE,EngineeringinMedicineandBiology27thAnnualConference.Shanghai,China,September1-4,2005.2DongGuo-Ya,WuXi-Yao,”ThecomparisonBetweenApproximateEntropyandComplexityintheStudyofSleepEEG”,BeijingUniversityofScienceandTechnolongy.3LuWeimin,LiuFubin,“AnalysisoftheNonlinearDynamicsforSleepEEG”,ChinaMedicalEquipment,2008,5(2):16-20.4FuXiaohua,LiHongpei,“SleepandHealth”,ChinaMedicalJournals,2003,38(8).5DingBaoxi,ChenZhihua,ZhaoLi,“CorrelationAnalysisofEEGData”,ProgressinModernBiomedicine,2008,8(1).6LIUHui,HEWei-xing,CHENXiao-ping,“EEGtime-seriesanalysisusingnonlineardynamicsmethodforsleepmonitoring”,JournalofJiangsuUniversity(NaturalScienceEdition),Vol.26No.2Mar.2005.7Y.Shen,E.Olbrich,P.A.chermann,P.F.Meier,“DimensioncomplexityandspectralpropertiesofthehumansleepEEG”,ClinicalNeurophysiology114(2003)199-209.8S.Janjarasjitt,M.S.Scher,K.A.Loparo,“NonlineardynamicalanalysisoftheneonatalEEGtimeseries:Therelationshipbetweensleepstateandcomplexity”,ClinicalNeurophysiology119(2008)1812-1823.9ErnestoPereda,DulceMdeLaCruz,SoledadManas,JoseM.Garrido,SantiagoLopzez,JulianJ.Gonzalez,“TopographyofEEGcomplexityinhumanneonates:Effectofpostmenstrualageandthesleepstate”,NeuroscienceLetters394(2006)152-157.10MateoAboy,Member,IEEE,RobertoHornero,Member,IEEE,DanielAbasolo,Member,IEEE,andDanielAlvarez,“InterpretationoftheLempl-ZivComplexityMeasureintheContextofBiomedicalSignalAnalysis”,IEEETRANSACTIONSONBIOMEDICALENGINEERING,VOL.53,VOL.53,NO.11,NOVEMBER2006.