醫(yī)學(xué)外文翻譯--利用獨(dú)立分量分析對(duì)自然連續(xù)音樂(lè)刺激下功能性磁共振（fmri）分解中關(guān)鍵性問(wèn)題的研究（英文）

1、This article appeared in a journal published by Elsevier. The attached copy is furnished to the author for internal non-commercial research and education use, including for instruction at the authors institution and shar

2、ing with colleagues.Other uses, including reproduction and distribution, or selling or licensing copies, or posting to personal, institutional or third party websites are prohibited.In most cases authors are permitted to

3、 post their version of the article (e.g. in Word or Tex form) to their personal website or institutional repository. Authors requiring further information regarding Elsevier’s archiving and manuscript policies are encour

4、aged to visit:http://www.elsevier.com/authorsrightsAuthor's personal copyF.Congetal./JournalofNeuroscienceMethods223 (2014) 74–84 75fMRIdataincludetheblockdesignandtheevent-relateddesign(Panetal.,2011).Fortheblockdes

5、ign,thecontrastoffMRIdatabetweenthestimulusonsetandthestimulusoffsetisanalyzed.Fortheevent-relatedone,thedesignmatrixcanbeusedforregressionduringwhichthetemporalcourseofavoxelandthecorrespondingspatialmaparelearned.Witht

6、hedevelopmentoffMRIresearch,somestudiesevenreportedfMRIdataobtainedduringarealisticexperiencewherethestimulusisnaturalistic,continuousandlong(Allurietal.,2012;Hassonetal.,2004;HaynesandRees,2006;Kauppietal.,2010;Kayetal.

7、,2008;SpiersandMaguire,2007).Suchnaturalisticbraindatacanprovidemuchricherbrainresponsesforresearchandittendstobedifficulttodirectlyobtaintheprecisecontrastordesignmatrixaccordingtotheexperimentaldesign.Inordertoprocessa

8、ndanalyzesuchnaturalisticbraindata,theinter-subjectcorrelation(ISC)(Hassonetal.,2004)hasbeenwidelyused.ISCisbasedonthecorrelationbetweentwotemporalcoursesoftwoparticipantsgiventhesamespatiallocation,i.e.,thevoxelwiththes

9、amecoordinates.Recently,basedonacousticalfeatureextractionalgorithmsusedinmusicinformationretrieval,musicalfeaturesofthemusicstimulushavebeenextractedandcorrelatedtothetemporalcourseofeachvoxelofthefMRIdata(Allurietal.,2

10、012).DuetothelargeamountofvoxelsinfMRIdata,thenum-berofmultiplecomparisonsinsuchcorrelationanalysesislargeaswell.Therefore,somestatisticalmethodsaretypicallyusedtoavoidthefalsealarm.Onestraightforwardmethodistoreducethen

11、umberoftimesofcorrelations.Forexample,whenindependentcomponentanalysis(ICA)isappliedtodecomposefMRI(McKeownetal.,1998),thenumberofICAcomponents(usuallylessthanhundreds)ismuchsmallerthanthenumberofvoxels(hundredsofthousan

12、ds).Thedatadrivendataprocessingmethods,likeICA,havebeenusedtoprocessnaturalisticbraindata(Malinenetal.,2007;Ylipaavalniemietal.,2009)andthesimilaritybetweenthetemporalcoursesofthestimulusandthetemporalcoursesofICAcompone

13、nts(i.e.,spatialmaps)wasexamined.We findthatsomekeyissuesinapplyingICAtodecomposenaturalisticbraindatahavenotbeenwelladdressedyet.Thisstudyisdevotedtoanalyzingeverystepfortheapplicationofthisadvancedmethod.ForICA,theFas

14、tICAalgorithm(Hyvärinen,1999)wasused.Since1998(McKeownetal.,1998)ICAhasbeenextensivelyusedforthefMRIdataprocessing.Fordifferentdefinitionsofsamplesandvariablesinthelineartransformmodel,theapplicationofICAcanbedivide

15、dintotemporalICAandspatialICA(McKeownetal.,1998;Erhardtetal.,2010;Calhounetal.,2001;Huetal.,2005;Leeetal.,2011).Intheformer,anindependentcomponentisatem-poralcourse.Forthelatter,anindependentcomponentisavoxelseries,which

16、canbeassembledintoaspatialmapoffMRI.GiventhetypicaldimensionsoffMRIdatasets,thespatialICAisusuallypre-ferredbothfortheplausibilityoftheunderlyingneurophysiologicalmodelandforcomputationalrequirements.Hence,thespatialICAi

17、schosenforthefMRIdataanalysisinthisstudy.Hereinafter,whenICAismentioned,itisreferredtospatialICA.ICAcanbefurtherdividedintoindividualICAforanindi-vidualdataset(e.g.,includingoneparticipant’sdata)andgroupICAfortheconcaten

18、ateddataset(e.g.,includingmultiplepar-ticipants’data)(Calhounetal.,2009).GroupICAcanbeevencategorizedasthetemporalconcatenationapproach(e.g.,multi-pleparticipants’dataareconcatenatedinthetimedomain)andthespatialone(e.g.,

19、multipleparticipants’dataareconcatenatedinthespatialdomain)(Calhounetal.,2009).Thetemporalandspatialapproachesallowexaminingindividualtemporalcoursesandindividualspatialmaps,respectively,andtheyprovidecom-monspatialmapsa

20、ndcommontemporalcoursesovermultipleparticipants,respectively.Actually,groupICArequiresadditionalassumptionsbesidesthoseneededbyindividualICA(Congetal.,2013).ItisunknownwhetherfMRIdataduringreal-worldexpe-riencescanmeetth

21、eadditionalassumptions.Consequently,bothindividualICAandgroupICAareappliedtodecomposethefMRIdataheretoexaminewhethersimilarfindingscanbeobtainedbybothmethods.NomatterwhichmeansofICAisapplied,itisverycriti-caltodeterminet

22、henumberofextractedcomponents.Modelorderselection(MOS)hasbeenappliedforthispurpose(Lietal.,2007)andtheinformationtheorybasedMOSalgorithmsareoftenused,forexample,Akaike’sinformationcriterion(AIC)(Akaike,1974),MinimumDista

23、nceLength(MDL)(Rissanen,1978),andKullback–Leiblerinformationcriterion(KIC)(Cavanaugh,1999).ThistypeofMOSalgorithmsassumesthedataareindependentlyandidenticallydistributedandthecollectedbraindatahavetoberesampledtosatisfyt

24、hisassumptionforMOS(Lietal.,2007).Inthisstudy,weexamineanotherrecentlydevelopedalgorithmcalledSORTE(Heetal.,2010)forMOSoffMRIdata.SORTEisveryeffi-cientinthecomputinganddoesnotrequiretheresamplingprocess(Heetal.,2010).Alt

25、houghMOShasbeenextensivelyusedforfMRIdata,therearefewexplicitmethodstovalidatewhethertheesti-mationofMOSisaccurateornotfortherealfMRIdata.Recently,asimulationstudyhasshownthatMOScannotpreciselyestimatethenumberofsourcesi

26、nthelineartransformmodelwhensignal-noise-ratio(SNR)islow(e.g.,lessthan0dB),andthatwhenSNRislowSORTEandMDLtendtooverestimateandunderestimatethetruenumberofsources,respectively(Congetal.,2012).Inthisstudy,SORTE,AIC,MDLandK

27、ICareperformedontheconvention-allypreprocessedfMRIdataandfurtherpreprocessed(byadigitalfilter)fMRIdatatoexaminetheirperformanceinestimatingthenumberofsourcesinfMRIdataofindividualparticipants.ForindividualICA,clusteringt

28、heextractedICAcomponentsoffMRIdataisusuallyappliedtofindthecommoncomponentsacrossdifferentparticipants,andthesimilaritymatrixbasedhierarchicalclusteringhasbeenoftenused(Calhounetal.,2009;Espositoetal.,2005).ThenumberofIC

29、Acomponents(n)isalwaysmuchsmallerthanthenumberofvoxels(p).InfMRIdata,pcanbehundredsofthousands.Fortheveryhigh-dimensionaldata,dimensionreduc-tiontendstobeperformedbeforemachinelearning,likeclusteringandclassification.Int

30、hisstudy,arecentlydevelopeddimensionreductionmethodcalleddiffusionmap(DM)(CoifmanandLafon,2006)isappliedtoreducethedimensionofthedatatobeclustered(i.e.,thenICAcomponentshere)frompto2,andthen,thedegreeofclosenessofthenICA

31、componentscanbevisualizedbythescatterplotofthetwodimensionaldata.Furthermore,thespectralclus-tering(Nadleretal.,2006)isusedtofindthecommoncomponentsacrossmultipleparticipantsinthisstudy.ForgroupICA,thetemporalconcatenati

32、onseemstooutperformthespatialconcatenation(Calhounetal.,2009).Indeed,thiscon-clusionisbasedongroupICAforfMRIdatamostlyintheblockorevent-relateddesigns.ItisunknownwhethertheconclusionisvalidforthefMRIdataduringreal-worlde

33、xperiences.Therefore,bothapproachesaretriedtodecomposethefMRIhere.Inordertoaddresstheissuesmentionedabove,fMRIdataofelevenmusiciansinafree-listeningexperiment(Allurietal.,2012)areusedinthisstudy.2.Method2.1.Datadescripti

34、on2.1.1.fMRIElevenhealthyparticipants(withnoneurological,hearingorpsychologicalproblems)withformalmusicaltrainingparticipatedinthestudy(meanage:23.2±3.7SD;5females).TheparticipantswerescannedwithfMRIwhilelisteningto