用于診斷輔助的醫(yī)學(xué)數(shù)據(jù)挖掘的特點_第1頁
用于診斷輔助的醫(yī)學(xué)數(shù)據(jù)挖掘的特點_第2頁
用于診斷輔助的醫(yī)學(xué)數(shù)據(jù)挖掘的特點_第3頁
用于診斷輔助的醫(yī)學(xué)數(shù)據(jù)挖掘的特點_第4頁
用于診斷輔助的醫(yī)學(xué)數(shù)據(jù)挖掘的特點_第5頁
已閱讀5頁,還剩76頁未讀, 繼續(xù)免費閱讀

下載本文檔

版權(quán)說明:本文檔由用戶提供并上傳,收益歸屬內(nèi)容提供方,若內(nèi)容存在侵權(quán),請進行舉報或認領(lǐng)

文檔簡介

Speci?csofMedicalDataMiningforDiagnosisAid:ASurvey

SarahItania,b,*,FabianLecronc,PhilippeFortempsc

aFundforScienti?cResearch-FNRS(F.R.S.-FNRS),Brussels,Belgium

bFacultyofEngineering,UniversityofMons,DepartmentofMathematicsandOperationsResearch,Mons,Belgium

cFacultyofEngineering,UniversityofMons,DepartmentofEngineeringInnovationManagement,Mons,Belgium

Abstract

Dataminingcontinuestoplayanimportantroleinmedicine;speci?cally,forthedevelopmentofdiagnosisaidmodelsusedinexpertandintelligentsystems.Althoughwecan?ndabundantresearchonthistopic,cliniciansremainreluctanttousedecisionsupporttools.Socialpressureexplainspartlythislukewarmposition,butconcernsaboutreliabilityandcredibilityarealsoputforward.Toaddressthisreticence,weemphasizetheimportanceofthecollaborationbetweenbothdataminersandclinicians.Thissurveylaysthefoundationforsuchaninteraction,byfocusingonthespeci?csofdiagnosisaid,andtherelateddatamodelinggoals.Onthisregard,weproposeanoverviewontherequirementsexpectedbytheclinicians,whoareboththeexpertsandthe?nalusers.Indeed,webelievethattheinteractionwithcliniciansshouldtakeplacefromthevery?rststepsoftheprocessandthroughoutthedevelopmentofthepredictivemodels,thusnotonlyatthe?nalvalidationstage.Actually,againstacurrentresearchapproachquiteblindlydrivenbydata,weadvocatetheneedforanewexpert-awareapproach.Thissurveypaperprovidesguidelinestocontributetothedesignofdailyhelpfuldiagnosisaidsystems.

Keywords:DataMining;Medicine;DiagnosisAid;ExplainableArti?cialIntelligence

1.Introduction

Asoneofthetrendiestresearchtopicsofourcentury,DataMining(DM)makeskeycontribu-tionstothescienti?candtechnologicaladvanceinaconsiderablenumberof?elds(

Gupta

,

2014

;

PhridviRajandGuruRao

,

2014

).Coinedduringthenineties,thedisciplineissubjecttoatoughcompetitionforthedevelopmentofalgorithmsalwaysmorepowerful,whichaimatprocessingdata

*Correspondingauthor.UniversityofMons,DepartmentofMathematicsandOperationsResearch,RuedeHoudain,9,7000Mons,Belgium.

Emailaddresses:sarah.itani@umons.ac.be(SarahItani),fabian.lecron@umons.ac.be(FabianLecron),philippe.fortemps@umons.ac.be(PhilippeFortemps)

2

Numberofpublications

1200

1000

800

600

400

200

0

199019952000200520102015

Year

Figure1:EvolutionoftheannualnumberofpublicationsrelatedtomedicaldataminingintheScopusdatabase(Sco

-

pus

)onaquarterofacentury,from1990to2015

toinfersomeknowledgeintheformofpatternsand/orrelationships(

BellazziandZupan

,

2008

).Theassociatedtechniquesarederivedfromthe?eldsofbothstatisticsandMachineLearning(ML),thelatterwhichaimsatdevelopingcomputationalmethodsabletoextractgeneralizationsfromasetofdata(

Giudici

,

2005

).

MedicalapplicationsfeatureamongtheconcernsoftheDMcommunity,withasigni?cantin-creaseinresearchinterestoverthelastyears(seeFigure

1

).Thisinteractioncomesindi?erentdisciplines(

Bellazzietal.

,

2011

):atthecellularandmolecularlevel(bioinformatics);atthetis-sueandorganlevel(imaginginformatics);atthesinglepatientlevel(clinicalinformatics);atthepopulationandsocietylevel(publichealthinformatics).

Forhalfacenturynow,diagnosispredictionhasbeenaveryactiveresearchareaofclinicalinformatics(

Wagholikaretal

.,

2012

).Inthisregard,withtheadventofDM,researchhasprogres-sivelyshiftedawayfromthestatisticalapproachlongconsideredasastandardpractice.Actually,underahypothetico-deductiveprocess,statisticalanalysesaredriventocheckahypothesisstatedbeforehandanddatasamplesarecollectedforthisspecialpurpose(

Yooetal.

,

2012

).Thisstatis-ticalapproachissurelyadaptedtoraisedi?erencesbetweenpathologicalandcontrolgroups,butnottosetanindividualassessment,i.e.aclinicalexaminationpersubject.Incontrast,enrichedbyMLtechniques,DMinductivelyprocessesavoluminousamountofdata,tobothextractknowledgeanddeveloppredictivemodelsabletohelpindiagnosingpathologies(

Vieiraetal.

,

2017

;

Yooetal.

,

2012

;

BellazziandZupan

,

2008

).Insuchaprocess,statisticsmay?nditsplaceinfeatureengineer-

3

ing,beforethestageofmodelbuildingwhichismainlybasedonMLmethodsofclassi?cationorregression(

Esfandiarietal.

,

2014

).

Inthatrespect,itisthroughdataminingthatrecentworksweredevotedtotheearlydetectionofcancer,e.g.see

LyuandHaque

(

2018

);

Aliˇckovi′candSubasi

(

2017

);

Cichoszetal.

(

2016

);

Nahar

etal.

(

2016

);

Esfandiarietal.

(

2014

);

Krishnaiahetal.

(

2013

);

Parvinetal.

(

2013

);

Guptaetal

.(

2011

).Otherpathologies,suchascardiacandpulmonarydiseases,diabetes,hypertension,meningi-tisformbesidesasigni?cantpartoftheresearchformoreprecisediagnoses(

Esfandiarietal.

,

2014

).Severalpsychiatricdisorders,suchasAttentionDe?citHyperactivityDisorder(ADHD)(

Itanietal.

,

2018a

;

Abrahametal.

,

2017

;

Milhametal.

,

2012

),Alzheimer(

Papakostasetal

.,

2015

),autism(

Kos-

mickietal

.,

2015

),schizophrenia,depressionandParkinson(

Wooetal.

,

2017

)arealsotheobjectofextensiveinvestigation.

Asprobablyperceivedbymostofresearchers,andcertainlybytheauthorsofthepresentpaper,diagnosticdecisionsupportsystemsthathavebeenproposedsofararenotunanimouslyapprovedbyclinicians(

Wagholikaretal

.,

2012

).Suchsystems,andtheunderlyingpredictivemodels,arenotablyfoundasbeingfarfromthe?eldreality.Itisthusmostlikelythatdataminersarenotenoughattentivetothespeci?csofmedicaldiagnosticdecisionsupport.Inparticular,thoughtheDMcommunitywassensitizedaboutthedistinctivenatureofmedicalapplications(

CiosandMoore

,

2002

),thepredictiveperformanceremainspracticallythelonelyparameterwithinthescopeofdataminers,whichencouragescompetition.Thistrendhasbeenaccentuatedwiththegreateravailabilityofopenmedicaldatabases,sharedbydi?erentmedicalandresearchcentersworldwide(

DiMartino

etal.

,

2017

;

Wooetal.

,

2017

;

DiMartinoetal.

,

2014

;

Esfandiarietal.

,

2014

;

Mennesetal.

,

2013

;

Ihleetal.

,

2012

;

Kerretal.

,

2012

;

Milhametal.

,

2012

;

Polineetal.

,

2012

).Someofthesedatasetswerelaunchedattheoccasionofo?cialcontests,e.g.theADHD-200collection(

Milhametal.

,

2012

).Infocusingalmostexclusivelyonperformance,theseresearchworks(1)misschallengesofbetterperceivingandunderstandingtheissuespropertothemedical?eld,(2)areexposedtotheriskofyieldinginconsistentmodels,sincenotably,recentstudiesshowedthattheremaybenologicbehindthepredictionsofaccuratemodels(

Ribeiroetal.

,

2016

).

Itisourstrongconvictionthattheclinicianshavetobeinvolvedinthewholedevelopmentprocessoftheirdecisionsupportsystems.Indeed,theybringexpertiseandknowledgetocontributetointelligentandexpertsystems.Thatiswhy,inthepresentpaper,wewillshedlightuponthespeci?csofmedicaldataminingfordiagnosisaidandraisetherelateddatamodelinggoals.Forsuchapurpose,wewilladdressthefollowingquestions.

4

(1)Howcandecisionsupportmodelsbemoreattractivetoclinicians?Whataretheexpressedrequirementsinthisregard?

(2)Whataretheobjectivescorrespondingtosuchrequirementsintermsofmathematicalmod-eling?

(3)Inwhatwaymedicaldata,particularlyinthiseraofopenmedicaldataproliferation,makesdataminingmorechallenging?

(4)Towhatextentarethecurrentdataminingtechniquesabletosatisfytheclinicians’needsandtohandletheparticularnatureofmedicaldatasimultaneously?

Inansweringthesequestions,weareledtodescribeacomprehensiveexpert-awareapproachwhichstandsoutfromtheexistingliterature,throughthreemaincontributionsexposedbelow.

·Becauseofthelimitede?ectivenessofsomemodels,

Karpatneetal

.(

2017

)pushforatheory-

guideddatascience.SuchDMmodelsaregroundedintheoreticalbases,inthedomainsofPhysicsandChemistrymainly.Inthecontextofmedicaldiagnosis,wecanadoptasimilarapproach,notguidedbytheory,butratherbytheexperts’domainknowledge.Ourpaperlaysthebasesforsuchanapproach,inbuildingakindofbridgebetweenboththemedicalanddataminingdomains.

·Wenotonlyexpressthattheissueofdiagnosisaidisofaparticularnature,wealsopropose

thetranslationoftheassociatedspeci?csintomodelinggoals.Indeed,mostofthepapersthathaveinterestonthespeci?csofthemedicaldomainhaveawidescope,andarethusnotspeci?callyfocusedondiagnosis,butalsoonprognosisandmonitoringnotably,whichinvolvesthatmodelingisnotdiscussedwithenoughdepth(

BellazziandZupan

,

2008

;

Cios

andMoore

,

2002

;

Lavraˇc

,

1999

).Besides,webringamorerecentpointofviewcomparedtothepapersthatspeci?callyaddressedaidedmedicaldiagnosis(

Wagholikaretal

.,

2012

;

Kononenko

,

2001

).

·WedonotprovideanoverviewofDMtechniquesandtherelatedworks;thiswaswidelyproposedinprevioussurveys(

Kalantarietal.

2018

;

Kourouetal.

2015

;

Esfandiarietal.

2014

;

Wagholikaretal

.

2012

;

Yooetal.

2012

).WeratherquestiontheexistingDMtechniques,giventhemodelinggoalsraisedfollowingtheunderstandingoftheproblemanddata.Thisallowsustoraisesomesolidfutureresearchdirections.

5

PREDIcTEDAs>

N

P

Negative(N)

TN

FP

Positive(P)

FN

TP

Figure2:Confusionmatrix

Thepaperisorganizedasfollows.Insection

2

,weexposethematerialsweconsideredtostructureandmakeoursurvey.Theresultsarepresentedinsection

3

anddiscussedinsection

4

.Finally,weconcludethisreportinsection

5

.

2.Materials

2.1.Terminology

Medicaldiagnosisistheresultofachallengingtaskwhichconsistsofcollectingandconciliatingdi?erentinformation(

Donner-Banzho?etal.

,

2017

;

HommersomandLucas

,

2016

;

Miller

,

2016

).Thelatterincludethesymptoms(subjectivedata)andthesigns(objectivedata)ofthetroubleprovidedbyclinicalexaminationsandlaboratorytests.Inquestofexplanationsforthesesymptomsandsigns,theclinicianscometotheconclusionoftheexistence/absenceofatrouble,i.e.thediagnosis.

Atestisoneamongotherelementsthatmotivatesadiagnosis(

Gordis

,

2014

;

CiosandMoore

,

2002

).Thepredictionsofaclinicaltestareofseveraltypes.Apatientwith(respectivelywithout)thediseaseDpredictedassuchisdesignatedastruepositive(resp.truenegative).Incaseofwrongpredictions,thepatientsarefalsepositivesorfalsenegativesrespectively.LetTP(resp.TN)denotethenumberofTruePositives(resp.TrueNegatives)andFP(resp.FN)thenumberofFalsePositives(resp.FalseNegatives);thesequantitiesareusuallyexposedinamatrixofconfusion(seeFigure

2

)(

Wittenetal.

,

2005

).Di?erentscalarmetricsarecomputedfromTP,TN,FPandFNtoassesstheperformanceofclinicaltests;theyareexposedinTable

1

(

LalkhenandMcCluskey

,

2008

;

Akobeng

,

2007a

,

b

).Letusnotethatpositiveandnegativepredictivevaluesdependontheprevalenceofthedisease(

Akobeng

,

2007a

):theyareeasilydeducedfromtheknowledgeofsensitivityandspeci?city,whicharefreefromsuchanin?uence.

Whenseveraltestsarerequiredtocheckthepresenceofamedicalcondition,thesetestsmaybeassessedgloballyintermsofnetsensitivityandnetspeci?city.Thevaluesoftheseindicatorsdependonthewayinwhichthetestswereadministered,i.e.sequentiallyorsimultaneously(

Gordis

,

2014

).Figures

3

and

4

presentthemechanismsofsequentialandparalleltesting.Forillustration

6

Test2(tp2,tn2)

Positive

Test2

(tp2,tn2)

Test2

(tp2,tn2)

METRIc

DEFINITIoN

FoRMuLA

Accuracy(A)

Rateofsuccessfulpredictions

A=TP+TN

TP+FP+TN+FN

Sensitivityor

truepositiverate(tp)

>Abilitytodetectpatientswithagivendisease.

>Probabilitythatapatientwithdis-easetestspositive.

tp=

Speci?cityor

truenegativerate(tn)

>Abilitytodetectpatientswithoutagivendisease.

>Probabilitythatapatientwithoutdiseasetestsnegative.

tn=

PositivePredictiveValue(PPV)

Chancethatapatient,predictedashavingagivendisease,istrulyso.

PPV=

NegativePredictiveValue(NPV)

Chancethatapatient,predictedasfreefromagivendisease,istrulyso.

NPV=

Table1:Performancemetricsofscreeningtests

Negative

Test1

(tp1,tn1)

Figure3:Sequentialtesting

Positive

Negative

Test1

(tp1,tn1)

NegativeNegative

Negative

Test1

(tp1,tn1)

PositiveNegative

Positive

Figure4:Paralleltesting

7

purposes,theexamplepresentsthecaseoftwotests;theassociatedreasoningmaybegeneralizedtosituationsinvolvingmoretests.Incaseofsequentialtesting,apatientissubmittedtoanotherroundofexaminationifhe/shetestedpositive,inordertosettlede?nitelyhis/hermedicalcondition.Ifthepatienttestspositivefollowingasecondroundofexamination,thesubjectisdiagnosedwiththediseaseinquestion.Thus,ifoneofbothtestspresentsanegativeresult,thepatientisconsideredasdisease-free.Theassociatednetsensitivityandspeci?cityareexpressedas:

tp=tp1.tp2andtn=tn1+tn2-tn1.tn2.

Incontrast,incaseofparalleltesting,apatientisconsideredasnegativeoncealltestscon?rmthisconditionsimultaneously.Inthiscase,theassociatednetspeci?cityandsensitivityaregivenby:

tn=tn1.tn2andtp=tp1+tp2-tp1.tp2.

Inthesamewaythatacliniciancanaskfortheopinionofanexpertconfrere,he/shecanresorttomodelsfordiagnosisaid.Theonlydi?erencebetweenbothscenariosrestsontheexternalnatureofthediagnosticsupport,eitherhumanorcomputerized.Thedataofoneorseveraltest(s)arepotentialinputsfordiagnosisaidmodels.Itshouldbenotedthatnon-interpretedoutcomesoftesting(e.g.acholesterollevel,ascan)constitutethemodelinputs,andnotthevalueofthetest(s),i.e.positiveornegative.Actually,itistheroleofthepredictivemodeltodetermineapatient’smedicalconditioninoutput.

Inlightoftheforegoing,inthepresentsurvey,whatwerefertoasamodelisdi?erentfromatest,thelatterbeingapotentialinputoftheformer.Amodelprovidesarecommendationofdiagnosis;atestprovidesaresultthatallows,amongotherpotentialinformation,tomakeadiagnosis.

2.2.Theknowledgediscoveryprocess

Theextractionofknowledgeforthepurposeofdiagnosisaid?tsintoaKnowledgeDiscoveryProcess(KDP).Sinceitspioneerformalizationby

Fayyadetal

.(

1996

),alternativemodelswereproposed,eitheracademically-orindustrially-minded(

KurganandMusilek

,

2006

).Inparticular,theKDPwasadaptedformedicalapplicationsandillustratedfortheissueofdiagnosisaidby

Cios

etal.

(

2007

,

2000

).Theassociatedstepsaresummarizedbelow.

UnderstandingoftheproblemTheprocessisinitiatedbytheproblemstatement,thede?ni-tionoftheobjectives,andthesu?cientappropriationofadomain-speci?cvocabulary.Obvi-

8

ously,interactionswithdomainexpertsareessential.Atthislevel,thechoiceofdataminingtechniquesispartiallyforeseengiventheexpressedrequirements.

UnderstandingofthedataThisstepconsistsofcollectingandexploringdata,i.e.observingandanalyzingtheinformation.

PreparationofthedataThecreationoftargetdatasets(

Fayyadetal

.,

1996

)involvesnotablynoiseremovalaswellascheckingthecompletenessandconsistencyofdata.Then,dataareprocessedthroughengineering,selectionandpossiblereductionofpertinentfeatures.

DataminingThisprocessreceivestheprepareddatasets,andextractsknowledge,i.e.patterns,relationships(

BellazziandZupan

,

2008

).

EvaluationofthediscoveredknowledgeTheresultsarecloselyconsidered:theyareexpectedtobringnewandinterestingelements,tobeunderstoodandtomakesense.Here,domainexpertshavetoplayanimportantroleintheirabilitytointerpretandassesstheresults.

UseofthediscoveredknowledgeItcanleadtoactiontaking,decisionmakingorsystemsde-ployment(

Fayyadetal

.,

1996

).

TheKDPisnotstrictlyaone-wayprocessasitisnotexcludedtoreconsidertheworkofpreviousstages:thisallowstoreinforcetheconsistencyoftheresults(

Ciosetal.

,

2007

).Forexample,the?nalevaluationmayaskforre?ningtheresults.Ortobetterunderstandthedata,are-understandingoftheproblemmaystrengthenthedomain-speci?cknowledge.

2.3.Acceptancecriteria

Onedi?cultyrelatedtomedicalDMisthatitmaytargetdi?erentpublicswiththeresultingnecessitytoaddressdi?erentexpectations.

Actually,aDMapproachmayberequestedinthemedical?eldbyresearchersandspecialistsinordertostudyagivenpathologythroughtheidenti?cationofexplanatoryfactors.Inthatcase,theextractedknowledgeisvalidatedifitcarriesacertainlevelofcredibility,measuredbymeansofcriteriarelatedtostatisticalpowernotably.Ifendorsedbythescienti?ccommunity,suchresultsmaybetakenintoconsideration(directlyorindirectly)bycliniciansfacedwithadiagnosistask.

Assuggestedinsection

2.2

,theextractedknowledgemayalsobedeployedintheformofacomputerizeddiagnosisaid.Despitetheyarethelonelyusersofsuchtechnologies,thecliniciansarein?uencedintheirexpectations,e.g.bythepatientswhoplacealotofhopeinafairdiagnosis.

9

Di?erentmodelsweredevelopedinane?orttoexplainhowaclinicianmayacceptatechnologyandintegrateittohis/herworkingpractices(

Andargolietal

.,

2017

;

Ketikidisetal.

,

2012

;

Holden

andKarsh

,

2010

;

YarbroughandSmith

,

2007

).ThemostpopularistheTechnologyAcceptanceModel(TAM),introducedby

Davisetal.

(

1989

)andrevisedby

VenkateshandDavis

(

2000

)(TAM2).Enjoyedforitsconcisestructure,themodeldepictsthepsychologicalprocesswhich,in?uencedbymaterialandsocialfactors,leadstotheintentionofusingacomputerizedapplicationindi?erent

contexts(

YarbroughandSmith

,

2007

).

VenkateshandDavis

(

2000

)reportthattheacceptanceoftechnologyisacquiredinpracticeonceitsusefulnessandeaseofusearebothperceivedbytheuser.Moreover,theeaseofuseisoneofthefactorsin?uencingtheuser’sperceptionoftheusefulnessoftheapplication.Theperceptionofusefulnessrestsalsoonsocialfactors:thesubjectivenorm,i.e.theuser’s(professionalorprivate)surroundings’opinionregardinghis/herdecisiontoadopt(ornot)theapplication,andtheimage,i.e.thesocialstatustheapplicationprovidestotheuser(

Munetal.

,

2006

;

Chismarand

Wiley-Patton

,

2002

).

Thesubjectivenormimpactsdirectlytheintentionofuse.Thisin?uenceisexertedontheclini-cianbyhis/herpatientsbutalsobytheprofessionalenvironment.Indeed,thephysicianissensitivetotheopinionofconfreres,particularlyofreferencepeopleinthedomain,eventhoughthisopinionmaybecontrarytothephysician’sbeliefs(

Munetal.

,

2006

).Asforthein?uenceofthepatients,thestudyof

Sha?eretal.

(

2013

)showstheyoftentendtodemonizecomputerizeddiagnosticsupport.Conversely,noncomputer-assistedpracticesareperceivedasapledgeofprofessionalism;maytheclinicianresorttotheopinionofanexpertconfrereisevenperceivedasanintelligentact.Yetinbothlastcases,thecliniciansmightbasetheirdecisiononelementsprovidedintheliteratureandextractedfromaDMapproach.Thus,theinvolvementofcomputinginthediagnosticprocess,ifonlytohaveanadvice,wouldinitselfleadthephysician’simagetotakeahittowardscolleaguesand/orpatients(

Munetal.

,

2006

).

Inthepresentsurvey,wewillhighlightthespeci?csofDMtodevelopdiagnosticdecisionsupportmodelswhichmeettherequirementsoftheclinicians.Wewilldealwithhowtomakecomputerizeddiagnosisaidful?llcriteriaofoutputqualityandresultsdemonstrabilityadvocatedbyTAM.Nev-ertheless,itmustberecognizedthatadoptingasuitableapproachofmodelingdoesnotguarantee

exclusivelytheacceptanceofthemodelssincesomerelatedfactors(e.g.subjectivenorm)donotfallwithinDMconcerns.

10

cKnowledgeDiscoveryProcessc

NatureofMedicalData

OverviewofDMTechniques

PerformanceEvaluation

Speci?csofMedicalDM

√√

√√

√√

Selectedtechniquesfordatamininginmedicine

Machinelearningformedicaldiagnosis:history,stateoftheartandperspective

Theuniquenessofmedicaldatamining

Predictivedatamininginclinicalmedicine:currentissuesandguidelines

Introductiontotheminingofclinicaldata

Clinicaldatamining:a

溫馨提示

  • 1. 本站所有資源如無特殊說明,都需要本地電腦安裝OFFICE2007和PDF閱讀器。圖紙軟件為CAD,CAXA,PROE,UG,SolidWorks等.壓縮文件請下載最新的WinRAR軟件解壓。
  • 2. 本站的文檔不包含任何第三方提供的附件圖紙等,如果需要附件,請聯(lián)系上傳者。文件的所有權(quán)益歸上傳用戶所有。
  • 3. 本站RAR壓縮包中若帶圖紙,網(wǎng)頁內(nèi)容里面會有圖紙預(yù)覽,若沒有圖紙預(yù)覽就沒有圖紙。
  • 4. 未經(jīng)權(quán)益所有人同意不得將文件中的內(nèi)容挪作商業(yè)或盈利用途。
  • 5. 人人文庫網(wǎng)僅提供信息存儲空間,僅對用戶上傳內(nèi)容的表現(xiàn)方式做保護處理,對用戶上傳分享的文檔內(nèi)容本身不做任何修改或編輯,并不能對任何下載內(nèi)容負責(zé)。
  • 6. 下載文件中如有侵權(quán)或不適當(dāng)內(nèi)容,請與我們聯(lián)系,我們立即糾正。
  • 7. 本站不保證下載資源的準(zhǔn)確性、安全性和完整性, 同時也不承擔(dān)用戶因使用這些下載資源對自己和他人造成任何形式的傷害或損失。

評論

0/150

提交評論