1326页英语原文Empirical evaluation of a hybrid intelligent monitoring system using different measurWord文档下载推荐.docx
《1326页英语原文Empirical evaluation of a hybrid intelligent monitoring system using different measurWord文档下载推荐.docx》由会员分享,可在线阅读,更多相关《1326页英语原文Empirical evaluation of a hybrid intelligent monitoring system using different measurWord文档下载推荐.docx(15页珍藏版)》请在冰豆网上搜索。
1326页英语原文Empiricalevaluationofahybridintelligentmonitoringsystemusingdifferentmeasur@#@3.2.Establishmentofthevalidationstages(phaseII)@#@Inaccordancewiththeproposedmethodology,thevalidationobjectiveswerefirstestablishedonthebasisofthesystemrequirements.Thesewerethenorderedinsixlevels,asrepresentedbydifferenttonesofgreyinFig.4.Thefirstfourlevelsdealwiththesystem’scapacityforobtainingcorrectdataonthebasisofananalysisoftheCTGrecordsorientedtodiagnosticuse.Theselevelshavebeenorderedtakingintoaccountwhicharethebasicparametersneededtodeterminetheothers.Thisway,inthefirstplaceweneedtoknowtheFHRbaselineinordertoobtainitsvariability.Basedonthesetwoparameterswecanapproachthedetectionofaccelerativeanddecelerativepatterns.Andfinally,aclassificationiscarriedoutoverthislastkindofpatterns.@#@Level5referstothesystem’sfinalinterpretationofthetest(reactive/non-reactive,normal/abnormal)and,therefore,ofthefoetalstate.Finally,level6referstotheappropriatenessofthetherapeuticrecommendationsandthefoetaloutcomeprognosismadebythesystem.Aswementioned,wewillfocusourattentiononthevalidationoflevels1-4.@#@3.3.Selectionofthevalidationstrategy(phaseIII)@#@Theadvantagesofaprospectivevalidationagainstfoetaloutcomeareevident,butsincethesystemoperatesinacriticalenvironmentitisnotpossibletoobtainareferencetoestablishwhethertheinterpretationoftheFHRsignalmadebythesystemiscorrect.Therearefollowingtwomainreasonsforthis:
@#@@#@1.Someoftheindicationsformulatedbythesysteminordertomanipulatetheenvironmentwillnotbeexecuted.@#@2.Abadprognosismadebythesystemcouldnotresultinarealbadoutcome,duetoclinicalintervention.@#@Ourvalidationwas,thus,approachedfromaretrospectivepointofview.@#@3.4.Constructionofalibraryofexamples(phaseIV)@#@Thevalidationwascarriedouton3450minofCTGrecordscorrespondingto53patients.Thesewerehigh-riskpregnancieswithgestationalageabove31weeks.Patientsweretakenastheyentrythemonitoringunitwithnootherselectioncriterion.Thevalidationstartedwhenitwasdeterminedthatthesetofavailableexamplesincludedallthekindofpatternswewantedtotest.@#@Threeexpertsparticipatedinthevalidationprocess.ExpertAisdirectorofahospitalMaternal-FoetalHigh-RiskSectionandhasbeenpractisinginthefieldsofobstetricsandgynaecologysince1967.ExpertsBandCarebothdoctorsinthesameunit,expertBsomewhatmoreexperiencedthanexpertC(thealphabeticalorderofthenomenclaturereflecttheirlevelsofexperience).AllofthemperformNSTsonadailybasisandareregardedasexpertsbytheirpeersandsuperiors.@#@Thethreecliniciansanalysedtheavailablerecordsseparately.Theywerefirstrequiredtoindicatethegeneralstateoftheparametersincludedinthetest.Then,theywereaskedtocarryoutamoredetailedanalysisofthesignals,bymarkingallthedifferentpatternstheycouldidentify(accelerations,decelerations,typeofdecelerations,abruptchangesinthebaseline,suddentachycardiasandbradycardias,theappearanceofsilentorsaltatoryrhythms,etc.).Thesedataformedthebasisforthevalidation.@#@3.5.Initialvalidationwithunlabelledexamples;@#@crossed-validation;@#@analysisandinterpretationofresults(phasesV-VII)@#@Forthecalculationofthepairs’test,contingencytablesandgrouptest,SHIVA[31],avalidationtoolforintelligentsystems,wasemployed.@#@AsshallbeseeninSection4,thisinitialvalidationpermittedanadequatecharacterisationofthesystemcomparedtothegroupofexperts.Nevertheless,thereweredisagreementsthatremainedunresolved,andsoinaccordancewiththeproposedmethodology,anothervalidation,thistimecrossed,wascarriedout.Thereasonwhyacrossed-validationisnecessaryisthatinthefirstvalidationcliniciansonlyhavetoconfirmpatternsbutnotrejectanyone.Ontheotherhand,anautomaticsystemactssystematicallyinordertodetectallexistingpatterns,afactthatapparentlycausesittogenerateahighnumberoffalsepositivescomparedtotheexperts.Wesuspectedthatsomeofthesefalsepositiveswererealpatternsthatclinicianssimplyhadoverlooked.ThisaffectsespeciallytothedetectionofaccelerationsanddecelerationsandoccurssometimeswhentheanalysisofthesignalgetstoapointinwhichitisalreadypossibletoestablishadiagnosiswithouthavingtoexploretherestoftheCTGrecord.Sotheaimofthecrossed-validationwastogivethesepatternstoclinicianstoobtainfromthemanexplicitconfirmationorrejection.Furthermore,buttoalesserextent,thecalculationoftheindexoffalsenegativesproducedbythesystemisalsoaffected.@#@Inordertoresolvethisproblem,themost