统计学文献11 11页

AdvancesinBiomedicalInformaticsfortheManagementofCancerSYEDHAQUE,DINESHMITAL,ANDSHANKARSRINIVASANDepartmentofHealthInformatics,SchoolofHealthRelatedProfessions,UniversityofMedicineandDentistryofNewJersey,Newark,NewJersey07107,USAABSTRACT:Increasedaccesstohealthcare,andadvancesineducationandtechnologyhaveresultedinalargerproportionofthepopulationhavinglongerlifeexpectancy.Thestrongcorrelationbetweenageandcancerhasresultedinamajorhealthcareproblemforthiscentury,anduntilrecentlycancerhasdefiedanylong-lastingcure.However,progress,especiallyinthefieldofbio-medicalinformatics,promisesasuccessfulpredictionandpossiblyaperma-nentcureforcancerwithinthenexttwodecades.Biomedicalinformatics—withitsrootsincomputerscience,biomedicalengineering,biostatistics,andmathematics—helpstobringthepatientclosertothephysician,facilitatesaccesstospecialistinformationandknowledgebasesacrosstheworld,andmakesitpossibletoidentifygeneticexpressionprofilesformalignantorcan-cerouscells.Thispaperreviewsthenewresearchfindingsinbiomedicalinfor-matics,workingtowardtheultimategoalofsuccessfullypredictingcancer,solvingcomplexproblemsinpreventionandtreatmentofcancer,andperhapscompletelycuringthescourgeofcancer.KEYWORDS:biomedicalinformatics;cancermanagement;imageprocessing;computer-aideddiagnosis;artificialintelligence;decisionsupportsystems;computernetworkingINTRODUCTIONCanceristhemostfeareddiseaseinthepublicmind.Currentlyaround10millionnewcasesofcancerarediagnosed,andthisisexpectedtodoublebytheyear2020.1,2Thepossibilityofpermanentlycuringcancerinthetwenty-firstcenturyrequiresanappreciationofthecontemporarynatureofourknowledge.The1950sand1960sheraldedtheeffectivecontrolofinfectiousdiseases.Cancer,however,becauseofanagingpopulationandastrongassociationbetweencancerandage,hasemergedasthemajorhealthcareproblemofthetwenty-firstcentury.Until1960,noonehadpro-posedordemonstratedthatasystemicormetastaticformofcancercouldbecured.Inthefollowingthreetofourdecades,techniquesfortheearlydetection,prevention,andtreatmenttherapieshavesignificantlyimproved,andatleast15–20%ofpatientswithsystemic/metastaticcancerscannowbecuredwiththecurrentsystemictreat-ments.Progress,especiallyinthefieldofbiomedicalinformatics,extrapolatestoapredictionofcancercontrolinthetwenty-firstcentury,andthispaperreviewstheAddressforcorrespondence:SyedHaque,DepartmentofHealthInformatics,SchoolofHealthRelatedProfessions,UniversityofMedicineandDentistryofNewJersey,65BergenStreet,Newark,NJ07107-3001,USA.Voice:973-972-6871;fax:973-972-1054.haque@umdnj.eduAnn.N.Y.Acad.Sci.980:287–297(2002).©2002NewYorkAcademyofSciences.\n288ANNALSNEWYORKACADEMYOFSCIENCESmajorresearchadvancesthataremostpromisingforthesuccessfulmanagementofcancer,fromthedetectionstagetotherapyandprevention.ROLEOFBIOMEDICALINFORMATICSINCANCERMANAGEMENTBiomedicalinformatics,makinguseofthetremendousadvancesingenetics,engineering,andcomputerscience,ispoisedtoallowfortheaccuratepredictionofcancerandalsooffersapotentialsolutionforitscure.Ithasitsrootsinthehistoryandprinciplesofinformationtheoryandhasnowbecomeanindispensableagentforgatheringandintegrationofmedicalinformation.Ithelpsovercomehumanlimita-tionsinrememberingorprocessinginformationintheformofcomplexmedicalobjects,suchassignalsorimages,acquiringandreconstructingimages,optimizingmedicationdosagesfortherapeuticpurposes,andmaintainingandupgradinglargemedicaldatabases.BytheextensiveuseofcommunicationnetworksandtheInternet(i.e.,telemedicine),biomedicalinformaticshelpsbringthepatientclosertothephy-sicianandgreatlyfacilitatesaccesstospecialistinformationandknowledgebasesacrosstheworld;knowledgebasesthatmayberequiredtoexpediteanalysisand/ortreatment.Decisionmakingbythephysicianorthespecialistisgreatlyaidedbypointofcaredecisionsupportsystems,usingtraditionalstatisticaltechniquesoradaptiveandintelligentprograms,suchasexpertsystems,geneticalgorithms,andartificialneuralnetworks.Furthermore,sequencingandtheanalysisofthehumangenomemakesitpossibletoidentifygeneticexpressionprofiles,notonlyformalig-nantorcancerouscells,butalsofortheirnormalcounterparts.Becausecancerispri-marilyageneticdisorder,theexpandingfieldofgenomicswillcertainlyaccelerateourprogresstowardunderstandingandsuccessfulcontrolofcancer.Thefollowingsectionsreviewthenewresearchfindingsineachofthebroadareasofbiomedicalinformaticsmentionedaboveandthatcontributetotheultimategoalofsuccessfullypredictingandcuringthescourgeofcancer.IMAGEPROCESSINGANDCOMPUTERAIDEDDIAGNOSISImagingtechnologyhaschangedrapidlyduringthepastfourdecades.Thisisespeciallytrueintheareaofcanceroustumordetection.Anatomicdelineationoftumorswasconsideredacrowningachievementofthelatetwentiethcentury,butmorerecentlythemajorfocushasshiftedtowardphysiologicandevenmolecularleveltumordetection.Extensiveuseisbeingmadeofnewdigitalimageprocessingtechniquesassociatedwithimagingmodalities,suchascomputedtomography(CT),powerDopplerultrasound,magneticresonanceimaging(MRI),methoxyisobutyl-isonitrile(MIBI)isotopeimaging,andpositronemissiontomography(PET)somuchsothattumorimagingisnowanessentialpartofthepracticeofoncology.Itplaysacrucialroleincancerscreeningprograms,indiagnosis,andstagingofanestablisheddiseasecase.3Furthermore,theassessmentoftumorsizebyimaging,usuallywithcomputertomography(CT)scanning,isakeycomponentindeter-miningthetumorresponsetotherapy,bothinclinicaltrialsandindailyoncologypractice.CT,ultrasound,andMRIprovidehigh-resolutionanatomicimageswith\nHAQUEetal.:CANCERMANAGEMENT289detailedstructuralinformation.Computer-assistedimaging,therefore,comprisescomputer-basedanalysisofdigitizedimagesresultinginpromptstoaidtumorclas-sificationandcomputerizedstereotacticlocalizationdevicestoimprovelocationaccuracy.However,theseimagingmodalitiesyieldlimitedfunctionalinformationontumortissuesandoftencannotdistinguishresidualdiseasefromnon-viableornecrotictumormasses,24norcantheydetectminimalresidualdisease.Incontrast,radiopharmaceuticalimagingand,inparticular,PETcangivefunctionalinformationabouttheunderlyingtissues.Thepossibilityofrefiningthesetechniquesandalsotheemergenceofnewerimagingmodalitiesthatcandetectchangesincancers(suchasMIBI)atthephysiologic,cellular,ormolecularlevels,givesrisetothenotionthatthesemethodswillhaveimplicationsfordrugdevelopmentstrategiesandalsofutureclinicalmanagement.Theoverallgoalsofimagingwithrespecttocancermanagementare,therefore,essentiallythreefold:(1)theearliestpossibledetectionofbenignormalignantlesionsortumors,(2)correlationofimagingresultswithotherclinicalparameterstoassessdiseasebiology,and(3)accuratestagingandfollow-upaftertreatment.Toaidtheradiologist,ortoautomatetheprocessofdifferentiatinglesionsasmalignantorbenign,variouscomputer-basedmethodshavebeendeveloped.Thebasisofcomput-er-aideddetectionoftumorsistheuseofcertainvisualcriteriatodetectabnormali-tiesoncancerouscells,tocharacterizethemasmalignantorbenign.However,itisoftendifficulttotranslatethesemeasuresintocomputeralgorithmsthatexactlymatchwhatradiologistsvisuallyperceive.Thus,amoredesirableapproachtocom-puter-baseddetectionistoselectappropriatefeaturesfordescribingmalignantandbenignlesions.Classifiersaredesignedtopredictthemembershipofagivensamplebasedonthefeaturevectormembershipinafeaturespace.Ofcourse,properselec-tionoffeaturesisaveryimportantstep,sinceinclusionofinappropriatefeaturesoftencanadverselyaffectclassifierperformance.4,5Theroleofacomputeris,there-fore,toextractimagefeaturesfromsuspiciousregionscontaininglesions,andesti-matethelikelihoodthatthelesionismalignantorbenign.Forexample,intheareaofbreastimagingormammography,oneofthemostsignificantfeaturesusedbytheradiologistand/oracomputer-basedtechniqueliesinthedetectionofmicrocalcifi-cationsinbreasttissue.6–8Microcalcificationsarecalciumdepositsinbreasttissuethatareimagedassmallbrightspotswithabout0.1mmto0.5mmdiameteronamammogram.Manymicrocalcificationsarerelatedtobenignbreastdiseases,butsomearecausedbymalignancy.Malignantmicrocalcificationsoftenhaveirregularshapesandsizes,tendingtogrouptogetherformingclusters(MCCs).IndeedMCCsareindicatedtobeoneofthesuresignsinmammogramthatsignifythepresenceofbreastcancer.Becauseofthelowspecificityofmicrocalcificationfeatures,9,10radi-ologistsrecommendbiopsyformostcasesinordernottomissbreastcancer.Useofcomputer-aidedimageprocessingtechniquescanobviatetheneedfor,attheveryleastminimize,theuseofsuchinvasivetechniques.11,12Thetypesoffeaturesextractedareusuallymorphologicalortextural.Morphologicalfeaturesareusedtodescribethesize,shape,andcontrastoftumorswiththeirsurroundings,forexample,thefeaturesofmicrocalcificationsdetectedinamammogramandtheirvariationwithinacluster.Dobroetal.haveusedtwomorphometricfeatures(shortaxisandopticaldensity)forthepredictionofregionallymphnodesmetastasesinpatientswithlaryngealcancer.13Texturefeaturesmaybeextractedfromthebreasttissueto\n290ANNALSNEWYORKACADEMYOFSCIENCESdescribethetexturalchangesofthetissueduetoadevelopingmalignancy,asusedbyKegelmeyeretal.14Huoetal.15used14features,somemorphological,sometex-tural,andothersfromthecentralbreastregionondigitizedmammograms,tochar-acterizethemammographicparenchymalpatternsofwomenatdifferentrisklevels.Longatoetal.16developedanimageanalysissystemsthatusesquantitativenuclearmorphometryinformationformakingclinical,diagnostic,andprognosticoutcomepredictionsinbothprostateandbladdercancer.Theyusedmostlytexturalfeaturesquantifyingnuclearsize,shape,DNAcontentandchromatinorganizationfortheirclassificationpurposes.ThestudyofBertheetal.17investigatedifbiologicalfea-turesdeterminedthroughimagecytometryareabletocharacterizeclinicalsubpop-ulationsoflipomas.Theiranalysisgeneratedthreegroupsofquantitativebiologicalvariablesdescribingmorphonuclearaspects(i.e.,thechromatinpatternofthecellnuclei),thenuclearDNAcontent(DNAploidylevel),andarchitectural.Possiblerelationsbetweenclinicalandbiologicalfeaturesoflipomaswereinvestigatedbymeansofunivariateandmultivariatestatisticalanalyses.Manystudiesevenextendtheimageanalysistoaidinbiopsyprocedures,asisaptlyillustratedbythetwostud-iesthatfollow.Zengetal.18developedathree-dimensional(3D)modelingandsim-ulationtechniquetoevaluatemostofthebiopsyprotocolsincurrentclinicaluse,andtocorrelatetheresultswiththosefromclinicalcases.Usingdeformablemodelingtechniques,3Dcomputerizedprostatesurfacemodelswerereconstructedfromstep-sectioned,whole-mountedradicalprostatectomyspecimenswithlocalizedprostatecancer.A3Dcomputersimulationsystemwasdevelopedtoaccuratelydepicttheanatomyoftheprostateandallindividualtumorfoci.Aphysiciancouldeasilyandinteractivelyuseitforprostateneedlecorebiopsybymeansofuser-friendlygraphicuserinterface(GUI).Atotalof281prostatemodelswerereconstructed,and18biop-sieswereperformedbyaurologistoneachmodeltodeterminethedetectionratesofsevendifferentbiopsyprotocols.Clinicalbiopsiesfrom35patientcaseswerealsoreviewedandcorrelatedwiththesimulationresults.Similarly,Nakajimaetal.19pre-sentedthedesignandmethodsofavideo-basedguidancesystemtoassistwithbraintumorbiopsyprocedures.Theirsystemalsocomprisesa3Dpositionsensor,aCCDcamera,aliquidcrystaldisplay(LCD),andagraphicscomputerequippedwithavideocaptureboard.ThemarkersforpositionalmeasurementareattachedtotheCCDcamera,asurgicaltool,andthepatienthead.Inordertobeintegratedintothecamera-viewingcapacity,3Dimagesoftheheadandguidanceindicatorsarevisual-izedtoformalifelikevideoimage.Beyondthenowubiquitousroleofaidingtheacquisitionandprocessingofimag-estodetecttumor-relatedpatterns,theapplicationofcomputersystemsandnet-worksnowextendtotraditionallyhumantasks,suchasdecisionmaking,therapymanagement,andpatienteducation.Indeed,Chanetal.20haveevensuggestedthatcomputerbaseddecisionscouldbemorereliableandconsistentthandecisionsmadebydoctors.Additionalstudieshavecorroboratedtheuseofcomputer-aideddiagno-sisinaidingorimprovingthedetectioncapabilitiesofradiologists.Forexample,astudyconcentratingonthefalse-negativerateinscreeningmammography,thecapa-bilityofcomputer-aideddetectiontoidentifythesemissedlesions,andwhetherornotsuchcomputermethodsincreaseradiologistsrecallrate.21Theoriginal-attend-ingradiologistshadafalse-negativerateof21%andthecomputeraideddetectionmethodcouldhavepotentiallyhelpedreducethisfalse-negativerateby77%.In\nHAQUEetal.:CANCERMANAGEMENT291anotherstudy22theobjectivewastoassesstheeffectofcomputer-aideddetection(CAD)ontheinterpretationofscreeningmammogramsinacommunitybreastcen-ter.TheuseofCADintheinterpretationofscreeningmammogramscanincreasethedetectionofearly-stagemalignancieswithoutundueeffectontherecallrateorpos-itivepredictivevalueforbiopsy.Therewasasignificantimprovementintheaccura-cyofscreening.Inyetanotherstudy,23itwasfirmlyestablishedthatmammographiccharacteristicsofcancersmissedatscreeningweredetectedbyaCADtechnique.Detectionerrorsaffectedcaseswithcalcificationsandmasses.CADmarkedmost(77%;88of115)cancersmissedatscreeningthatradiologistsretrospectivelyjudgedtomeritrecall.Ithasbeenshown24thatthedetectionofsensitivityandspecificitycanbeconsiderablyincreasedbyusingcomputerbaseddecisiontoolsforbreastcan-cer.Suchcomputerbaseddecisionsupportsystemsforaidingorautomatingthepro-cessofdetection,diagnosis,andtreatmentformanactiveareaofresearch,asisshownbelow.USEOFDECISIONSUPPORTSYSTEMSINCANCERDETECTIONANDTREATMENTMANAGEMENTDecisionsupportsystemshelpprovidethedirectorindirectassistanceconstantlyrequiredbyaphysicianinmakinglogicallyrelateddecisionsconcerningagivenpatient.However,bothtypesofassistancecouldbedirectedtowardimprovingthequalityofdiagnosisortheefficiencyoftherapy.Indirectassistanceisintheformofhospitalinformationsystemsthatprovideaccesstopatientrecordsandlaboratoryresults,orgettinginformationfrombibliographic,legal,orknowledgedatabasespossiblystoredelsewherethanthepointofcare.Directassistance,ontheotherhand,typicallyincorporatesreasoningmechanismsthatapplymedicalknowledgetothespecificcaseofaparticularpatientandsuggestsolutionsthatofferthebestcost/benefitratio.Suchsystemsessentiallyfollowthegeneralaspectsofmedicalinter-ventionintheirpredictive,preventative,curative,andassistancecapabilities.25−27Thefollowingparagraphsdiscussinterestingworkdoneinthisarea.Manydecisionsupportsystemsaredesignedtobeautonomous,asintheworkofBreitfieldetal.,28whodevelopedapoint-of-useportabledecisionsupporttooltoautomatethecancerclinicaltrialprocess.Thesupporttoolconsistsofahand-heldcomputerwithaprogrammablerelationaldatabase.Atwo-levelhierarchicdecisionframeworkisusedfortheidentificationofeligiblesubjectsfortwoopenbreastcan-cerclinicaltrials.Thisdecisionsupporttoolandthedecisionframeworkonwhichitisbasedcouldbeusedformultipletrialsatvariouscancersites.Floydetal.29presentcasebased,reasoningcomputersoftwaredevelopedfrommammographicfindingstoprovidesupportfortheclinicaldecisiontoperformbiopsyofthebreastThecase-basedreasoningsystemisdesignedtosupportthedecisiontoperformbiopsyinthosepatientswhohavesuspiciousfindingsondiagnosticmammography.Acom-puter-aideddetectionsystemhasbeenproposed30fortissuecellnucleiinhistologicsectionsandvalidatedaspartofthebiopsyanalysissupportsystem(BASS).Cellnucleiareselectivelystainedwithmonoclonalantibodies,suchastheanti-estrogenreceptorantibodiesthatarewidelyappliedaspartofassessingpatientprognosisinbreastcancer.Thedetectionsystemusesareceptivefieldfiltertoenhancenegatively\n292ANNALSNEWYORKACADEMYOFSCIENCESandpositivelystainedcellnucleiandasquashingfunctiontolabeleachpixelvalueasbelongingtothebackgroundortoanucleus.Detectionandclassificationofindi-vidualnucleiaswellasbiopsygradingperformance,hasbeenshowntobepromis-ingcomparedtotheperformanceofexperts.Sensitivityandpositivepredictivevaluesweremeasuredtobe83%and67.4%,respectively.OnemajoradvantageofBASSstemsfromthefactthatthesystemsimulatestheassessmentproceduresrou-tinelyemployedbyhumanexperts;thus,itcanbeusedasanadditionalindependentexpert.Indeedresearchintodesigningdecisionsupportsystemswithhumanexpert-likedecisionmakingisincreasinglyusingtheframeworkofartificialintelligence(AI).ManysuchAIbasedsystemsarelistedbelowthatemployadaptiveorlearningtechniques,suchasgeneticalgorithms,productionrules,andneuro-fuzzylearningrules.ARTIFICIALINTELLIGENCEBASEDDECISIONSUPPORTSYSTEMSApartfromusingthenowwell-establishedmathematical(hemodynamicsandpharmacokinetics)andstatistical(Bayesian)models,thenewAIbaseddecisionsup-portsystemsdevelopedforthemanagementofcanceruseawidevarietyoftech-niques,suchasgeneticalgorithms(GA),fuzzylogic,andartificialneuralnetworks(ANN).GAarewellsuitedforfeatureselectionproblems,whereoptimalsolutionispracticallyimpossibletocomputeandanearoptimalsolutionisthebestalternative.GAthatcandetectMCCsandclassifythemasbenignormalignanthavebeendesignedandimplementedbyZhangetal.,Yamanyetal.,Anastosioetal.,31–33andothers.SimilartoGAtechniquesaretheartificialneuralnetworkandfuzzylogictechniquesforclassificationanddetection.Bothofthesetechniqueshavebecomewellestablishedasviable,multipurpose,robustcomputationalmethodologieswithsolidtheoreticalsupportandwithstrongpotentialtobeeffectiveintheareaofcancerpredictionandcontrol.34,35Forexample,neuralnetworkscanextractnewmedicalinformationfromrawdata,buildcomputermodelsthatareusefulformedicaldeci-sion-making,andaidinthedistributionofmedicalexpertise.InthedomainofcancermanagementresearchresultsarebecomingavailablefromthetechniquesofANNandfuzzylogicduringthepastdecade.YuandGuan36presentaneuralnetworkbaseddiagnosissystemfortheautomaticdetectionofMCCsindigitizedmammo-grams.Theproposedsystemconsistsoftwomainsteps.Usingfeaturesconsistingofamixtureofwaveletandgraylevelstatisticalmeasuressegments,potentialMCCpixelsinthemammogramsarelocated.Thesearethenlabeledintopotentialindivid-ualmicrocalcificationobjectsbytheirspatialconnectivity.Individualmicrocalcifi-cationsaredetectedbyusingasetof31featuresextractedfromthepotentialindividualmicrocalcificationobjects.Thediscriminatorypowerofthesefeaturesisanalyzedusinggeneralregressionneuralnetworksviasequentialforwardandsequentialbackwardselectionmethods.AsimilartechniqueisemployedbyGavri-elidesetal.,37whohavedevelopedamultistagecomputer-aideddiagnosis(CAD)schemefortheautomatedsegmentationofsuspiciousMCCsindigitalmammo-grams.Theschemeconsistsofthreemainprocessingsteps.First,thebreastregionissegmentedanditshigh-frequencycontentenhancedusingnon-sharpmasking.Inthesecondstep,individualmicrocalcificationsaresegmentedusinglocalhistogram\nHAQUEetal.:CANCERMANAGEMENT293analysisonoverlappingsubimages.Forthisstep,eighthistogramfeaturesareextractedforeachsubimageandareusedasinputtoafuzzyrule-basedclassifierthatidentifiessubimagescontainingmicrocalcificationsandassignstheappropriatethresholdstosegmentanymicrocalcificationswithinthem.AnotherexampleisthatduetoPenedoetal.38whohavedevelopedacomputer-aideddiagnosissystem,basedonatwo-levelANNarchitecturefordetectinglungcancernodulesfoundondigitizedchestradiographs.ThefirstANNperformsthedetectionofsuspiciousregionsinalow-resolutionimage.TheinputstothesecondANNarethecurvaturepeakscomputedforallpixelsineachsuspiciousregion.Thisisbasedonthefactthatsmalltumorspossessanidentifiablesignatureincurvature-peakfeaturespace,wherecurvatureisthelocalcurvatureoftheimagedatawhenviewedasareliefmap.Theyemployedafree-responsereceiveroperatingcharacteristicsmethodwiththemeannumberoffalsepositives(FPs)andthesensitivityasperformanceindicestoevaluateallthesimulationresults.Thecombinationofthetwonetworksprovideresultsof89%–96%sensitivityand5–7FPs/image,dependingonthesizeofthenodules.Wuetal.39alsoconstructedanartificialneuralnetwork(ANN)baseddiag-nosissystemforcancerpredictionusinganoptimallyminimizednumberofinputfeatures.AbackpropagationANNmergednineinputfeatures(ageandeightradio-graphicfindingsextractedbyradiologists)topredictbiopsyoutcomeasitsoutput.Thefeatureswereranked,andmoreimportantoneswereselectedtoproduceanopti-malsubsetoffeatures.SimilarapplicationsofANNsforbothbreastandovariancan-cerscanbefoundintheworkofWildingetal.40andofLoetal.41,42ANNshavealsobeenusedfortreatmentsetups,asintheworkofRowbottometal.43whodevel-opedanANNbasedmethodologyforthecustomizationofcoplanarbeamorienta-tionsinprostatecancerradiotherapy.Thegeometryofthepatientswasmodeledbyreducingtheexternalcontour,planningtargetvolume,andorgansatrisktoasetofcuboids.ThecoordinatesandsizeofthecuboidsweregiventotheANNasinputs.Evenusingaverysimplemodelforthegeometryofthepatient,anANNwasabletoproducebeamorientationsthatweresimilartothoseproducedbyconventionalcom-puteralgorithms.ROLEOFCOMPUTERNETWORKINGINCANCERMANAGEMENTThedramaticprogressinthecomputernetworkingindustry(especiallytheonsetoftheWorldWideWeb)isalreadyhavingastronginfluenceonallmedicalpractice(telemedicine),includingoncology.Withinformationonmanydiseasesnowavail-ableandrapidlyexpandingontheweb,mostpatientswanttoobtainmaximumknowledgerelatedtotheirdiseases,diseasesymptoms,medications,treatmentoptions,andsoforth.IndeedInternetbasededucationisgreatlyinfluencingpatientself-educationanddecisionmakingaboutscreeningandtreatmentofvarioustypesofcancer.ThestudiesofLondonandGomella,44Pautleretal.,45andKimetal.46areexcellentexamplesandsourcesofinformationontherelevanceandapplicationoftheInternetincancermanagement.Apartfromgeneralinformationdissemina-tion,personalizedcomputerrecordscouldalsobeachievedforpatientmanagementacrosstheInternet,asisshownbytheworkofHolzeretal.47andothers.48–49Rickeetal.50summarizethemostimminentinfluencesoftelemedicineonthefutureofthe\n294ANNALSNEWYORKACADEMYOFSCIENCESoncologist.Specifically,thehistoryoftelemedicine,recenttechnologies,theperfor-manceofelectronicpatientrecords,andtheeffectsoftelemedicineservicesandelectronicinfrastructuresonclinicalworkflowandonmedicalqualitymanagementarereviewedindetail.Aswellaswellknownandestablishedtelemedicineservices,suchasvideoconferencing,themostinfluentialtrendsarethespreadofdigitalhospitalinfrastructureswithunlimited,securedaccesstorelevantpatientdata.Becausethevolumeandsizeofsuchdataincreasesdramatically,anadequatenetworkinfrastructureiscrucial.Sowhatdowedonowtoprepareforthefuture?Lowe51suggeststhefollowing:(1)establishanoncologyinformaticsgroupwithinthecancercentertoprovidethenecessaryexpertiseandbegintheplanningprocess;(2)beginimplementingasecureintranetbasedonstandardInternettechnologies;(3)workwiththehostmedicalcenterandexternalagenciestodeterminewhowillpayforandimplementahigh-bandwidthnetworkinfrastructure;(4)recruitbioinfor-maticsspecialist/technicianswhocanhelpimplementtechnologies;(5)ensurethatthecancercenter’sEMRsystemcansupportcancerprotocoldataandfacilitatetheretrievalanddeliveryofthecomplexdigitalimagingdatathatbecomeavailableinfuture.CONCLUSIONSAdvancesinbioinformaticsthatwillhavethegreatestimpactonthepracticeoflaboratorymedicineduringthenextdecadeincludemoleculardiagnostics,nearpatienttesting,medicalimageanalysis,robotics,andpatientinformationmanage-ment.Thelistofmolecularpathologytestswithpotentialclinicalutilityexpandsdaily.Some,suchastestsforhumanimmunedeficiencyvirus(HIV)andhepatitisCvirus,arealreadyavailableascommercialkits.DNAtestsinoncology,suchasT-andB-cellgenerearrangementsandtranslocations,haveprovenusefulindetectingsmallnumbersoftumorcellsandhavedemonstratedclinicalutilityinsomecircum-stances.ScreeningtestsforthegeneticpredispositiontocertainformsofcolonandbreastcancerandAlzheimer’sdiseasearenowpossible.Thissuggeststhepotentialforlargescalescreeningofpopulationsatrisk.Continuedimprovementsinbiosen-sortechnologyandminiaturizationwillincreasetheabilitytotestformanyparam-etersatornearthepatient.Computerizedimageanalysiswillradicallychange,andinsomecaseseliminate,manualclinicalmicroscopyinurinalysis,hematology,immunohistochemistry,andcytology.Roboticswillgreatlydecreasepersonnelrequirementsforrepetitivetasks,suchasspecimentransport,processing,andallo-cating.Imagemanagementsystemswillallowarchivingofdiagnosticgrossandmicroscopicimages,togetherwithtraditionaltextdescriptionsanddiagnosis.Tele-pathologywilllinksmallercenterswithexpertconsultantsintertiarycenters.Voicerecognitionsystemswillobviatetheneedfortranscriptionists.Finally,advancesinbiotechnologyandbioinformaticsoffergreatpromise,withnewbreakthroughsingenediscoveryandelucidationofgenefunction.Atpresent,manycandidategenesrelatedtocancerpathogenesishavebeenidentifiedinseveraltypesofhumancancer,yetfrequentlytheirfunctionremainselusive.Thisisparticularlytrueasitrelatestotheprogressionofhumancancer.High-throughputmolecularapproachesareemerg-ingthatmaybecomeaccurate,automated,andcosteffective.Forexample,DNA\nHAQUEetal.:CANCERMANAGEMENT295arraysonmicrochipsareunderdevelopmentwithnumerousapplications,includingtheabilitytoscreengenesrapidlyformutationsandtostudypatternsofgeneexpres-siononalargescale.52Automatedsystemsformicrodissectionandsequencingarealsointheirimplementationstages.Commensuratewiththeirintegrationandevolu-tion,theseinformationandtechnologicaltoolshavethepotentialtoofferamorecomprehensiveunderstandingofmultiplegeneticandcellularalterationsoccurringduringcancerinitiation,development,andprogression.53Molecularepidemiologyhasenormouspotentialincancerpreventionthroughtheearlyidentificationofatriskpopulationsandtherapidassessmentofinterventionefficacy.54Allofthesetechnologieswillbeexpensivetoimplement,butwell-planneddeploymentwillbothdecreasecostandimprovethequalityofmedicalcare.Withbioinformatics,there-fore,acompleteunderstandingoftheprocessofcancer,ifnotitspermanentcure,iswellwithinreach.REFERENCES1.SIKORA,K.1999.Developingaglobalstrategyforcancer.Eur.J.Cancer35:1870–1877.2.GREENLEE,R.T.,etal.2000.Cancerstatistics.2000.CACancerJ.Clin.50:7–33.3.PISANO,E.D.&C.A.PARHAM.2000.Digitalmammography,sestamibibreastscintig-raphy,andpositronemissiontomographybreastimaging.Radiol.Clin.NorthAm.38:861–869.4.JIANG,Y.,etal.1999.Improvingbreastcancerdiagnosiswithcomputeraideddesign.Acad.Rad.6:22–23.5.SAHINER,B.,etal.2000.Featureselectionandclassifierperformanceincomputeraideddiagnosis:theeffectoffinitesamplesize.Med.Phys.27:1509–1522.6.FEIG,S.A.&R.E.HENDRICK.1993.Risk,BenefitandControversiesinMammographicScreening.119–135.RadiologicalSocietyofNorthAmerica,Inc.,OakBrooks,IL.7.DUFFY,S.W.&L.TABAR.2000.Screeningmammographyreevaluated.Lancet355:747–748.8.SMART,C.R.,etal.1995.Benefitofmammographyscreeninginwomenages40–49years.Cancer75:1619–1626.9.ADLER,D.&M.A.HELVIE.1992.Mammographicbiopsyrecommendations.Curr.Op.Radiol.4:123–129.10.KOPANS,D.B.1991.Thepositivepredictivevalueofmammography.AJR.158:521–526.11.BOCCIGNONE,G.,A.CHIANESE&A.PICARIELLO.2000.Computeraideddetectionofmicrocalcificationsindigitalmammograms.Comput.Biol.Med.30:267–286.12.MATA,C.R.,etal.2000.Detectionofmicrocalcificationsbymeansofmultiscalemethodsandstatisticaltechniques.J.Digit.Imaging13:221–225.13.DOBRO,S.W.,etal.2000.Nuclearmorphometryforthepredictionofregionallymphnodesmetastasesinpatientswithcancerofthelarynx.Otolaryngol.HeadNeckSurg.123:770–774.14.KEGELMEYER,W.P.,etal.1994.Computeraidedmammographicscreeningforspicu-latedlesions.Radiolology191:331–337.15.HUO,Z.,etal.2000.Computerizedanalysisofmammographicparenchymalpatternsforbreastcancerriskassessment:featureselection.Med.Phys.27:4–12.16.LONGATTO,F.A.,A.SANTINELLI&R.MONTIRONI.2000.Chromatintexturebycom-puter-assistedimageanalysis.Pathologica92:94–95.17.BERTHE,J.V.,etal.2001.Thecontributionofofimagecytometrytothecharacteriza-tionofclinicalsubgroupslipomas.Intl.J.Oncol.18:1315–1321.18.ZENG,J.,etal.2001.Prostatebiopsyprotocols:3Dvisualization-basedevaluationandclinicalcorrelation.Comput.AidedSurg.6:14–21.19.NAKAJIMA,Y.,etal.2000.Enhancedvideoimageguidanceforbiopsyusingthesafetymap.Stud.HealthTechnol.Informat.70:230–232.\n296ANNALSNEWYORKACADEMYOFSCIENCES20.CHAN,H.P.,etal.1999.Improvementofradiologists’characterizationofmammo-graphicmassesbycomputeraideddiagnosis:anROCstudy.Radiology212:817–827.21.SANDERS,G.D.,R.F.NEASE,JR.&D.K.OWENS.2000.Designandpilotevaluationofasystemtodevelopcomputer-basedsite-specificpracticeguidelinesfromdecisionmodels.Med.DecisionMaking20:145–159.22.WARREN,B.L.,etal.2000.Potentialcontributionofcomputer-aideddetectiontothesensitivityofscreeningmammography.Radiology215:554–562.23.MOBERG,K.,etal.2001.Computedassisteddetectionofintervalbreastcancers.Eur.J.Radiol.39:104–110.24.BIRDWELL,R.L.,etal.2001.Mammographiccharacteristicsof115missedcancerslaterdetectedwithscreeningmammographyandthepotentialutilityofcomputer-aideddetection.Radiology219:192–202.25.VYBORNY,C.J.,M.L.GIGER&R.M.NISHIKAWA.2000.Computer-aideddetectionanddiagnosisofbreastcancer.Radiol.Clin.NorthAm.38:725–740.26.MICHAELSON,J.S.,E.HALPERN&D.B.KOPANS.1999.Breastcancer:computersimula-tionmethodforestimatingoptimalintervalsforscreening.Radiology212:551–560.27.ZWIGGELAAR,R.,etal.1999.Model-baseddetectionofspiculatedlesionsinmammo-grams.Med.ImageAnalysis3:39–62.28.BREITFELD,P.P.,etal.1999.Pilotstudyofapoint-of-usedecisionsupporttoolforcancerclinicaltrialseligibility.J.Am.Med.Inform.Assoc.6:466–477.29.FLOYD,C.E.,JR.,J.Y.LO&G.D.TOURASSI.2000.Case-basedreasoningcomputeralgorithmthatusesmammographicfindingsforbreastbiopsydecisions.Am.J.Roentgenol.175:1347–1352.30.SCHNORRENBERG,F.,etal.1997.Computer-aideddetectionofbreastcancernuclei.IEEETrans.Inform.Technol.Biomed.1:128–140.31.ZHANG,B.,etal.1999.Featureselectionforcomputerizedmassdetectionindigitizedmammogramsbyusinggeneticalgorithms.Acad.Rad.6:327–332.32.YAMANY,S.M.,K.J.KHIANI&A.A.FAROOQ.1997.Applicationofneuralnetworksandgeneticalgorithmsintheclassificationofendothelialcells.Patt.Recog.18:1205–1210.33.ANASTASIO,M.A.,etal.1998.Ageneticalgorithmbasedmethodforoptimizingtheperformanceofacomputerbaseddiagnosisschemefordetectionofclusteredmicro-calcificationsinmammograms.Med.Physics.25:1613–1620.34.DAYHOFF,J.E.&J.M.DELEO.2001.Artificialneuralnetworks:openingtheblackbox.Cancer91:1615–1635.35.VANOSDOL,W.W.,T.G.MYERS&J.N.WEINSTEIN.2000.Neuralnetworktechniquesforinformaticsofcancerdrug.Meth.Enzymol.321:369–395.36.YU,S.&L.GUAN.2000.ACADsystemfortheautomaticdetectionofclusteredmicrocalcificationsindigitizedmammogramfilms.IEEETrans.Med.Imaging19:115–126.37.GAVRIELIDES,M.A.,etal.2000.Segmentationofsuspiciousclusteredmicrocalcifica-tionsinmammograms.Med.Phys.27:13–22.38.PENEDO,M.G.,etal.1998.Computer-aideddiagnosis:aneural-network-basedapproachtolungnoduledetection.IEEETrans.Med.Imaging17:872–880.39.WU,Y.,etal.1993.ArtificialneuralnetworksimMammographyapplicationtodeci-sionmakingdiagnosisofbreastcancer.Radiol.187:81–87.40.WILDING,P.,etal.1994.Applicationofbackpropagationneuralnetworkstothediag-nosisofbreastandovariancancer.CancerLett.77:159–193.41.LO,J.Y.,etal.1995.Computer-aideddiagnosisofbreastcancer:artificialneuralnet-workapproachforoptimizedmergingofmammographicfeatures.Acad.Radiol.2:841–850.42.LO,J.Y.,etal.1997.Predictingbreastcancerinvasionwithartificialneuralnetworksonthebasisofmammographicfeatures.Radiology189:555–558.43.ROWBOTTOM,C.G.,S.WEBB&M.OLDHAM.1999.Beam-orientationcustomizationusinganartificialneuralnetwork.Phys.Med.Biol.44:2251–2262.44.LONDON,J.W.&L.G.GOMELLA.2000.OverviewoftheInternetandprostatecancerresources.Sem.Urol.Oncol.18:245–253.\nHAQUEetal.:CANCERMANAGEMENT29745.PAUTLER,S.E.,etal.2001.Useoftheinternetforself-educationbypatientswithpros-tatecancer.Urology57:230–233.46.KIM,H.S.,E.KIM&J.W.KIM.2001.Developmentofabreastself-examinationpro-gramfortheInternet:healthinformationforKoreanwomen.CancerNurs.J.24:156–161.47.HOLZER,S.,etal.1999.Datawarehousingasatoolforqualitymanagementinoncol-ogy.Stud.HealthTechnol.Inform.68:432–435.48.LIU,J.M.,etal.2000.Integrationofcomputer-assembleddigitalimagesandtextdataasevidencefortheoncologicalrecord.J.Digit.Imaging13:55–59.49.BARTHELEMY-BRICHANT,N.,etal.1999.Evaluationoffrequencyandtypeoferrorsdetectedbyacomputerizedrecordandverifysystemduringradiationtreatment.Radiother.Oncol.53:149–154.50.RICKE,J.&H.BARTELINK.2000.Telemedicineanditsimpactoncancermanagement.Eur.J.Cancer36:826–833.51.LOWE,H.J.1999.Transformingthecancercenterinthe21stcentury.MDComp.16:40–42.52.PAPPALARDO,P.A.,etal.1998.Microdissection,microchiparrays,andmolecularanal-ysisoftumorcells.Sem.Rad.Oncol.8:217–223.53.WILKINSON,D.S.1997.Theroleoftechnologyintheclinicallaboratoryofthefuture.Review11:322–330.54.PERERA,F.P.&I.B.WEINSTEIN.2000.Molecularepidemiology:recentadvancesandfuturedirections.Carcinogenesis21:517–524.