CharacterisingpatternsoftumourinvasioninTimothyJLarkin1,NatalieRBoonzaier1,LailaAMohsen2,andStephenJTargetaudience:ClinicianstreatingpatientswithglioblastomaandimagingscientistsinterestedinimaginggliomaPurpose:Gliomaisthemostcommonprimarybraintumourandithasatendencytoinvadealonganddisruptwhitemattertracts.Diffusiontensorimaging(DTI)issensitivetowhitematterintegrityandinthecaseofglioma posingthediffusiontensorintoanisotropiccomponent(p)andananisotropiccomponent(q)hasbeenusedpreviouslytoinvestigatepatternsofgliomainvasion.Characterisationofdifferentinvasivephenotypesingliomaisofinterestasitmayindicateareaswhereatumourislikelytoreoccur,andadditionally,theextentofinvasioncorrelateswithprogressionsurvival1.Methods:Diffusiontensorimagingwasperformedpreoperativelyforacohortof46patientswithhigh-gradeglioma(meanage54.8years,range22-73years;male=33,female=13)at3TonaTrio.Regionsofinterest(ROIs)weredrawnaroundthepandqabnormalitiesforeachtumourbyanexperiencedneurosurgeonusingImageJ.Thepatternofinvasionwasthenclassifiedaseitherminimally(n=7),locally(n=14)ordiffusely(n=23)invasivebyvisualinspectionoftherelativeextentofthetwoabnormalitiesaccordingtopreviouslypublishedcriteria2.Theinter-raterreliabilityofassigninginvasivepatternswasassessedonasubsetof30patientsusingROIsdrawntlybyasecondexperiencedneuroradiologist.WesoughttodevelopanobjectivemethodofdefiningthedifferentinvasivephenotypesbymeasuringthedistancweentheedgesofthepandqROIsonaslicebyslicebasisintwodimensions.TheHausdorffdistancweenthecontoursdefiningtheregionsofinterestwascalculatedusing.RatherthanjustreporttheumdistancweentheROIswealsocalculatedthemedianandthe90thpercentileofthedistances.DifferencesbetweentheinvasivepatternswereexploredusingANOVAwiththeTukey-KramerHSDpost-hoctestandasignificancelevelof0.05(IBMSPSSver22).Results:Therewasagreementontheinvasivepatternbetweenthetworatersin90%ofcases,givingakappastatisticof0.81,suggestinggoodinter-raterreliability,withaconsensusbeingusedasesofdisagreement.Significantdifferenceswereobservedinthemediandistancesbetweentheedgesfortumoursthatwereminimally(2.4±0.4mm)versuslocally(4.3±2.0mm)invasive(P< DistanceDistance

90th

Figure1.Plotshowingcomparisonbetweenmedian,90thpercentileand HausdorffdistancesbetweenedgepixelsoftheofthepandqROIsfordifferentpatternsofglioblastomainvasion0.05),andbetweenminimallyanddiffusely(5.9±2.8mm)invasivetumours(P<0.01).Similarly,theumdistancweenROIswassignificantlydifferentfortheminimally(17±4mm)comparedtolocally(30±12mm)invasivetumours(P<0.05)andbetweentheminimallyanddiffusively(36±12mm)invasivetumours(P<0.05).The90thpercentileoftheHausdorffdistancewassignificantlydifferentbetweenlocally(13±5mm)anddiffusely(17±5mm)invasivetumours(P<0.05).Discussion:Theseresultsshowthatthevisuallyassignedinvasionpatternsareindeeddistt.WhilelocallyanddiffuselyinvasivetumoursdonothavesignificantlydifferentumHausdorffdistancesbyitispossibletodistinguishthemusingthefulldistancedistributionprofile.FurtherworkisunderwaytodevelopanautomatedclassifierforassigningdifferentpatternsofgliomainvasionfromDTIdata,whiayprovideprognosticinformationandidentifyareasofinvasionformoreaggressivetargetingoftherapy.Conclusion:Patternsofgliomainvasionclassifiedvisuallywereshowntobedifferentbyanobjectiveysis.MohsenLA,ShiV,JenaR,GillardJH,PriceSJDiffusiontensorinvasivephenotypescanpredictprogression-survivalinglioblastomas.BrJNeurosurg.2013;27:436-441PriceSJ,JenaR,Bu NG,CarpenterTA,PickardJD,GillardJH.Predictingpatternsofgliomarecurrenceusingdiffusiontensorimaging.Eur.Radio.2007;17:1675-1684Proc.Intl.Soc.Mag.Reson.Med.23 TimothyJLarkin1、NatalieRBoonzaier1、LailaAMohsen2和StephenJPrice1(DTI)(p)(q)方法：術前在西門子Trio上對46名高級別膠質瘤患者（平均年齡54.8歲，范圍22-73歲；3313）3TImageJpqROI2，通過目視檢查兩種異常的相對程度，將侵襲模式分為輕微侵襲（n=7）、局部侵襲（n=14）或彌漫性侵襲（n=23）ROI，30維上逐個切片地測量p和qROI邊緣之間的距離。使用計算定義感興趣區(qū)域的輪廓之間的豪距離。我們不僅報告ROI之間的最大距離，還計算了距離的中位數(shù)和第90個百分位ANOVATukey-KramerHSD事后檢驗以及0.05的顯著性水平(IBMSPSSver22)探90%，kappa0.81，表明評估者間的(2.40.4mm)腫瘤(4.3±2.0mm)邊緣之間的中位距離觀察到顯著差異(P<60.0050.0040.0030.0020.0010.001pqROI90Hausdorff0.05），以及最小和彌漫性（5.92.8mm）侵襲性腫瘤之間（P0.01）(174mm)(3012mm)(P0.05(3612mm)侵襲性腫瘤之間的ROI之間的最大距離存在顯著差異(P<0.05)。P<0.05