ThevalueofCTimagetextureanalysisinthedifferentialdiagnosisofbenignandmalignantparotidtumors / 实用放射学杂志

Objective ToinvestigatethevalueofCTimagetextureanalysisindifferentiatingbenignandmalignantparotidtumors inadults.Methods Thirty-threepatientswithparotidtumorswereretrospectivelyenrolledfromAugust2013toApril2018.Thehistological resultswereconfirmedthroughsurgeryorbiopsy,andtheyweredividedintothebenign(22cases)groupandthemalignant(11cases) groupaccordingly.AllpatientsunderwentCTscanswithcontrastenhancementbeforeanytreatment.ROIwasusedintextureanalysis forextractingfeaturesofthemaximumsliceofthetumorsinCTimages.Textureparametersbetweenbenignandmalignantgroups werecompared.ThestatisticallysignificantparameterswerethenanalyzedwithROC.Results CTtextureparametersshowedsignificant differencesbetweenthebenignandthemalignantgroups(P＜0.05),includingClusterShade,Sumentropy;GrayLevelNonUniformity, LargeAreaHighGrayLevelEmphasis,Contrast,ZoneEntropy,GrayLevelVariance,Sum Entropy;Entropy,IDMN,Sum Entropy, DifferenceEntropy,GrayLevelNonUniformityNormalized,andSmallAreaLowGrayLevelEmphasisthatoriginatedfromconventional, arterialandvenousphaseCTimagesrespectively.Sum Entropyfrom CTvenousphasewasconsideredasthebestdifferentiating parameter(AUC＝0.88,P＜0.05).WithSum Entropycutoffvalueof1.53,thesensitivityandspecificityindifferentiatingthebenign parotidtumorswere100%and72.70%respectively.Conclusion CTtextureanalysiscouldbeappliedclinicallyindifferentiatingbenign andmalignantparotidtumors.

MRtextureanalysisinthedifferentialdiagnosisofintracranialsolitaryfibroustumor/hemangiopericytomasandvascularmeningioma / 实用放射学杂志

Objective ToinvestigatethevalueofimagetextureanalysisbasedonconventionalMRimagesinthedifferentialdiagnosisof intracranialsolitaryfibroustumor/hemangiopericytoma(SFT/HPC)andvascularmeningiomas.Methods 12casesofSFT/HPCand 16casesofvascularmeningiomaconfirmedbypathologywerecollected.Variousdiscriminantanalysismethodsandimagetextureanalysis techniqueswereappliedtopre-operativeroutineMRIimagesofthebrain,andthebesttextureparameterswereselectedtoclassify cases,includinglineardiscriminantanalysis(LDA),nonlineardiscriminantanalysis(NDA),principalcomponentanalysis(PCA)and rawdataanalysis(RDA).Finally,thebestclassificationsequencetextureparameterswerechosenforstatisticalanalysis.Results The enhancedT1WIwasthebestclassificationsequence.ThePOE+ACC methodhadthelowestmisclassificationrate.Theabsolutegradient skewness,theautocorrelationofthecooccurrencematrix,andthevarianceofthehistograminthetexturefeaturesoftheSFT/HPC andvascularmeningiomasweredifferent(P<0.05).Conclusion ThatbasedontheconventionalMRimagetextureanalysiscanprovidemore quantitativeinformation,anew methodandideafortheidentificationofintracranialSFT/HPCandvascularmeningioma.