正交信号校正用于子宫内膜癌组织的近红外光谱诊断的研究.pdfVIP

正交信号校正用于子宫内膜癌组织的近红外光谱诊断研究 张卓勇*张 家 进相玉红 首都师范大学化学系，北京 100048 摘要 光谱预处理对于光谱的解析和建模具有重要的影响。讨论了一种加强正交信号校正方法(emphatic orthogonalsignalcorrection，EOSC)，并与主成分正交信号校正(principalcomponent-orthogonalsignalcor rection，PC-OSC)做了比较 。为了展示EOSC的性能，本工作根据需要做了模拟NIR光谱，建立SVM分类 器。与PC-OSC相比，EOSC校正需要更多的成分数，但是并不会出现像PC-OSC模型中的过拟合现象。这 个结果在模拟NIRS中得到证实。本工作还引入新的数据降维方法，并和PC-LDA(principalcomponent-line ardiscriminantanalysis)进行对比。本作中提出基于SVM的思想建立的SVM-DA方法，避免了经典LDA 的假设条件等。经过降维后的数据，用于建立Logistic回归模型。通过Logistic回归模型的结果，不仅可知 道样本的类别归属，而且可以知道该样本和别的类别之间的关系。 关键词 子宫内膜癌；近红外光谱分析技术；预处理方法；加强正交信号校正；支持向量机判别分析 OrthogonalSignalCorrectionAppliedtoDiagnosisofEndometriaiCancer BasedonNearInfraredSpectroscopy ZHANGZhuo-yong*ZHANGJia-jinXIANGYu-hong DepartmentofChemistry,CapitalNormalUniversity,Beijing 100048,China Abstract MethodsforpreprocessingofNIRS(Nearinfaredspectroscopy)playanimportantroleinchemometricsbecauseoflow absorptionintensity,seriousbandoverlapping.Inthiswork,chemomerticsmethodswerecombinedwithNIRSforendometrial tissuesections.TwoeffectiveNIRSpreprocessingmethodshavebeenintroduced:EOSC(Emphasisorthogonalsignalcorrec tion)andSVM(Supportvectormachine)-baseddiscriminantanalysis.EOSCwasevaluatedandcomparedwithprincipalcompo nentorthogonalsignalcorrection(PC-OSC)byusingsupportvectormachine(SVM)classifiers.EOSCcombinedwiththeSVM gavebetterpredictionswithrespecttoalargernumberofcomponentsandwasnotassusceptibletooverfittingthedataasthe classifierbuiltwithPC-OSCdata.AnovelmethodcomparedwithPC-LDA(Principalcomponent-lineardiscriminantanalysis) fordimensionalityreductionofNIRSdatasethasbeenintroduced.Accordingtothelimitations,SVM-baseddiscriminantanaly sismethodhasbeenintroducedinthiswork.Afterdimensionalityreduction,logisticregressionmodelwasbuilt.Fromthepre dictedresultsoflogisticregressionmodel,boththecategoriesofsamplesandthetrendsofsamplesbelongingtootherclasswere clear. Keywords Endometrialcancer;Nearinfraredspectroscopy;Preprocessingmet