基于CT影像组学特征的机器学习模型对症状性颈动脉斑块的诊断价值

打开文本图片集

CT radiomics-based machine learning models in diagnosis for symptomatic carotid artery plaque

LIJing1²，WANGKaifeng3，ZHANGMeilan1²，SONGBowen1²，ZHAO Fengqiao²，DUHai²

GraduateSchlBatouedicalolgoferongoliUiesityfence&chologyotouina;t ofRadiologydontalialdosoa;olofsicicalieeseicalUeity 350001，China.

[Abstract]Objective：Toexplorethediagnosticeficiencyofthemachinelearning modelsbasedonCTradiomics features for symptomaticcarotidplaques.Methods：TeclinicaldataandheadandneckCTAimagesof18Opatientswereretrospectively analyzed.Atotalof18Opatientsweredivided intoasymptomaticgroup（54cases）andanasymptomaticgroup（126cases） basedonwhethertherewereischemiceventsintheipsilateraleye（transientmonocular blindnessortransientamaurosisor retinalinfarction）orischemiceventsinthecerebralhemispheres（transientischemicatackorstroke）withinthepast6months. 180caseswererandomlydividedintoatrainingset（144cases）andatestset（36cases）ataratioof8：2.Theradiomics featuresextracted from CTA images of the training set were selected through Mann-Whitney υ test，Spearman correlation analysis，andLASSOalgorithm.Themostoptimalfeatureswereretainedtoconstructfivemachinelearningmodels，XGBoost， SVM，LR，LightGBMandExtraTres.ROCcurvewasused toevaluate themodels’eficiency.Results：Inthetestset，the AUCsof the five modelsreachedO.723-0.786.ThepredictiveperformanceofExtraTresandLightGBMmodelsperformed well.TheAUC，sesitivitysecificityofExtrareesmodelinthetrainingsetwreO.781.663，0.761，andtoseintstet were0.786，0.643，0.875.AndtheAUC，sensitivityspecificityofLightGBMmodelinthetrainingsetwereO.829，0.735， 0.804 and those in the test set were O.77，0.643，0.875，respectively. Conclusion ∵ Machine learning models based on CT radiomics features can effectively identify symptomatic carotid plaques.

[Keywords] Carotid plaque;Tomography，X-ray computed;Machine learning;Radiomics;Symptomatic

脑卒中是全球第二大死亡因素[1]。（剩余9139字）