25医疗装备2024年1月第37卷第1期MedicalEquipment,January.2024,Vol.37,No.1基于多模态影像组学方法在脑胶质瘤高低分级中的应用价值叶姣,宾芳,鲁波,蔡泽润,胡玲(通信作者)湘潭市中心医院(湖南湘潭411100)〔摘要〕目的探讨基于多模态影像组学方法在脑胶质瘤高低分级中的应用价值。方法选择258例脑胶质瘤患者作为观察对象,高分化脑胶质瘤210例,低分化脑胶质瘤48例。所有患者按9‥1的比例分为训练组和测试组,选择增强肿瘤区域作为感兴趣区域(ROI),通过Pyradiomics开源库提取T1、T2、T1ce和Flair4个模态MRI共428组影像组学特征。选出具有最高预测价值的影像组学特征,构建对数几率回归(LR)、支持向量机(SVM)和多层感知机(MLP)3种机器学习模型进行脑胶质瘤分级,并对测试组进行验证。结果利用LR、SVM和MLP3种机器学习算法构建的影像组学模型在训练集的曲线下面积(AUC)均>0.95,测试组均>0.90。基于LR构建的影像组学模型最优,其在测试组中的准确率、AUC、敏感度和特异度分别为92.0%、0.976、90.5%和100.0%。结论基于多模态MRI影像组学特征结合机器学习分类模型可准确预测脑胶质瘤的高低分级。〔关键词〕影像组学;脑胶质瘤;机器学习〔中图分类号〕R737.33〔文献标识码〕B〔文章编号〕1002-2376(2024)01-0025-04〔DOI〕10.3969/j.issn.1002-2376.2024.01.007收稿日期:2023-07-12·论著·Multi-modalMRIRadiomicsforDistinguishingHigh-GradeandLow-GradeGliomasYeJiao,Binfang,Lubo,Caizerun,HuLing(CorrespondingAuthor).XiangtanCentralHospital,XiangtanHunan411100,China【Abstract】ObjectiveToexploretheapplicationvalueofmultimodalradiomicsmethodsinthehighandlowgradesofgliomatoassistclinicaldiagnosis.MethodsAtotalof258casesofgliomawereanalyzed,included210casesofwell-differentiatedgliomaand48casesofpoorly-differentiatedglioma.Patientswererandomlydividedintotrainingandtestinggroupsinaratioof9‥1.TheenhancedtumorRegionswereselectedasRegionsofinterest(ROI),andatotalof428groupsofradiomicsfeaturesfromfourMRImodalities(T1,T2,T1ceandFlair)wereextractedthroughthePyradiomicsopensourcelibrary.SpearmancorrelationtestandLeastAbsoluteShrinkageandSelectionOperator(LASSO)wereusedtoselecttheradiomicsfeatureswiththehighestpredictivevalueinthetraininggroup.ThenthreeMachinelear...