摘要
目的探讨T2WI加权成像(T2 weighted imaging,T2WI)、对比增强T1加权成像(contrast enhanced T1 weighted imaging,CE-T1WI)瘤体和瘤周影像组学特征联合常规因素鉴别纤维型和非纤维型脑膜瘤的临床价值。材料与方法纳入经病理证实的108例脑膜瘤患者,包括30例纤维型脑膜瘤、78例非纤维型脑膜瘤,按7∶3的比例分为训练集(n=76)和测试集(n=32)。在训练集中,从T2WI、CE-T1WI序列的瘤体、瘤周中分别提取1132个影像组学特征,采用最大相关最小冗余方法(min-redundancy and max-relevance,mRMR)及最小绝对收缩和选择算子(least absolute shrinkage and selection operator,LASSO)筛选最佳影像组学特征子集,采用逐步逻辑回归(logistic regression,LR)机器学习方法构建影像组学模型:T2WI瘤体、T2WI瘤周、CE-T1WI瘤体、CE-T1WI瘤周、(T2WI+CE-T1WI)瘤体、(T2WI+CE-T1WI)瘤周和(T2WI+CE-T1WI)瘤体+瘤周7种模型。通过单因素、多因素逻辑回归分析方法筛选出有意义(P<0.05)的常规因素。然后,联合鉴别效能最佳的组学模型与常规因素生成列线图,通过受试者工作特征(receiver operating characteristic,ROC)曲线下面积(area under the curve,AUC)评价列线图的诊断效能,由决策曲线分析(decision curve analysis,DCA)的净收益值评估该模型的临床应用效能。在测试集中验证其鉴别效能。结果T2WI瘤体、T2WI瘤周、CE-T1WI瘤体、CE-T1WI瘤周、(T2WI+CE-T1WI)瘤体、(T2WI+CE-T1WI)瘤周和(T2WI+CE-T1WI)瘤体+瘤周影像组学模型在训练集的AUC值分别为0.925、0.803、0.837、0.872、0.902、0.894、0.908,在测试集的AUC值分别为:0.652、0.812、0.700、0.725、0.700、0.816、0.729。T2WI瘤体影像组学模型鉴别纤维型与非纤维型脑膜瘤的AUC在训练集、测试集中分别为0.92、0.65,出现过拟合;(T2WI+CE-T1WI)瘤周影像组学模型在测试集中的AUC值最高,该模型诊断效能最佳。建立(T2WI+CE-T1WI)瘤周组学模型与常规因素(T2WI信号强度和瘤周水肿)的联合模型鉴别效能有所提高,其在训练集、测试集的AUC分别为0.89、0.82。校准曲线显示该模型术前鉴别纤维型、非纤维型脑膜瘤预测概率与实际概率之间的一致性良好,DCA结果显示该模型有良好的临床应用效能。结论多模态MRI组学模型可有效鉴别纤维型及非纤维型脑膜瘤,联合常规因素后可进一步提升该模型鉴别效能。
Objective:To investigate the clinical value of T2WI-weighted imaging(T2WI),contrast-enhanced T1-weighted imaging(CE-T1WI)of the tumour body and peritumour in combination with conventional factors in identifyingfibrous and non-fibrous meningiomas.Materials and Methods:A total of 108 patients with pathologically confirmed meningiomas,including 30 fibrous meningiomas and 78 non-fibrous meningiomas,were enrolled and divided into a training set(n=76)and a validation set(n=32)in a ratio of 7:3.In the training set,1132 radiomics features were extracted from the tumour body and peri-tumour of T2WI and CE-T1WI sequences,respectively.The optimal subset of radiomics features was identified through the maximal correlation minimal redundancy method(mRMR)and the least absolute shrinkage and selection operator(LASSO).Logistic regression(LR)machine learning method to construct imaging genomics models:T2WI tumour,T2WI peritumour,CE-T1WI tumour,CE-T1WI peritumour,(T2WI+CE-T1WI)tumour,(T2WI+CE-T1WI)peritumour and(T2WI+CE-T1WI)tumour+peritumour.The conventional factors with significance(P<0.05)were screened by single-factor and multifactor logistic regression analysis methods.Then,the radiomics model with the best discriminatory efficacy was combined with the conventional factors to generate nomograms,and the diagnostic efficacy of the nomograms was evaluated by AUC,and the clinical efficacy of the model was assessed by the net benefit value of the decision curve analysis(DCA).the efficacy of this model was validated in the test set.Results:The AUC values for the T2WI tumour,T2WI peritumour,CE-T1WI tumour,CE-T1WI peritumour,(T2WI+CE-T1WI)tumour,(T2WI+CE-T1WI)peritumour and(T2WI+CE-T1WI)tumour+peritumour radiomics models in the training set were 0.925,0.803,0.837,0.872,0.902,0.894,0.908,respectively.In the test set,the corresponding values were 0.652,0.812,0.700,0.725,0.700,0.816,0.729.The AUC of the T2WI tumour radiomics model for identifying fibrous and non-fibrous meningiomas was 0.92 in the training set and 0.65 in the test set.This appeared to be an overfitting.The(T2WI+CE-T1WI)peritumour radiomics model had the highest AUC value in the test set,and the model demonstrated the best diagnostic efficacy.The discriminatory efficacy of the established(T2WI+CE-T1WI)peri-tumour radiomics model was improved from the combined model with conventional factors(T2WI signal intensity and peri-tumour oedema),and its AUCs in the training set and test set were 0.89 and 0.82,respectively.The calibration curves showed good agreement between the predicted and actual probabilities of the model's preoperative identification of fibrous and non-fibrous meningiomas,DCA results show good clinical efficacy for this model.Conclusions:Multimodal MRI radiomics models can effectively identify fibrous and non-fibrous meningiomas,and their discriminatory efficacy can be futher improved when combined with conventional factors.
作者
杨慧敏
李文鑫
姜兴岳
王倩倩
张濬韬
刘新疆
YANG Huimin;LI Wenxin;JIANG Xingyue;WANG Qianqian;ZHANG Juntao;LIU Xinjiang(Department of Radiology,Shanghai Pudong Hospital(Pudong Hospital of Fudan University),Shanghai 201399,China;Department of Radiology,the Affiliated Hospital of Binzhou Medical College,Binzhou 256603,China;GE Healthcare PDX GMS medical affairs,Shanghai 200203,China)
出处
《磁共振成像》
北大核心
2025年第8期50-57,共8页
Chinese Journal of Magnetic Resonance Imaging
基金
上海市卫生健康委员会卫生行业临床研究专项立项项目(编号:202140266)
上海市浦东新区系统学科建设项目(编号:PWZbr2022-16)。
关键词
脑膜瘤
影像组学
分型
瘤周组织
磁共振成像
meningioma
radiomics
typing
peri-tumour
magnetic resonance imaging
