摘要
目的探究基于常规多序列MRI的影像组学模型在术前预测纤维型和非纤维型脑膜瘤的价值,以助于临床术前准备及预后评估。方法回顾性分析了自2013年3月至2022年12月经手术后病理证实的共317例脑膜瘤患者的临床及多序列MRI(包括T1WI、T2WI、T2WIFLAIR、T1WI增强)资料。手动勾画获得脑膜瘤强化区作为感兴趣区域(EnHROI),并分别将勾画区域向外周各自膨胀3 mm、5 mm得到EnH3mmROI、EnH5mmROI,对每个MRI序列的三种ROI分别提取影像特征,分别采用5折交叉验证法和留一法交叉验证(LOOCV)分别进行特征筛选、模型验证及比较。使用相关系数法和最小绝对收缩和选择算子(least absolute shrinkage and selection operator,LASSO)算法进行特征选择,随后使用支持向量机的机器学习算法(AVM)构建模型,最后评估不同预测模型的效能。结果利用受试者工作特征曲线(receiver-operating characteristic curve,ROC)评估预测脑膜瘤分型的影像组学模型中,基于EnH3mmROI的预测模型均优于基于EnHROI的模型及EnH5mmROI的模型。基于EnHROI模型AUC值为0.801、准确率为0.842;基于EnH3mmROI模型AUC值为0.858、准确率为0.842;基于EnH5mmROI模型AUC值为0.841、准确率为0.868。结论(1)基于EnH3mmROI所建立的影像组学预测模型效能明显优于基于EnHROI及EnH5mmROI所建立的预测模型。(2)基于常规多序列MRI图像建立的术前预测脑膜瘤组织学分型的影像组学模型具有一定的临床价值,有助于临床判断预后以及为治疗计划提供依据。
Objective To explore the value of radiomics models based on conventional multi-sequence MRI in distinguishing fibrous and non-fibrous meningiomas to aid in clinical preoperative preparation and prognostic assessment.Methods Clinical and multi-sequence MRI(including T1WI,T2WI,T2WIFLAIR,T1WI enhancement)data of a total of 317patients with postoperative pathologically confirmed meningiomas from March 2013to December 2022were retrospectively analyzed.The enhanced area of meningioma was manually outlined as the region of interest(EnHROI),and the outlined area was expanded to the periphery by 3mm and 5mm to obtain EnH3mmROI and EnH5mmROI,respectively,and the radiomics features were extracted from the three kinds of ROIs of each MRI sequence,and the 5-fold cross-validation and LOOCV method was used for feature screening and model validation.Feature selection was performed using the correlation coefficient method and the least absolute shrinkage and selection operator(LASSO)algorithm,followed by model construction using the support vector machine algorithm(AVM),and finally the efficacy of the different prediction models was evaluated.Results The EnH3mmROI-based prediction model outperformed both the EnHROI-based model and the EnH5mmROI-based model in the assessment of radiomics models for predicting meningioma typing using the receiver-operating characteristic curve(ROC).The AUC value of the EnHROI-based model was 0.801,with an accuracy of 0.842;the AUC value of the EnH3mmROI-based model was 0.858,with an accuracy of 0.842;and the AUC value of the EnH5mmROI-based model was 0.841,with an accuracy of 0.868.Conclusion①The effectiveness of the radiomics model based on EnH3mmROI was significantly better than that based on EnHROI and EnH5mmROI.②The radiomics model in predicting preoperative histologic typing of meningiomas based on conventional multi-sequence MRI has clinical value,which can help to determine the prognosis and provide a basis for the clinical treatment plan.
作者
孔春雨
莫展豪
程斯文
隋赫
宛迎春
吴帅
范晓飞
吕忠文
KONG Chunyu;MO Zhanhao;CHENG Siwen;SUI He;WAN Yingchun;WU Shuai;FAN Xiaofei;LV Zhongwen(Department of Radiology,China-Japan Union Hospital of Jilin University,Changchun130033,China;Department of Ultrasound,China-Japan Union Hospital of Jilin University,Changchun130033,China;Department of Endocrinology,China-Japan Union Hospital of Jilin University,Changchun130033,China)
出处
《中国实验诊断学》
2024年第3期278-286,共9页
Chinese Journal of Laboratory Diagnosis
基金
吉林省卫生健康科技能力提升项目(2021JC022)。
关键词
脑膜瘤
影像组学
组织学分型
磁共振成像
meningioma
radiomics
histologic typing
magnetic resonance imaging
