基于多序列MRI影像组学模型预测脑膜瘤病理分级的价值

The value of radiomics models based on multi-sequence MRI in predicting the pathological grading of meningiomas

目的:探讨基于多序列MRI影像组学模型预测脑膜瘤病理分级的价值。方法:回顾性分析经手术病理证实的215例脑膜瘤患者的临床及MRI资料。其中,低级别组174例,高级别组41例。将所有患者按照7∶3的比例随机分为训练组和验证组。采用ITP-SNAP软件,分别在T_(2)WI、DWI和对比增强T_(1)WI图像上勾画肿瘤的三维ROI,使用AK软件提取影像组学特征。采用Spearman相关性分析及多元Logistic回归分析筛选组学特征并构建影像组学标签。使用ROC曲线下面积(AUC)评价影像组学模型的预测效能。结果:高级别组与低级别组之间年龄和性别构成的差异均无统计学意义(P>0.05)。基于T_(2)WI、DWI和对比增强T_(1)WI的单序列及多序列联合影像组学模型预测高、低级别脑膜瘤的AUC均大于0.700。基于单序列的影像组学模型中,增强T_(1)WI在训练组和验证组的AUC分别为0.942和0.913,均高于其它两个序列。基于MRI多序列联合的影像组学模型预测高、低级别脑膜瘤的AUC值最高,在训练组的AUC为0.950,在验证组的AUC为0.923。结论:MRI影像组学模型能够预测脑膜瘤的病理分级,尤其是多序列联合的影像组学模型对脑膜瘤病理分级具有较高的预测效能。

Objective:To explore the value of radiomics models based on multi-sequence MRI in predicting pathological grading of meningioma,providing reference for the choice of clinical treatment schemes.Methods:Clinical and MRI data of 215 cases of meningioma were retrospectively analyzed.The subjects were divided into the low-grade group and high-grade group according to pathological results.All patients were divided into the training group and validation group according to a ratio of 7∶3.Tumor volumes were segmented on T_(2)WI,DWI and enhanced T_(1)WI images respectively using ITP-SNAP software,and radiomics features were extracted using AK software.Spearman correlation analysis and multiple logistic regression analysis were used to select the features,and radiomics labels were constructed.The area under ROC curve(AUC)was used to evaluate the predictive performance of the radiomics models.Results:There was no statistical significance in age and gender between the high-grade group and the low-grade group of meningioma(P>0.05).The AUC values of T_(2)WI,DWI,enhanced T_(1)WI single sequence and multi-sequence combined radiomics models were all more than 0.700 for predicting the high and low grade meningiomas.In the single sequence radiomics models,the AUC values(training group:0.942;validation group:0.913)of enhanced T_(1)WI sequence radiomics model for predicting the pathological grading of meningioma were more than that ofthe other two single sequence radiomics models.AUC value of the radiomics model combined multi-sequence MRI in predicting the pathological grading was the highest,with 0.950 in the training group and 0.923 in the validation group.Conclusion:The MRI radiomics models can predict the pathological grading of meningioma,especially the radiomics model combined multi-sequence MRI with high predictive efficiency.