多参数磁共振成像影像组学鉴别高级别胶质瘤及单发脑转移瘤的价值

The value of multi-parameter magnetic resonance imaging radiomics in distinguishing high-grade gliomas from single brain metastases

目的:利用多参数磁共振成像(magnetic resonance imaging,MRI)影像组学鉴别高级别胶质瘤及单发脑转移瘤。方法:回顾性收集经病理活检证实为高级别胶质瘤或单发脑转移瘤患者的MRI结构序列及功能序列表观扩散系数(apparent diffusion coefficient,ADC)影像,共103例。手动勾画感兴趣区域容积(volumes of interest,VOI),包括肿瘤核心区域、周围水肿区域以及肿瘤全域。使用Python中的开源Pyradiomics包进行影像组学特征提取。然后选择t检验和最小绝对收缩和选择算子(least absolute shrinkage and selection operator,LASSO)方法进行特征筛选及降维。最后使用逻辑回归(Logistic regression,LR)、随机森林(random forest,RF)、支持向量机(support vector machines,SVM)及K-邻近(K-nearest neighbors,KNN)四种分类器建模,对比鉴别高级别胶质瘤及单发脑转移瘤的效能。结果:高级别胶质瘤及单发脑转移瘤患者的性别及年龄之间不存在统计学差异,肿瘤位置之间存在统计学差异。基于肿瘤核心区域鉴别两者取得最高的诊断效能,曲线下面积(area under curve,AUC)最高为0.924。在多序列组合模型中,ALL_TCR_LR模型的AUC值最高,为0.924,被选为最优分类器模型。结论:基于肿瘤核心区域的多序列影像组学使用LR机器学习分类器可实现高级别胶质瘤及单发脑转移瘤的鉴别,为临床决策和实践提供了较大的帮助。

Objective:To distinguish high-grade gliomas from single brain metastases using multi-parameter magnetic resonance imaging(MRI)imaging.Methods:Traditional MRI sequences and apparent diffusion coefficient(ADC)images of 103 patients with high-grade glioma or single brain metastases confirmed by pathological biopsy were retrospectively collected.Volumes of interest(VOI)is manually delineated,includingthe tumor core region,the peripheral edema region and the entire tumor region.Using Pyradiomics package in Python to extract radiomics features.Then t-test and least absolute shrinkage and selection operator(LASSO)method are selected for feature screening and dimensionality reductio.Logistic regression(LR),random forest(RF),support vector machines(SVM)and K-nearest neighbors(KNN)were used to establish the model and compare the effect of distinguishing these brain tumors.Results:No significant differences were observed in terms of gender and age among patients with high-grade glioma and single brain metastasis.The highest diagnostic efficiency was obtained based on tumor core region differentiation,and the highest area under curve(AUC)was 0.924.In the multi-sequence combination model,the ALL_TCR_LR combination exhibited the highest AUC value of 0.924 on the test set,which was chosen as the optimal classifier combination.Conclusion:The multi-sequence radiomics based on the tumor core region can realize the identification of high-grade glioma and single brain metastases using LR machine learning classifier,which provides a great help for clinical decision-making and practice.