目的 探讨基于T2WI及增强T1WI(contrast enhanced T1WI,CE-T1WI)MRI的影像组学模型在术前无创预测肺癌患者脑转移瘤表皮生长因子受体(epidermal growth factor receptor,EGFR)突变的价值.材料与方法 多中心回顾性分析2016年12月至2021年10月在中山大学孙逸仙纪念医院、中山大学肿瘤防治中心、安徽省立医院进行脑转移瘤手术切除并行EGFR基因检测的肺癌患者的临床及影像资料,共纳入103例患者(118个脑转移瘤),以中山大学肿瘤防治中心、安徽省立医院80例患者为训练集,共89个脑转移瘤(EGFR突变型49个,EGFR野生型40个),中山大学孙逸仙纪念医院23例患者为测试集,共29个脑转移瘤(EGFR突变型23个,EGFR野生型6个).在CE-T1WI图像勾画脑转移瘤轮廓获取感兴趣体积(volume of interest,VOI),将CE-T1WI-VOI复制至配准后的T2WI图像获取脑转移瘤的T2WI-VOI,采用PyRadiomics软件从上述VOI提取脑转移瘤影像组学特征,采用mRMR和LASSO-logistics筛选影像组学特征,构建EGFR基因突变的预测模型,采用受试者工作特征(receiver operating characteristic,ROC)曲线的曲线下面积(area under the curve,AUC)评价影像组学模型对EGFR基因突变的预测效能.采用校准曲线和Hosmer-Lemeshow test(HL)评价模型校准度.采用决策曲线(decision curve analysis,DCA)评价影像组学模型预测EGFR基因突变状态的临床净获益.结果 年龄、性别、病理类型、脑转移瘤数目、脑转移瘤直径在训练集和测试集中EGFR突变型及野生型的组间差异均无统计学意义(P>0.05).在CE-T1WI及T2WI图像上共提取了3 190个脑转移瘤的影像组学特征,最终筛选出了4个基于CE-T1WI图像和5个基于T2WI图像的影像组学特征构建预测EGFR基因突变状态的模型.该模型预测EGFR基因突变状态在训练集中AUC为0.828,准确度、敏感度、特异度分别为76.4%、81.6%、70.0%;在测试集中AUC值为0.783,准确度、敏感度、特异度分别为82.8%、95.7%、33.3%,均表现出较高的预测效能.影像组学模型训练集和测试集的校准曲线显示EGFR基因突变状态的预测概率与实际概率的一致性较好(HL>0.05).DCA显示影像组学模型在13.8%～87.2%的阈值范围内具有较强的临床适用性.结论 基于T2WI及CE-T1WI MRI构建的影像组学模型能够术前无创预测肺癌患者脑转移瘤EGFR基因突变状态.
A MRI-Radiomics-based model predicts EGFR mutation status in brain metastases in lung cancer patients
Objective:To explore the value of a radiomics model based on T2-weighted imaging(T2WI)and contrast-enhanced T1-weighted imaging(CE-T1WI)magnetic resonance imaging(MRI)in non-invasive preoperative prediction of epidermal growth factor receptor(EGFR)mutation status in brain metastases of lung cancer patients.Materials and Methods:We retrospectively collected clinical and MRI data of all lung cancer patients who underwent surgical resection of brain metastases and EGFR gene testing at Sun Yat-sen Memorial Hospital,Sun Yat-sen University,Sun Yat-sen University Cancer Center and Anhui Provincial Hospital from December 2016 to October 2021.A total of 103 patients(118 brain metastases)were included,with 80 patients(89 brain metastases)in the training set(Sun Yat-sen University Cancer Center and Anhui Provincial Hospital)and 23 patients(29 brain metastases)in the test set(Sun Yat-sen Memorial Hospital,Sun Yat-sen University).The enhanced edges of brain metastases were delineated on the CE-T1WI images to obtain the volume of interest(VOI),which was then copied to the registered T2WI images to obtain the T2WI-VOI of brain metastases.Radiomics features of brain metastases were extracted from these VOI using the PyRadiomics software,and feature selection was performed using mRMR and LASSO-logistics methods.A predictive model for EGFR mutation status was constructed,and the performance of the radiomics model in predicting EGFR mutations status was evaluated using the area under the receiver operating characteristic curve(AUC).The calibration curve and Hosmer-Lemeshow(HL)test were used to evaluate the calibration of the model,and decision curve analysis(DCA)was used to assess the clinical net benefit of the radiomics model in predicting EGFR mutation status.Results:In the training set and test set,there were no statistically significant differences in age,sex,pathological type,number of brain metastases,and diameter of brain metastases between the EGFR mutation and wild-type groups in terms of clinical data and MRI features(P>0.05).A total of 3 190 radiomics features of brain metastases were extracted from the CE-T1WI and T2WI images.Finally,four radiomics features based on CE-T1WI and five radiomics features based on T2WI were selected,and a radiomics model for predicting EGFR mutation status was constructed using these features.The radiomics model showed good predictive performance with an AUC of 0.828,accuracy of 76.4%,sensitivity of 81.6%,and specificity of 70.0%in the training set and an AUC of 0.783,accuracy of 82.8%,sensitivity of 95.7%,and specificity of 33.3%in the test set.The calibration curves of the radiomics model in the training and test sets showed good consistency between the predicted probabilities and the actual probabilities of EGFR mutation status(HL>0.05).DCA demonstrated the clinical utility of the radiomics model within a threshold range of 13.8%-87.2%.Conclusions:The radiomics model based on T2WI and CE-T1WI MRI can serve as an auxiliary tool for non-invasive preoperative prediction of EGFR mutation status in brain metastases of lung cancer patients.
国家科技期刊平台
NSTL
万方数据
