IVIM、DKI联合DCE-MRI的影像组学在预测乳腺癌HER-2表达状态中的应用价值

Application value of IVIM,DKI and DCE-MRI radiomics predicting HER-2 expression in breast cancer

目的 探讨联合体素内不相干运动(intravoxel incoherent motion, IVIM)、扩散峰度成像(diffusion kurtosis imaging,DKI)和动态对比增强磁共振成像(dynamic contrast-enhanced magnetic resonance imaging, DCE-MRI)、参数图构建影像组学模型预测乳腺癌患者人类表皮生长因子受体2(human epidermal growth factor receptor 2, HER-2)的表达状态。材料与方法 回顾性分析192例乳腺癌患者病例资料,根据患者的病理结果分为HER-2表达阳性组(48例)和HER-2表达阴性组(144例),术前均行IVIM、DKI及DCE-MRI。并按照8∶2的比例将病例随机分为训练集(154例)和测试集(38例)。在灌注分数(perfusion fraction, f)、灌注相关扩散系数(perfusion related diffusion coefficient, D^(*))、真实扩散系数(real diffusion coefficient, D)、平均扩散率(mean diffusivity, MD)和平均扩散峰度值(mean kurtosis, MK)参数图和第2期DCE-MRI(DCE-2)图像中勾画出病变区域的三维感兴趣区(region of interest, ROI),并提取其中的影像组学特征。采用Z分数归一化对特征进行标准化处理,并使用K最佳、最小冗余最大相关(max-relevance and min-redundancy, mRMR)、最小绝对收缩与选择算子回归(least absolute shrinkage and selection operator, LASSO)算法依次对特征进行降维和选择,通过logistic逻辑回归(logistic regression, LR)分类器分别建立参数图模型及联合模型,并采用5折交叉验证法验证模型的稳定性。通过受试者工作特征(receive operating characteristic,ROC)曲线和曲线下面积(area under the curve, AUC)对不同参数图像模型及联合模型的诊断效能进行分析,使用DeLong检验对各模型间ROC曲线进行比较,使用决策曲线分析(decision curve analysis, DCA)对模型的临床价值进行评估。结果 从每个ROI中提取了2286个MRI特征,在f、D^(*)、D、MD、MK参数图、第2期DCE-MRI和联合序列中分别筛选得到7、6、7、6、7、12、10个特征与HER-2表达状态相关。f、D^(*)、D、MD、MK参数图模型及第2期DCE模型在测试集中的AUC分别为0.693、0.679、0.586、0.682、0.661、0.732;联合模型在测试集中的AUC为0.861 (95%CI:0.775~0.958),敏感度和特异度分别为100.0%和71.4%,经DeLong检验,训练集中联合模型与f、D、D^(*)、MD、MK参数图模型及DCE-2模型之间AUC差异均有统计学意义(P均<0.05)。结果表明联合模型对预测HER-2的表达状态优于单一模型。结论 基于DCE-MRI、IVIM和DKI的影像组学联合模型可以在术前有效预测乳腺癌患者的HER-2表达状态,有助于临床对乳腺癌进行诊断、分型、制订治疗方案及预后。

Objective:To explore the intravoxel incoherent motion(IVIM),diffusion kurtosis imaging(DKI)and diagnostic value of radiomics models based on dynamic contrast-enhanced magnetic resonance imaging(DCE-MRI),in prediction of human epidermal growth factor receptor 2(HER-2)positive status in breast cancer patients.Materials and Methods:The clinical data of 192 patients with breast cancer were analyzed retrospectively.Patients were divided into HER-2 positive group(48 cases)and HER-2 negative group(144 cases)based on their pathological results.All patients underwent IVIM,DKI,and DCE-MRI scans before surgery.And then these data were randomly divided into training sets(n=154)and test sets(n=38)at a ratio of 8∶2.The three-dimensional volume region of interest of the tumor was manually delineated on the perfusion fraction(f),perfusion related diffusion coefficient(D^(*)),real diffusion coefficient(D),mean diffusivity(MD)and mean kurtosis(MK)parameter maps and the second phase of dynamic contrast-enhanced MRI,and radiomics features were extracted.The Z-score normalization was used for feature normalization,and the Select K Best,max-relevance and min-redundancy(mRMR)and least absolute shrinkage and selection operator(LASSO)were used to single out the most valuable radiomic features.The parametric map models and a combined model were established by logistic regression(LR)classifier,and the stability of the models was verified by the 5-fold cross-validation.The receive operating characteristic(ROC)curve and area under the curve(AUC)were used to evaluate the efficacy of the model.In addition,the DeLong test was used to compare the models,and decision curve analysis(DCA)was used to evaluate the models.Results:A total of 2286 radiomics features were extracted from each ROI,and 7,6,7,6,7,12 and 10 features were selected from the f,D^(*),D,MD,and MK parametric maps,the second phase of dynamic contrast-enhanced MRI(DCE-2)and combined sequence,respectively,which were related to breast cancer HER-2 status.The AUC of the f,D^(*),D,MD,and MK models and the DCE-2 model in the test group were 0.693,0.679,0.586,0.682,0.661 and 0.732,respectively.The AUC of the combined model in the test group was 0.861(95%CI:0.775-0.958).The sensitivity and specificity were 100.0%and 71.4%.By DeLong's test,in the training set there were statistically significant differences between combined model and the f model,the D model,the D^(*)model,the MD model,the MK model and the DCE-2 model(P<0.05).The results showed that the combined model was better than the single parameter diagram model in predicting the status of HER-2.Conclusions:The combined radiomics model based on DCE-MRI,IVIM and DKI can better predict the expression status of HER-2 in breast cancer patients,which is important for the diagnosis,treatment and prognosis of breast cancer.