DCE-MRI联合DKI的影像组学在预测三阴性乳腺癌中的应用价值  

Application value of radiomics based on DCE-MRI combined with DKI in predicting triple-negative breast cancer

作　　者：关传国 邵硕[2] 郑宁[2] 陈文静 赵晓萌 吴建伟 GUAN Chuanguo;SHAO Shuo;ZHENG Ning;CHEN Wenjing;ZHAO Xiaomeng;WU Jianwei(Clinical Medical College,Jining Medical University,Jining 272013,China;Magnetic Resonance Imaging Room,Jining First People's Hospital,Jining 272000,China;United Imaging Intelligence Medical Technology Co.,Ltd.,Beijing 100089,China)

机构地区：[1]济宁医学院临床医学院,济宁272013 [2]济宁市第一人民医院磁共振室,济宁272000 [3]联影智能医疗科技(北京)有限公司,北京100089

出　　处：《磁共振成像》2025年第2期35-43,共9页Chinese Journal of Magnetic Resonance Imaging

基　　金：济宁市重点研发计划项目(编号:2023YXNS117)。

摘　　要：目的基于动态对比增强MRI(dynamic contrast-enhanced MRI,DCE-MRI)和扩散峰度成像(diffusion kurtosis imaging,DKI)参数图构建影像组学模型,评估其在预测三阴性乳腺癌(triple-negative breast cancer,TNBC)中的应用价值。材料与方法回顾性分析165例乳腺癌患者病例资料,根据患者的病理结果分为非TNBC组(120例)和TNBC组(45例)。所有患者术前均接受DCE-MRI和DKI检查。按照8∶2的比例随机分为训练集(n=132)和测试集(n=33)。在第2期DCE-MRI图像、平均扩散峰度值(mean kurtosis,MK)和平均扩散率(mean diffusivity,MD)参数图中勾画出病变区域的三维感兴趣区(three-dimensional region of interest,3D ROI),并提取影像组学特征。使用K最佳、最小冗余最大相关(max-relevance and min-redundancy,mRMR)以及最小绝对收缩和选择算子回归(least absolute shrinkage and selection operator,LASSO)算法依次对特征进行降维和选择,然后,通过逻辑回归(logistic regression,LR)分类器分别建立第2期DCE-MRI模型、DKI参数图模型(MD+MK、MD、MK)及联合模型(DCE-MRI+MD+MK),并采用5折交叉验证法验证模型的稳定性。模型的预测性能通过受试者工作特征(receiver operating characteristic,ROC)曲线和曲线下面积(area under the curve,AUC)进行评估,并使用DeLong检验分析模型间的统计学差异。最后,通过决策曲线分析(decision curve analysis,DCA)评估影像组学模型在临床中的应用价值。结果从每个序列3D ROI中分别提取了2286个影像组学特征,从第2期DCE-MRI、MD+MK、MD、MK及DCE-MRI+MD+MK中分别选取了8、9、12、7、21个特征与TNBC相关。第2期DCE-MRI模型、MD+MK模型、MD模型和MK模型在测试集的AUC分别为0.810、0.769、0.676、0.625;联合模型(DCE-MRI+MD+MK)在测试集中的AUC是0.884,其准确率、敏感度和特异度分别为78.8%、79.2%和77.8%。最后,把临床特征与影像组学特征进行联合建立列线图模型。结果表明,影像组学联合模型(DCE-MRI+MD+MObjective:To construct a radiomics model based on dynamic contrast-enhanced MRI(DCE-MRI)and diffusion kurtosis imaging(DKI),and evaluate its diagnostic value for triple-negative breast cancer(TNBC).Materials and Methods:A retrospective analysis was performed on the clinical data of 165 breast cancer patients,who were divided into a non-TNBC group(120 cases)and a TNBC group(45 cases)based on pathological results.All patients underwent preoperative DCE-MRI and DKI scans.The patients were randomly split into a training set(n=132)and a test set(n=33)at a ratio of 8∶2.A three-dimensional(3D)region of interest(ROI)was delineated in the lesion area from the phaseⅡDCE-MRI images,the mean kurtosis(MK)map,and the mean diffusivity(MD)map,and radiomics features were extracted.Feature reduction and selection were performed using K-best,maximum relevance and minimum redundancy(mRMR),and least absolute shrinkage and selection operator(LASSO)algorithms.Logistic regression(LR)classifiers were used to build the phaseⅡDCE model,DKI parameter map models(MD,MK,MD+MK),and the combined model(DCE-MRI+MD+MK).The stability of the models was validated using five-fold cross-validation.The models'predictive performance was evaluated by receiver operating characteristic(ROC)curve and area under the curve(AUC),and statistical differences between models were analyzed using the DeLong test.Finally,decision curve analysis(DCA)was performed to assess the clinical utility of the radiomics models.Results:A total of 2286 radiomics features were extracted from the 3D ROIs of each sequence.From the PhaseⅡDCE-MRI,MD+MK,MD,MK,and DCE-MRI+MD+MK sequences,8,9,12,7,and 21 features were selected,respectively,that were associated with TNBC.The AUCs of the PhaseⅡDCE-MRI model,MD+MK model,MD model,and MK model in the test set were 0.810,0.769,0.676,and 0.625,respectively.The combined model(DCE-MRI+MD+MK)achieved an AUC of 0.884 in the test set,with an accuracy,sensitivity,and specificity of 78.8%,79.2%,and 77.8%,respectively.Finally,a nomogram model w

关 键 词：乳腺癌 影像组学 动态对比增强 扩散峰度成像 磁共振成像 诊断价值 

分 类 号：R445.2[医药卫生—影像医学与核医学] R737.9[医药卫生—诊断学]