基于药代动力学对比增强MRI影像组学模型列线图预测乳腺癌腋窝淋巴结转移的价值  

Predicting axillary lymph node metastasis in breast cancer using pharmacokinetic dynamic contrast-enhanced MRI radiomics nomogram

作　　者：苏迪 王琪[2] 高婧 张忠胜 张海程 初同朋 毛宁[2] 谢海柱[2] SU Di;WANG Qi;GAO Jing;ZHANG Zhongsheng;ZHANG Haicheng;CHU Tongpeng;MAO Ning;XIE Haizhu(School of Medical Imaging,Binzhou Medical University,Yantai 264003,China;Department of Imaging,Yantai Yuhuangding Hospital,Yantai 264000,China)

机构地区：[1]滨州医学院医学影像学院,山东烟台264003 [2]烟台毓璜顶医院影像科,山东烟台264000

出　　处：《中国中西医结合影像学杂志》2025年第2期216-222,共7页Chinese Imaging Journal of Integrated Traditional and Western Medicine

基　　金：国家自然科学基金(82371993);山东省自然科学基金(ZR2021MH120);泰山学者青年专家计划(tsqn202211378);烟台市科技创新计划(2023YD015)。

摘　　要：目的:探究基于药代动力学对比增强(Pk-DCE)MRI影像组学模型列线图在乳腺癌腋窝淋巴结(ALN)转移预测中的价值。方法:回顾性收集术前行动态对比增强MRI(DCE-MRI)的乳腺癌患者591例,按照9∶1的比例随机分为训练集531例和测试集60例。将DCE-MRI图像导入定量分析软件获取容量转移常数(K^(trans))、流出速率常数(K_(ep))、血管外细胞间隙体积分数(V_(e))、血浆容积分数(V_(p))参数图。用ITK-SNAP软件分别在DCE-MRI原始图和参数图上勾画ROI,并用Pyradiomics提取特征。通过方差阈值、Select-K Best、最小绝对收缩和选择算子(LASSO)算法筛选特征并降维,通过logistic回归分析建立影像组学模型并计算模型的影像组学评分(Radsocre)。利用单因素和多因素logistic回归分析筛选差异有统计学意义的临床特征和Radsocre建立联合模型,并绘制列线图。应用ROC曲线评估模型的预测效能,用决策曲线分析(DCA)和校准曲线评估模型的一致性。通过De Long检验比较临床特征模型、影像组学模型及联合模型诊断效能的差异。结果:利用肿瘤直径、Radscore DCE、Radscore V_(e)、Radscore V_(p)建立联合模型,其在ALN转移预测中表现较好。联合模型在训练集中AUC为0.877(95%CI 0.848~0.906),敏感度为0.826(95%CI 0.779~0.866),特异度为0.723(95%CI 0.656~0.782);在测试集中AUC为0.889(95%CI 0.800~0.978),敏感度为0.850(95%CI 0.695~0.938),特异度为0.889(95%CI 0.639~0.981)。联合模型预测效能优于临床特征模型和影像组学模型。DCA显示,联合模型有显著净效益,具有较高的应用价值。联合模型的校准曲线一致性较好。结论:基于Pk-DCE-MRI影像组学模型列线图可用于术前预测乳腺癌ALN转移,为乳腺癌的诊疗提供新的有效工具。Objective:To develop a pharmacokinetic dynamic contrast-enhanced MRI(Pk-DCE-MRI)radiomics model nomogram for preoperative prediction of axillary lymph node(ALN)metastasis in breast cancer.Methods:This retrospective study analyzed 591 breast cancer patients who underwent preoperative DCE-MRI.Patients were randomly divided into training(531 cases)and validation(60 cases)cohorts at a 9∶1 ratio.K^(trans),K_(ep),V_(e),V_(p) parametric maps were generated using quantitative analysis software.ROIs were delineated on DCE-MRI images and parametric maps using ITK-SNAP,with radiomics features extracted via Pyradiomics.Features were selected and dimensionality reduced through variance,Select-K Best,and LASSO algorithm.A radiomics model was constructed using logistic regression to calculate radiomics score(Radscore).Clinical predictors were integrated through univariate and multivariate logistic regression to establish a combined model.Model performance was evaluated using ROC curve analysis,decision curve analysis(DCA),and calibration curves.The DeLong test was used to compare the differences between the clinical features model,the radiomics model and the combined model.Results:The combined model incorporated tumor diameter,Radscore DCE,Radscore V_(e),and Radscore V_(p) had a good performance for ALN metastasis.In the training cohort,it achieved an AUC of 0.877(95%CI 0.848—0.906),with a sensitivity of 0.826(95%CI 0.779—0.866)and a specificity of 0.723(95%CI 0.656—0.782);while in the validation cohort,it achieved an AUC of 0.889(95%CI 0.800—0.978),with a sensitivity of 0.850(95%CI 0.695—0.938)and a specificity of 0.889(95%CI 0.639—0.981).The diagnostic efficiency of the combined model was better than both the clinical feature model and the radiomics model.DCA revealed that the combined model had a high net benefit and significant clinical application value.The calibration curve showed the combined model had a good consistency.Conclusions:The nomogram based on Pk-DCE-MRI radiomics can be used for preoperative pre

关 键 词：影像组学 列线图 乳腺肿瘤 腋窝淋巴结转移 磁共振成像 

分 类 号：R73[医药卫生—肿瘤]