基于DCE-MRI动力学分析构建的列线图模型预测乳腺癌脉管侵袭  

DCE-MRI kinetic heterogeneity-based nomogram for predicting lymphovascular invasion in breast cancer

作　　者：龚海鹏 张茹 段书峰[1] 邢金丽[1] 朱政锜 冯峰[1] GONG Hai-peng;ZHANG Ru;DUAN Shu-feng(Department of Radiology,Nantong Tumor Hospital,Jiangsu 226361,China)

机构地区：[1]南通市肿瘤医院/南通大学附属肿瘤医院影像科,江苏南通226000 [2]南通市中医院影像科,江苏南通226000

出　　处：《放射学实践》2025年第4期470-477,共8页Radiologic Practice

基　　金：南通市卫生健康委员会科研课题(QN2022029;QNZ2024041)。

摘　　要：目的:探讨基于DCE-MRI动力学分析构建的列线图模型对乳腺癌LVI状态的预测价值。方法:回顾性将2018年1月-2023年6月有明确LVI结果(病理证实)的187例乳腺癌住院患者纳入本研究。其中,LVI阳性组51例,LVI阴性组136例。搜集每例患者的临床危险因素;回顾性判读病灶的MR影像征象;使用MATLAB和SPM软件自DEC-MRI图像上提取病灶的动力学参数:与增强前对比增强第一期肿瘤内增强>50%体素的体积(volume)、增强第一期与增强前肿瘤内最大增强比(peak)、增强最后一期与第一期信号强度比增加大于10%的体素占比(persistent component)、最后一期与早期峰值强化期信号强度比降低大于10%的体素占比(washout component)、增强最后一期与第一期信号强度比增加或降低都不超过10%的体素占比(plateau component)、肿瘤内优势体素类型(predominant)和肿瘤增强异质性(heterogeneity)。采用独立样本t检验、Mann-Whitney U检验及χ^(2)检验分析乳腺癌LVI阳性组与阴性组之间临床危险因素、MRI征象和DCE-MRI动力学参数的差异。分别采用单因素和多因素logistic回归分析筛选乳腺癌LVI状态的独立预测因子,并构建动力学参数模型、临床-MRI特征模型以及动力学参数-临床-MRI特征联合模型,并绘制联合模型的列线图。利用ROC曲线评估各模型对乳腺癌LVI状态的预测效能。采用bootstrap抽样1000次对模型进行内部验证,采用Hosmer-Lemeshow检验对模型的拟合优度进行检验,并采用决策曲线分析(DCA)和临床影响曲线(CIC)评价模型的效能及临床应用价值。结果:多因素logistic回归分析显示peak、heterogeneity、瘤周水肿和ADC值是预测乳腺癌LVI状态的独立危险因素。动力学参数模型及临床-MRI特征模型预测乳腺癌LVI的AUC分别为0.877(95%CI:0.824~0.930)和0.821(95%CI:0.756~0.885);结合动力学参数和临床-MRI特征所构建联合模型具有最高的预测效能,当诊断阈值为0.379Objective:To investigate the predictive value of a nomogram model based on DCE-MRI kinetic heterogeneity for assessing lymphovascular invasion(LVI)status in breast cancer.Methods:This retrospective study included 187 breast cancer patients(51cases of LVI-positive,136 cased of LVI-negative)with pathologically confirmed LVI status from January 2018 to June 2023.Cli-nical risk factors and MR imaging features were collected.DCE-MRI kinetic heterogeneity parameters were extracted using MATLAB and SPM software,including:volume(proportion of tumor voxels with>50%enhancement in the first post-contrast phase compared to pre-contrast),peak(maximum enhancement ratio in the first post-contrast phase),persistent component(proportion of voxels with>10%signal increase in the final phase compared to the first),washout component(proportion of voxels with>10%signal decrease in the final phase compared to the peak),plateau component(proportion of voxels with signal changes≤10%between phases),predominant,and tumor enhancement heterogeneity(heterogeneity).Differences between LVI-positive and LVI-negative groups were analyzed using independent t-test,Mann-Whitney U-test andχ^(2)-test.Univariate and multivariate logistic regression identified independent predictors of LVI.Three models were constructed:kinetic parameter model,clinical-radiological model,and combined model.Nomograms were developed for the combined model.ROC curve analysis was used to evaluate predictive performance,and 1000 bootstrap samples were performed for internal validation.Hosmer-Lemeshow test was used to assess goodness-of-fit,and decision curve analysis(DCA)and clinical impact curve(CIC)were used to evaluated clinical utility.Results:Multivariate logistic regression analysis showed that peak,heterogeneity,peritumoral edema and ADC value were independent risk factors for predicting LVI in breast cancer.The AUCs of kinetic heterogeneity parameter model and clinical radiology characteristic model for predicting LVI in breast cancer were 0.877(95%CI:0.824~0.930)a

关 键 词：乳腺癌 脉管侵袭 动态对比增强 磁共振成像 列线图 

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]