基于DCE-MRI影像组学列线图预测乳腺浸润性导管癌PD-L1表达状态的研究  

Study on the prediction of PD-L1 expression in invasive ductal carcinoma of breast with radiomics nomogram based on DCE-MRI

作　　者：张慧 赵楠楠 朱芸[2] 张舒妮 李阳[2] 杨静茹 汤晓敏[2] 杨丽[2] 王玲玲 谢宗玉[2] ZHANG Hui;ZHAO Nan-nan;ZHU Yun;ZAHNG Shu-ni;LI Yang;YANG Jing-ru;TANG Xiao-min;YANG Li;WANG Ling-ling;XIE Zong-yu(School of Medical Imaging,Bengbu Medical College,Bengbu Anhui 233030;Department of Radiology,The First Affiliated Hospital of Bengbu Medical College,Bengbu Anhui 233004,China)

机构地区：[1]蚌埠医学院医学影像学院,安徽蚌埠233030 [2]蚌埠医学院第一附属医院放射科,安徽蚌埠233004

出　　处：《蚌埠医学院学报》2023年第8期1090-1097,共8页Journal of Bengbu Medical College

基　　金：安徽省重点研究与开发计划项目(2022e07020033);安徽省高校自然科学研究重点项目(2022AN051473);蚌埠医学院临床研究专项(2022byflc008);国家级大学生创新训练项目(202210367006)。

摘　　要：目的:探讨基于动态对比增强磁共振成像(DCE-MRI)的影像组学列线图在预测乳腺浸润性导管癌PD-L1表达状态中的价值。方法:回顾性分析蚌埠医学院第一附属医院术前经乳腺MRI检查且经病理证实的91例浸润性导管癌病人,按照6∶4的比例随机分为训练集(n=54)及测试集(n=37)。选择DCE-MRI的第2期在病灶所有层面逐层勾画3D感兴趣区,通过最小绝对收缩和选择算子回归筛选最优影像组学特征,通过支持向量机的方法获取影像组学评分(Rad-score),构建影像组学模型,通过单-多因素logistic回归筛选临床、影像特征作为独立危险因素构建临床模型,最终使用Rad-score联合临床-影像特征危险因素构建影像组学列线图模型。通过受试者工作特征曲线分析各模型的预测效能并计算相应的曲线下面积(AUC),得到最优模型。使用校准曲线、决策曲线评估模型的预测效能及临床实用性。结果:影像组学列线图模型诊断性能最佳,其训练集AUC、敏感度、特异度分别为0.938、81.8%、95.2%;测试集分别为0.904、65.2%、100.0%。结论:基于DCE-MRI影像组学列线图模型在预测乳腺浸润性导管癌PD-L1表达状态具有重要价值,有助于评估病人治疗效果及预后。Objective:To investigate the value of radiomics nomogram based on DCE-MRI in predicting PD-L1 expression in breast invasive ductal carcinoma of breast.Methods:A retrospective analysis was conducted in 91 patients with invasive ductal carcinoma who underwent preoperative breast MRI examination and confirmed by pathology at The First Affiliated Hospital of Bengbu Medical College.The patients were randomly divided into a training set(n=54)and a test set(n=37)at the ratio of 6∶4.The second stage of DCE-MRI was selected to delineate the 3D region of interest(ROI)layer by layer at all levels of lesion.The optimal radiomics features were screened through minimum absolute contraction and regression with the least absolute shrinkage and selection operator,the radiomics score(Rad-score)was obtained by support vector machine to construct the radiomics model,independent risk factors of clinical and imaging features were screened through univariate and multivariate logistic regression to construct a clinical model,and Rad-score combined with risk factors of clinical and imaging features was used to construct a radiomics nomogram model finally.The prediction efficacy of each model was analyzed by the receiver operating characteristic curve,and the corresponding area under the curve(AUC)was calculated to obtain the optimal model.The predictive efficacy and clinical utility of the model were evaluated using the calibration curve and the decision curve.Results:The radiomics nomogram model had the best diagnostic performance.The AUC,sensitivity,and specificity of the nomogram model were 0.938,81.8%,95.2%for the training set and 0.904,65.2%,100.0%for the test set,respectively.Conclusions:The radiomics nomogram model based on DCE-MRI is highly valuable in predicting the PD-L1 expression in invasive ductal carcinoma of breast,and is helpful to evaluate the treatment effect and prognosis of patients.

关 键 词：动态对比增强磁共振成像 影像组学 列线图 乳腺浸润性导管癌 PD-L1 

分 类 号：R445[医药卫生—影像医学与核医学]