基于MRI影像组学列线图预测肝外胆管癌程序性死亡配体1表达  

Prediction of Programmed Death Ligand 1 Expression in Extrahepatic Cholangiocarcinoma Based on MRI Radiomics Nomograms

作　　者：刘炯 王小勇 王利敏 黄新荞 杨春梅 舒健 LIU Jiong;WANG Xiaoyong;WANG Limin;HUANG Xinqiao;YANG Chunmei;SHU Jian(Department of Radiology,the Affiliated Hospital of Southwest Medical University,Luzhou 646000,China;Nuclear Medicine and Molecular Imaging Key Laboratory of Sichuan Province,Luzhou 646000,China)

机构地区：[1]西南医科大学附属医院放射科,四川泸州646000 [2]核医学与分子影像四川省重点实验室,四川泸州646000

出　　处：《中国医学影像学杂志》2025年第2期179-185,共7页Chinese Journal of Medical Imaging

基　　金：国家自然科学基金(82272077);四川省科学技术厅重点研发项目(2022YFS0070)。

摘　　要：目的探讨MRI影像组学联合临床特征的列线图术前无创预测肝外胆管癌程序性死亡配体1表达状态的价值。资料与方法回顾性搜集2011年1月—2021年12月西南医科大学附属医院经手术病理确诊的肝外胆管癌87例,以7∶3随机拆分为训练集和测试集。使用3D-Slicer软件在MRI图像上逐层勾画感兴趣区,并提取影像组学特征,然后进行数据标准化、特征降维及筛选。采用高斯朴素贝叶斯构建影像组学模型,同时获取影像组学评分。利用多因素Logistic回归筛选临床特征,并分别构建临床模型和联合模型。采用受试者工作特征曲线下面积(AUC)对3种模型预测效能进行评估,通过校准曲线和决策曲线评估联合模型列线图拟合优度和临床净收益。结果最终筛选出9个影像组学特征和3个临床特征,临床特征分别为谷丙转氨酶(P=0.020)、谷草转氨酶(P=0.025)、总胆红素(P=0.026)。联合模型的预测效能(训练集AUC 0.813,测试集AUC 0.818)优于单独的临床模型(训练集AUC 0.711,测试集AUC 0.705)和影像组学模型(训练集AUC 0.769,测试集AUC 0.767)。校准曲线与决策曲线表明联合模型列线图拟合优度佳,并可取得较好的临床净获益。结论基于术前多序列MRI图像的影像组学评分以及谷丙转氨酶、谷草转氨酶、总胆红素3种临床特征,构建联合模型列线图,能有效预测肝外胆管癌程序性死亡配体1表达状态,为患者精准个性化的免疫治疗提供指导。Purpose To investigate the value of non-invasive preoperative prediction of programmed death ligand 1 expression status in extrahepatic cholangiocarcinoma using MRI radiomics combined with clinical features through nomograms.Materials and Methods A retrospective collection was made of 87 cases of extrahepatic cholangiocarcinoma diagnosed through surgical pathology in the Affiliated Hospital of Southwest Medical University from January 2011 to December 2021.These were randomly divided into training and testing sets at a 7∶3 ratio.Using 3D-Slicer software,regions of interest were manually delineated layer-by-layer on MRI images,and radiomic features were extracted.Data normalization,feature dimensionality reduction and selection were then performed.A Gaussian naive Bayes classifier was used to construct the radiomics model,and radiomics scores were obtained.Multivariate Logistic regression was used to screen clinical features,and individual clinical and combined models were constructed.The predictive performances of the three models were evaluated using the area under the receiver operating characteristic curve(AUC),and the goodness of fit and clinical net benefit of the combined model nomogram were assessed through calibration and decision curves.Results Nine radiomic features and three clinical features were finally selected.The clinical features included alanine transaminase(P=0.020),aspartate transaminase(P=0.025)and total bilirubin(P=0.026).The predictive performance of the combined model(training set AUC 0.813,testing set AUC 0.818)was superior to that of the individual clinical model(training set AUC 0.711,testing set AUC 0.705)and the radiomics model(training set AUC 0.769,testing set AUC 0.767).Calibration and decision curves indicated good fit and better clinical net benefit for the combined model nomogram.Conclusion Based on preoperative multi-sequence MRI images and the radiomics score,along with alanine transaminase,aspartate transaminase and total bilirubin as clinical features,the constructed combi

关 键 词：肝外胆管癌 磁共振成像 程序性死亡配体1 B7-H1抗原 影像组学 列线图表 

分 类 号：R735.8[医药卫生—肿瘤] R445.2[医药卫生—临床医学] R392.1