基于增强CT影像组学与机器学习算法预测肿块型肝内胆管癌微血管侵犯

Prediction of microvascular invasion of mass-forming intrahepatic cholangiocarcinoma based on contrast-enhanced CT radiomics and machine learning

目的 研究肿块型肝内胆管癌发生微血管侵犯的影像组学及临床特征,并建立基于机器学习算法的预测模型。方法 回顾性收集2015年1月至2023年2月期间就诊于温州医科大学附属第一医院(75例)和温州医科大学附属第二医院(29例)的肝内胆管癌(ICC)患者资料,提取增强CT的影像组学特征,使用多种机器学习方法进行分析并比较,结合最佳影像组学机器学习方法与临床资料,建立预测模型并进行检验。结果 多种影像组学机器学习方法中,门脉期影像组学特征的朴素贝叶斯分类表现相对较好,曲线下面积(AUC)为0.818,结合筛选出的2个临床特征(瘤内动脉穿行,CEA>5 ng/mL)建立预测模型,训练组和测试组的AUC分别为0.883和0.891,训练组的敏感度为0.978,特异度为0.656,测试组的敏感度为0.909,特异度为0.700。结论 基于增强CT影像组学机器学习结合临床资料的模型可用于预测肝内胆管癌的微血管侵犯状态,具有较好的诊断价值。

Objective To study the radiomics and clinical characteristics of intratumoral microvascular invasion in mass-forming intrahepatic cholangiocarcinoma(ICC)and establish a predictive model with the best machine learning algorithm.Methods We retrospectively collected the data from 75 patients with ICC who were treated at the First Affiliated Hospital of Wenzhou Medical University and 29 patients with ICC who were treated at the Second Affiliated Hospital of Wenzhou Medical University between Jan.2015 and Feb.2023.Radiomic features were extracted from contrast-enhanced CT images,and various machine learning methods were used for analysis.The best radiomics machine learning method was combined with clinical data to establish a predictive model,which was then validated.Results Among various radiomic machine learning methods,naive Bayes classification based on portal venous phase imaging features performed relatively well,the area under the cure(AUC)was 0.818.By combining two selected clinical features(tumor arterial penetration,CEA>5 ng/mL),a predictive model was established with the AUC of 0.883 in the training group and 0.891 in the testing group.In the training group,sensitivity was 0.978,while specificity was 0.656,and in the testing group,sensitivity was 0.909,while specificity was 0.700.Conclusion The model based on radiomic machine learning combined with clinical data from contrast-enhanced CT has good diagnostic value for predicting microvascular invasion status in ICC.