术前鉴别混合型肝细胞-胆管癌和肝细胞癌的诊断效能:基于临床-超声影像组学的机器学习模型  

Diagnostic efficacy of a clinical-ultrasound radiomics-based machine learning model in preoperative differentiation of combined hepatocellular-cholangiocarcinoma from hepatocellular carcinoma

作　　者：张秀梅 马会会 韩森 纵瑞龙 ZHANG Xiumei;MA Huihui;HAN Sen;ZONG Ruilong(Department of Ultrasound,Xuzhou Central Hospital,Xuzhou 221000,China;Department of Radiology,Xuzhou Central Hospital,Xuzhou 221000,China)

机构地区：[1]徐州市中心医院超声科,江苏徐州221000 [2]徐州市中心医院影像科,江苏徐州221000

出　　处：《分子影像学杂志》2025年第4期509-515,共7页Journal of Molecular Imaging

摘　　要：目的探讨基于临床-超声影像组学特征构建的机器学习模型在术前鉴别混合型肝细胞-胆管癌(cHCC-CC)和肝细胞癌(HCC)中的诊断效能。方法回顾性纳入徐州市中心医院在2010年1月~2024年10月期间经病理证实的42例cHCC-CC患者作为研究对象,按照1∶2的比例采用倾向性评分匹配收集同期病理证实为HCC的84例患者作为对照组。对肿瘤和瘤周区域提取影像组学特征,计算Rad-score。通过单因素和多因素Logistic回归分析筛选与cHCC-CC显著相关的独立危险因素。使用支持向量机(SVM)、随机森林(RF)及逻辑回归(LR)3种机器学习算法构建模型,并选择AUC值最高的模型作为最优模型。按照7∶3的比例将患者随机分为训练集(n=89)和测试集(n=37),通过10折交叉验证方法验证最优模型的性能。结果肿瘤形态、肝硬化病史、CA19-9水平、Rad-score_(tumor)、Rad-score_(10 mm)是鉴别2种肿瘤的独立危险因素。LR模型在3种机器学习模型中表现最佳,曲线下面积为0.883(95%CI:0.826~0.951)。LR模型在训练集、验证集和测试集的曲线下面积分别为0.888(95%CI:0.805~0.971)、0.841(95%CI:0.633~0.994)和0.893(95%CI:0.793~0.992)。校准曲线显示良好的一致性。决策曲线分析显示具有较高的净收益。结论基于临床-超声影像组学的LR模型在鉴别cHCC-CC和HCC中具有术前诊断价值,有助于临床精准诊疗。Objective To explore the diagnostic efficacy of a machine learning model based on clinically-ultrasound radiomics in preoperatively differentiating combined hepatocellular-cholangiocarcinoma(cHCC-CC)from hepatocellular carcinoma(HCC).Methods A retrospective analysis was conducted on 42 patients with pathologically confirmed cHCC-CC in Xuzhou Central Hospital from January 2010 to October 2024.The control group consisted of 84 patients with pathologically confirmed HCC during the same period,selected using propensity score matching at a 1:2 ratio.Radiomic features were extracted from both the tumor and peritumoral regions,and the Rad-score was calculated.Independent risk factors associated with cHCC-CC were identified through univariate and multivariate logistic regression analyses.Three machine learning algorithms,including support vector machine(SVM),random forest(RF),and logistic regression(LR),were employed to develop predictive models.The model with the highest AUC value was selected as the optimal model.Patients were randomly divided into a training set(n=89)and a testing set(n=37)in a 7:3 ratio,and the performance of the best model was validated using the 10-fold cross-validation method.Results Tumor shape,cirrhosis,CA19-9 levels,Rad-score_(tumor),and Rad-score_(10 mm) were identified as independent factors for differentiating the two tumor types.Among the three machine learning models,the LR model demonstrated the best performance,achieving an AUC of 0.883(95%CI:0.826-0.951).The LR model achieved AUCs of 0.888(95%CI:0.805-0.971),0.841(95%CI:0.633-0.994),and 0.893(95%CI:0.793-0.992)on the training,validation,and test sets,respectively.The calibration curve indicated good consistency,and the decision curve analysis revealed a high net benefit.Conclusion The LR model based on clinically-ultrasound radiomics demonstrates significant preoperative diagnostic value in differentiating cHCC-CC from HCC,contributing to precise clinical diagnosis and treatment.

关 键 词：机器学习 混合型肝细胞癌-胆管癌 肝细胞癌 模型 

分 类 号：R735.7[医药卫生—肿瘤]