文献速递腹部介入MR

基于支持向量机的临床-影像组学MRI模型预测不可切除肝内胆管癌首次经动脉化疗栓塞疗效

Clinical MRI-radiomics model based on support vector machine to predict the efficacy of first transarterial chemoembolization for unresectable intrahepatic cholangiocarcinoma.
PubMed 文献 · Radiol Med · 2025-11-08

本研究回顾性纳入107例以TACE为首治、有增强MRI的不可切除肝内胆管癌患者,按7:3随机分为训练组75例和验证组32例,从动脉期、门静脉期和延迟期提取肿瘤及瘤周5 mm、10 mm的影像组学特征,经随机森林筛选后构建影像组学模型,并结合CA19.9构建临床-影像组学模型。联合瘤周10 mm的影像组学模型表现最佳,临床-影像组学模型在训练组AUC=0.941、验证组AUC=0.903,按模型预测的肿瘤反应分组后两组总生存期差异有统计学意义(P<0.011)。

要点

  • 基于肿瘤+瘤周10 mm的影像组学模型表现最佳。
  • 联合CA19.9的临床-影像组学模型训练组AUC=0.941、验证组AUC=0.903。
  • 按模型预测的肿瘤反应分组,两组总生存期差异有统计学意义(P<0.011)。

核心结论

  • 该模型在预测不可切除肝内胆管癌首次TACE后的肿瘤反应方面表现优异,可帮助临床医生对患者治疗做出更科学的指导。

研究设计与样本

  • 回顾性研究,纳入107例以TACE为首治且具有增强MRI的不可切除肝内胆管癌患者,按7:3随机分为训练组75例和验证组32例。

关键影像表现/征象

  • 影像组学特征从增强MRI的动脉期、门静脉期和延迟期提取,分别针对肿瘤、瘤周5 mm和瘤周10 mm。
  • 基于肿瘤+瘤周10 mm构建的影像组学模型表现最佳。

诊断标准与关键数值

  • 影像组学模型评分与临床变量经单因素和多因素logistic回归分析构建临床-影像组学模型。
  • 临床-影像组学模型联合CA19.9,训练组AUC=0.941,验证组AUC=0.903。

临床应用与阅片要点

  • 模型可辅助临床医生对不可切除肝内胆管癌患者首次TACE后的肿瘤反应进行预测,并为治疗决策提供更科学的指导。

局限性

  • 采用单中心队列,缺乏外部验证。
  • 回顾性设计固有特征导致不可避免的选择偏倚。

以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。

英文原文(摘要)· 点击展开

BACKGROUND: Transarterial chemoembolization (TACE) is an effective treatment for patients with unresectable intrahepatic cholangiocarcinoma (iCCA), but tumor heterogeneity affects the efficacy of treatment. This study aimed to construct a Clinical-Radiomics (CR) model for predicting tumor response after the first TACE in patients with unresectable iCCA.

METHODS: A total of 107 unresectable iCCA patients who received TACE as the first treatment with available contrast-enhanced MRI (CEMRI) were retrospectively enrolled. Patients were randomly assigned to the training (N = 75) and validation cohorts (N = 32) in a 7:3 ratio. Radiomics features were extracted from CEMRI (arterial, portal venous, and delayed phases) for tumor, peritumor 5 mm, and peritumor 10 mm, respectively, and then the features were selected by random forest before constructing the radiomics model. Radiomics model score and clinical variables were analyzed using univariate and multivariate logistic regression to construct the CR model. Kaplan-Meier method was utilized to assess OS.

RESULTS: The radiomics model constructed based on the tumor plus peritumor 10 mm demonstrated the best performance. The CR model developed by combining with CA19.9 showed excellent performance in both the training (AUC = 0.941) and the validation cohorts (AUC = 0.903). Dividing groups based on the model predicted tumor responses, the Kaplan-Meier curves demonstrated a significant difference in OS between the two groups (P < 0.011). The main limitations of this study include the use of a single-center cohort, which lacks external validation, and the inherent characteristics of a retrospective design, leading to an unavoidable selection bias.

CONCLUSION: The model in this study demonstrated excellent performance in predicting tumor response after first TACE in patients with unresectable iCCA. The model could support clinicians to make more scientific guidance on the treatment of patients.

来源:Radiol Med,查看原文。版权归原作者所有。

引用

  1. Zhang Z, Xu A, Zhang Y, et al. Clinical MRI-radiomics model based on support vector machine to predict the efficacy of first transarterial chemoembolization for unresectable intrahepatic cholangiocarcinoma. Radiol Med. 2025. PMID: 41205141. doi:10.1007/s11547-025-02141-5

供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。