文献速递腹部MR

肝胆期影像组学识别代偿性慢性肝病亚临床肝功能障碍的开发与外部验证

Hepatobiliary phase radiomics for identifying subclinical liver dysfunction in compensated chronic liver disease: development and external validation.
PubMed 文献 · Abdom Radiol (NY) · 2026-05-04

本研究基于钆塞酸增强MRI肝胆期图像提取全肝影像组学特征,构建模型识别代偿性慢性肝病患者的亚临床肝功能障碍,参考标准整合ALBI评分、MELD-Na和FIB-4。共纳入138人,内部测试集ROC-AUC为0.86,外部验证集ROC-AUC为0.89,提示该影像组学模型有潜力作为无创生物标志物,但验证队列样本量较小,需谨慎解读。

要点

  • 肝胆期影像组学模型在外部验证中ROC-AUC达0.89,可用于识别代偿性CLD的亚临床肝功能障碍。
  • 亚临床肝功能障碍定义为ALBI评分>-2.60、MELD-Na≥10或FIB-4≥2.67的复合标准。
  • 验证队列样本量小,模型稳健性和泛化性仍需多中心大样本验证。

核心结论

  • 基于钆塞酸增强MRI肝胆期图像的影像组学模型在代偿性慢性肝病患者中识别亚临床肝功能障碍表现出良好判别能力,但验证队列样本量有限,需谨慎解读。

研究设计与样本

  • 回顾性研究,纳入Child-Pugh A级且无临床失代偿的代偿性慢性肝病患者,共138人(亚临床肝功能障碍90例,对照48例)。
  • 从钆塞酸增强MRI肝胆期图像提取全肝影像组学特征,采用Boruta特征选择与类加权逻辑回归在嵌套交叉验证框架内开发模型。
  • 亚临床肝功能障碍采用复合参考标准:ALBI评分>-2.60、MELD-Na≥10、FIB-4≥2.67。
  • 内部测试集21例,外部验证集26例,外部验证使用固定决策阈值。

诊断标准与关键数值

  • 亚临床肝功能障碍定义:ALBI评分>-2.60、MELD-Na≥10或FIB-4≥2.67的复合参考标准。
  • 最终选择6个影像组学特征构建模型,包括一阶特征和纹理特征。
  • 内部测试集ROC-AUC为0.86(95%CI:0.66-1.00),准确度为0.81(95%CI:0.62-0.95)。
  • 外部验证集ROC-AUC为0.89(95%CI:0.73-0.99),准确度为0.77(95%CI:0.58-0.92),使用固定决策阈值。

临床应用与阅片要点

  • 在阅读代偿性慢性肝病患者的肝胆期MRI时,可关注全肝影像组学特征提示的亚临床肝功能障碍风险,辅助临床决策。
  • 模型输出可作为无创补充信息,但不应替代ALBI评分、MELD-Na、FIB-4等临床实验室指标的综合评估。
  • 外部验证使用固定决策阈值,提示模型阈值在独立队列中具有可操作性,但需注意验证队列样本量小。

局限性

  • 验证队列样本量相对较小(内部测试集21例,外部验证集26例),结果需谨慎解读。
  • 需在更大规模多中心人群中进一步验证模型的稳健性和泛化性。
  • 回顾性研究设计可能存在选择偏倚。

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

PURPOSE: To develop and externally validate a radiomics model derived from gadoxetic acid-enhanced hepatobiliary phase (HBP) MRI for detecting subclinical liver dysfunction in patients with compensated chronic liver disease (CLD).

METHODS: In this retrospective study, patients were included based on Child-Pugh Class A status and absence of clinical decompensation. Subclinical liver dysfunction was defined using a composite reference standard integrating ALBI score ( > - 2.60), MELD-Na ≥ 10, and FIB-4 ≥ 2.67. Whole-liver radiomic features were extracted from HBP images. Feature selection and model development were performed using Boruta and class-weighted logistic regression within a nested cross-validation framework.

RESULTS: A total of 138 individuals (90 patients with subclinical liver dysfunction and 48 controls) were analyzed. Six radiomic features were selected for model development. In the internal test cohort (n = 21), the model achieved a ROC-AUC of 0.86 [95% CI: 0.66-1.00] with an accuracy of 0.81 [95% CI: 0.62-0.95]. In the external validation cohort (n = 26), the model achieved a ROC-AUC of 0.89 [95% CI: 0.73-0.99] and an accuracy of 0.77 [95% CI: 0.58-0.92] using a fixed decision threshold. Both first-order and texture-based features were represented among the selected predictors.

CONCLUSION: HBP-derived radiomics demonstrates potential as a noninvasive biomarker for identifying subclinical liver dysfunction in compensated CLD patients. While the results are promising, the relatively small validation cohorts warrant cautious interpretation, and further validation in larger multicenter populations is required to confirm robustness and generalizability.

来源:Abdom Radiol (NY),查看原文。版权归原作者所有。

引用

  1. Işık F, Sipahioğlu S, Işıktaş ME, et al. Hepatobiliary phase radiomics for identifying subclinical liver dysfunction in compensated chronic liver disease: development and external validation. Abdom Radiol (NY). 2026. PMID: 42081100. doi:10.1007/s00261-026-05537-7

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。