文献速递外周血管头颈技术与安全US

基于影像组学的颈总动脉壁病理模式分类

Radiomics-Based Classification of Pathological Patterns in Common Carotid Artery Wall.
PubMed 文献 · Ultrasound Med Biol · 2026-03-31

本研究基于普通人群队列的784例颈动脉超声图像,提取178个影像组学特征,用逻辑回归和支持向量机区分颈总动脉壁的健康与异常回声模式。逻辑回归在测试集上AUC为0.72、敏感度0.71、特异度0.59、准确率0.64,表明机器学习有助于识别血管壁异常模式。

要点

  • 颈总动脉壁异常回声模式(如三信号模式)与纤维肌性发育不良、高血压及心血管风险相关。
  • 基于40个影像组学特征的逻辑回归模型在测试集AUC为0.72。
  • 视觉判读耗时且依赖操作者,影像组学方法有望辅助标准化识别。

核心结论

  • 机器学习分类器可区分颈总动脉壁的健康与异常回声模式。

研究设计与样本

  • 研究类型为基于普通人群队列的影像组学分类研究,共纳入784例参与者的颈动脉超声数据。
  • 从颈总动脉远壁标准化感兴趣区提取178个影像组学特征。
  • 血管壁模式由医生视觉分类为健康或异常,作为参考标准。
  • 采用80/20训练-测试集划分,训练逻辑回归和支持向量机模型。

关键影像表现/征象

  • 颈总动脉壁异常回声模式包括三信号模式等,原文提及与纤维肌性发育不良、原发性高血压及心血管风险因素相关。

诊断标准与关键数值

  • 逻辑回归基于40个筛选特征,训练集准确率0.70、敏感度0.62、特异度0.76、AUC 0.78。
  • 测试集准确率0.64、敏感度0.71、特异度0.59、AUC 0.72。
  • 健康类与异常类分别占数据的56%和44%。

临床应用与阅片要点

  • 阅片时可关注颈总动脉远壁回声模式是否呈健康或异常(如三信号模式)。
  • 影像组学特征筛选需考虑操作者间可重复性,ROI重叠度增加可提高可重复性。

局限性

  • 原文指出该工具未来可用于更大队列研究颈动脉壁模式的临床相关性。
  • 视觉分类作为参考标准存在操作者依赖性。
  • 本研究为单一人群队列,样本量784例。

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

OBJECTIVES: Abnormal echogenic patterns such as the triple signal pattern can be identified in the common carotid artery (CCA) using ultrasound. These patterns have been associated not only with fibromuscular dysplasia but also with primary hypertension and cardiovascular risk factors, suggesting their potential as markers of vascular aging. Typically identified through visual inspection, their detection is time-consuming and operator-dependent. This study aimed to develop a machine learning approach to identify carotid wall patterns based on carotid-ultrasound image-derived features from a general population cohort.

METHODS: Ultrasound data from 784 participants were analyzed, and 178 radiomic features were extracted from a standardized region of interest on the far wall of the CCA. Vascular wall patterns were visually classified by the physician as healthy or abnormal. Features were filtered based on inter-operator reproducibility, Pearson correlation and feature-relevance. Logistic regression (LR) and support vector machine models were trained using an 80/20 train-test split.

RESULTS: Healthy and abnormal class represented 56% and 44% of the data, respectively. The inter-operator reproducibility increased with greater overlap in ROI placement. Logistic regression achieved accuracy: 0.70, sensitivity: 0.62, specificity: 0.76, and AUC: 0.78 on the training set, and maintained strong performance on the test set (AUC: 0.72, sensitivity: 0.71, specificity 0.59, accuracy: 0.64), based on 40 selected features.

CONCLUSION: These results demonstrate that machine learning classifier can discriminate between healthy and abnormal vascular wall patterns and in the future, this tool may be used to investigate clinical relevance of carotid wall patterns in larger cohorts.

来源:Ultrasound Med Biol,查看原文。版权归原作者所有。

引用

  1. Jadoon M, Poli F, Boutouyrie P, et al. Radiomics-Based Classification of Pathological Patterns in Common Carotid Artery Wall. Ultrasound Med Biol. 2026. PMID: 41916789. doi:10.1016/j.ultrasmedbio.2026.02.017

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。