文献速递神经外周血管US

基于实时可解释深度学习的颈动脉狭窄横向B型超声扫查视频人机交互实现更高准确度分级

Human‒machine interaction based on real-time explainable deep learning for higher accurate grading of carotid stenosis from transverse B-mode scan videos.
PubMed 文献 · Eur J Radiol · 2025-09-19

这项前瞻性研究纳入三家医院疑似颈动脉狭窄≥50%的患者,基于颈动脉横向超声视频训练深度学习模型CaroNet-Dynamic 2.0,以专家诊断为参考标准,并将模型整合到网页图形界面实现可解释性与人工监督。模型分类性能接近高年资放射科医师(所有比较P>0.05),人机交互后低年资与高年资医师的AUROC分别达0.868-0.896和0.875-0.904,显著优于低年资医师单独判读(所有比较P<0.05)。

要点

  • CaroNet-Dynamic 2.0基于颈动脉横向超声视频进行狭窄分类,性能接近高年资放射科医师。
  • 低年资与高年资医师分别对47和37个斑块与AI判断不一致,人机交互中分别采纳38和28个AI诊断。
  • 人机交互后低年资医师AUROC为0.868-0.896,高年资医师为0.875-0.904。

核心结论

  • CaroNet-Dynamic 2.0尝试向放射科医师解释深度学习模型决策所依据的信息,并主动将医师纳入决策环以进一步提升其表现。

研究设计与样本

  • 前瞻性研究,纳入2022年2月至2022年10月三家医院疑似颈动脉狭窄≥50%的患者,共311例(平均年龄71.3岁±8.3;247例男性)。
  • 深度学习模型CaroNet-Dynamic 2.0基于颈动脉横向超声视频训练,以具有15年颈动脉超声评估经验的专家诊断为参考标准。
  • 模型整合到用户友好的网页图形界面以支持人工智能可解释性与人工监督,人机交互策略由5名高年资和5名低年资放射科医师评估。

关键影像表现/征象

  • 原文未报告具体的影像征象描述。

诊断标准与关键数值

  • 以专家诊断为参考标准计算AUROC;CaroNet-Dynamic 2.0分类性能与高年资放射科医师接近(所有比较P>0.05)。
  • 人机交互后低年资医师AUROC为0.868-0.896,高年资医师AUROC为0.875-0.904,均显著优于低年资医师单独判读(所有比较P<0.05)。

临床应用与阅片要点

  • 原文未据此给出推荐。

局限性

  • 原文未报告局限性。

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: We aim to develop an explainable deep learning (DL) model to assist radiologists in carotid stenosis classification by providing understandable or explainable output.

MATERIALS AND METHODS: This prospective study included patients suspected ≥50 % carotid stenosis from three hospitals between February 2022 and October 2022. The DL model CaroNet-Dynamic 2.0 was trained based on carotid transverse ultrasound (US) videos. Model performance was evaluated using expert (with 15 years of experience in carotid US evaluation) diagnoses as the reference standard. Finally, CaroNet-Dynamic 2.0 was integrated into a user-friendly web graphical user interface to support artificial intelligence (AI) explainability and human supervision. The human‒machine interaction strategy was evaluated with five senior and five junior radiologists. Area under the receiver operating characteristic curve (AUROC) were calculated.

RESULTS: A total of 311 patients (mean age ± standard deviation, 71.3 years ± 8.3; 247 men) were included. CaroNet-Dynamic 2.0 showed robust performance in carotid stenosis classification and approached that of senior radiologists (P > 0.05 for all comparisons). Junior and senior radiologists initially disagreed with AI on 47 and 37 plaques, respectively. Using the human‒machine interaction, they adopted AI diagnoses for 38 and 28 plaques, overruling 9 each. The AUROCs of human‒machine interaction achieved 0.868-0.896 and 0.875-0.904 for junior and senior radiologists respectively, substantially outperforming junior radiologists alone (P < 0.05 for all comparisons).

CONCLUSION: CaroNet-Dynamic 2.0 attempted to explain to radiologists the information the DL model used to make decisions and proactively involved them in the decision loop to further improve their performance.

来源:Eur J Radiol,查看原文。版权归原作者所有。

引用

  1. Liu J, Zhou X, Lin H, et al. Human‒machine interaction based on real-time explainable deep learning for higher accurate grading of carotid stenosis from transverse B-mode scan videos. Eur J Radiol. 2025. PMID: 40997634. doi:10.1016/j.ejrad.2025.112441

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。