数据与计算发展前沿 ›› 2026, Vol. 8 ›› Issue (5): 34-47.

CSTR: 32002.14.jfdc.CN10-1649/TP.2026.05.003

doi: 10.11871/jfdc.issn.2096-742X.2026.05.003

• 专刊：“东方”超级计算系统，助力青年科学家科技创新 • 上一篇    下一篇

基于边缘检测的气管插管内窥镜图像气道边界提取

王博1(),徐玉灿2,高继涵3,佘伟光3,梁存霞3,靳冰3,王鑫春3,周艳聪3,孙宝3,*()   

  1. 1 中国科学院力学研究所， 北京 100190
    2 郑州大学附属第一医院麻醉与围术期医学部， 河南 郑州 450000
    3 河南驼人医疗器械集团有限公司， 河南 长垣 453000
  • 收稿日期:2026-05-27 出版日期:2026-10-20 发布日期:2026-10-20
  • 通讯作者: *孙宝（E-mail: sunboy1989@126.com）
  • 作者简介:王博，中国科学院力学研究所，博士，高级工程师。主要研究方向为医学影像技术与无损检测技术。长期深耕图像处理与智能检测算法研究。
    本文主要承担工作为核心算法开发、实验数据采集与分析、论文撰写。
    WANG Bo, Ph.D., is a senior engineer at the Institute of Mechanics, Chinese Academy of Sciences. His research interests mainly focus on medical imaging and non-destructive testing.
    In this paper, he is responsible for core algorithm development, experimental data acquisition and analysis, and manuscript preparation. E-mail: wangbo@imech.ac.cn|孙宝，河南驼人医疗器械集团有限公司，硕士，高级工程师。主要研究方向为医疗器械开发，长期从事医疗器械项目管理与产品研发工作。
    本文主要承担工作为全程参与产品立项、方案设计、落地实施及项目统筹等工作。
    SUN Bao, with a master’s dergree, is a senior engineer at Henan Tuoren Medical Device Group Co., Ltd. His research focuses on medical device development. He is engaged in project management and product development of medical devices, participating in project initiation, scheme design and implementation.
    In this paper, he is responsible for the whole process of product initiation, scheme design, implementation and overall project coordination. E-mail: sunboy1989@126.com
  • 基金资助:
    “东方之星”青年科学家创新基金(DFZX202508)

Edge-Based Airway Boundary Extraction from Endotracheal Endoscopic Images

WANG Bo1(),XU Yucan2,GAO Jihan3,SHE Weiguang3,LIANG Cunxia3,JIN Bing3,WANG Xinchun3,ZHOU Yancong3,SUN Bao3,*()   

  1. 1 Institute of Mechanics, Chinese Academy of Sciences, Beijing 100190, China
    2 Department of Anesthesiology, Perioperative Medicine and Pain Medicine, First Affiliated Hospital of Zhengzhou University, Zhengzhou, Henan 450000, China
    3 Henan Tuoren Medical Device Co., Ltd, Changyuan, Henan 453000, China
  • Received:2026-05-27 Online:2026-10-20 Published:2026-10-20

RichHTML

1

PDF

7

摘要：

【目的】针对胸外科手术过程中双腔气管导管支气管套囊移位可能导致通气功能受损及围术期并发症增加的问题，建立一种基于可视化双腔气管导管的智能位置监测方法，实现支气管套囊位置的连续定量评估。【方法】利用集成于双腔气管导管前端的微型摄像头获取气道图像，通过图像增强预处理提高边界可见性，分别采用多尺度边缘检测方法和U-Net分割模型提取气道边界，并结合基于中轴变换（Medial Axis Transform，MAT）的距离测量算法，对导管与气道壁之间的空间关系进行定量分析。基于实时视频提取图像对两种边界检测方法的性能进行比较，并评估系统的实时运行能力。【结果】多尺度边缘检测方法能够在无需标注训练数据的情况下实现气道边界识别，具有较低的计算复杂度和良好的部署便利性。相比之下，U-Net分割模型获得了更加连续、稳定且准确的边界结果，更适用于后续几何测量与定量分析。经过模型量化后，U-Net模型大小为16 MB，在标准CPU平台上可实现4 FPS的实时处理速度。【结论】多尺度边缘检测方法适用于低计算资源条件下的快速边界提取，而U-Net分割模型在边界精度、鲁棒性及定量分析能力方面具有明显优势。结合MAT距离测量算法的U-Net框架能够为胸外科手术中支气管套囊位置的实时监测和气道几何定量评估提供可靠技术方案。

关键词: 可视双腔气管导管, 深度学习, 图像分析

Abstract:

[Objective] This study aims to develop a vision-based method for continuous and quantitative monitoring of bronchial cuff position in double-lumen endotracheal tubes during thoracic surgery, where cuff displacement may compromise ventilation and increase perioperative complications. [Methods] Airway images acquired from an integrated camera mounted on a visualized double-lumen endotracheal tube are processed using image enhancement techniques to improve boundary visibility. Airway boundaries are extracted using both a multiscale edge detection method and a U-Net segmentation model. The spatial relationship between the tube and airway wall is quantitatively evaluated using a medial axis transform (MAT)-based distance measurement algorithm. The performance of the two boundary extraction methods and the real-time capability of the system are experimentally evaluated using images extracted from live video. [Results] The multiscale edge detection method successfully identifies airway boundaries without requiring annotated training data and exhibits low computational cost. In contrast, the U-Net model produces more continuous, robust, and accurate boundary representations, providing superior performance for geometric measurement and quantitative analysis. After model quantization, the U-Net model achieves a compact size of 16 MB and a processing speed of 4 FPS on a standard CPU platform. [Conclusions] Although multiscale edge detection offer advantages in computational efficiency and ease of deployment, the U-Net-based framework demonstrates superior boundary extraction accuracy, robustness, and quantitative analysis capability. Combined with MAT-based distance measurement, the proposed method provides a reliable solution for real-time bronchial cuff position monitoring and quantitative airway assessment during thoracic surgery.

Key words: visualized double-lumen endotracheal tube, deep learning, image analysis