【作者】 张震；

【导师】 钱晋武；

【作者基本信息】 上海大学 ， 机械制造及其自动化， 2006， 博士

【摘要】 智能化医疗设备和医疗机器人技术研究，是一个多学科交叉的具有重要意义的研究领域。其研究和应用的一个重要方面是无创微创外科手术(MIS)，即利用人体的天生管腔或手术小孔来导入医疗器械或医疗机器人以实现对人体进行诊断、治疗、检测和手术等，从而减少手术期间对人体其他完好组织的伤害，缩短康复时间，并减轻患者的生理痛苦和医疗人员的手术操作时的心理压力。内窥镜诊疗术是微创外科的重要手段，其典型器材是内窥镜。肠道内窥镜是诊疗结肠癌等疾病的重要手段，但目前手工操作存在诸如穿孔等问题。因此，利用机器人和计算机技术对内窥镜诊疗设备进行智能化改造具有重要的研究意义和应用前景。 本论文以克服内窥镜检查穿孔、实现内窥镜智能化为目的，主要研究了智能内窥镜视觉导航技术，研究内容如下： (1)本文针对内窥镜诊疗环境的特殊性和复杂性，通过调研，分析医生在内窥镜检查中的动作和决策方式，在此基础上提出了采用计算机视觉方法引导肠道内窥镜的介入的主要步骤：采集肠道内窥镜头部CCD观测的肠道实时图像并传输到计算机，计算机自动处理分析图像，自动控制和调节肠道内窥镜头部的姿态或给医生适当的建议使内窥镜顺利的进入肠道。 (2)图像底层处理算法是进行计算机视觉导航研究的基础，针对智能内窥镜视觉导航方案，本文从图像信息的采集开始，到图像信息的预处理、图像特征的提取等基本导航技术进行了概述和研究。 (3)在暗区提取方法方面：分析了暗区提取法的内窥镜导航原理，研究了简单阈值算法、基于最大类间方差法的自适应阈值分割算法和基于小波变换的阈值分割算法等三种暗区提取算法，并通过试验对三种算法进行了比较和分析。 (4)在纹理分析方法方面：分析了内窥镜图像暗区的形成过程以及利用暗区进行寻径的不稳定因素，根据肠道内存在闭合肌肉线的情况，研究了采用纹理分析法进行引导内窥镜头部的方法，并且给出了两种纹理分析算法——组织连接算法和边缘能量算法。通过试验，对两种算法进行对比分析。 (5)基于Dempster-Shafer证据推理法的传感器信息融合的导航策略：对暗区提取法和纹理分析法两种方法进行了具体的对比分析，并在此基础上提出了一种基于D-S证据推理法的能对两种导航方法进行信息融合的导航策略。经过试验证明，该导航策略能较好的对两种视觉导航方法进行信息融合。 (6)设计了一种新的智能内窥镜导航系统，该系统采用基于计算机视觉技术的主动避障、有力觉传感器的肠道内窥镜主动介入。更多还原

【Abstract】 The research on intelligent medical devices and medical robots is currently an important area. Among its applications is the minimal invasive surgery (MIS), which has improved levels of traditional diagnosis and treatments. More and more forbidden zones of surgery could be reached and many kinds of diseases could be cured effectively. Endoscopy, an important technology of MIS, mainly uses endoscope as a diagnosis device. While colonoscopic diagnosis has become an effective therapy to cure colon illness, there exist some problems as perforation. Therefore, it is worthwhile to do research on intelligent modification for endoscopes.This dissertation is mainly about the visual navigation technology of the endosope. The main contents of the dissertation are as follows:(1) The particularity and complexity of the endoscopy diagnosis is introduced. By references and interviews with doctors, the manipulation strategy of doctors is analyzed and main steps for vision navigation in endoscopy diagnosis are proposed. The real-time images are picked up by CCD camera and sent to a computer. By analyzing these images, the computer automatically controls the pose of the tip of endoscope or gives suggestions to the doctor who pushes the endoscope into the colon successfully and safely.(2) As a basis for detailed research on visual navigation, a series of techniques are investgated including endoscope images acquisition, pretreatment of endoscope images, feature extraction of image and so on. Several algorithms of these techniques are introduced as well.(3) Dark region analysis: The principle of endoscope navigation based on dark region is introduced firstly. Three algorithms of auto-thresholding are presented which are used to find the path in the colon. These three auto-thresholding algorithms, known as the method of simple thresholding, the auto-thresholding algorithm based on Ostu’s method and the histogram segmentation algorithm based on wavelet, are used to extract the lumen from colon images. The comparison among three algorithms is carried out.(4) Contour analysis: At first, the limitation of dark region method is presented. According to the fact that there are many close muscle lines in colon surfaces, the method of contour analysis is introduced. Two algorithms, based on contour grouping and edge energe detecting respectively, are described to detecting path of the endoscope. The comparison between these algorithms is carried out.(5) Navigation strategy: The application of multi-sensor navigation is introduced更多还原