【作者】 张婧；

【导师】 刘允才；

【作者基本信息】 上海交通大学 ， 模式识别与智能系统， 2007， 博士

【摘要】 随着医学理论的不断进步与发展,外科手术也正朝着更加精细和复杂的方向发展。近几十年来高速发展的计算机、机器人、电子信息以及网络通信等技术已经越来越多地应用于医学领域。“机器人辅助外科手术”(RAS, Robot Aided Surgery)系统在现代临床医疗中已经被越来越多地采用。每年成千上万患有膝关节疾病的病人接受全膝关节置换手术(TKR, Total Knee Replacement),期望在一定程度上恢复行动,减少痛苦。目前,手术中假体位置主要由临床经验和专用模板保证,是误差主要来源。针对基于传统机械导航模板的人工膝关节手术的诸多缺陷,计算机或机器人辅助全膝关节置换手术应运而生。相比人工膝关节置换手术,采用计算机与机器人进行辅助手术,并使用专家系统进行受力、运动的评估,获取膝关节假体的最优放置位置,可以使机器人辅助膝关节手术具有更好的操作精度,提高膝关节置换手术的质量与成功率。传统机器人辅助全膝关节置换手术系统中存在的初始定位难,以及病人必须接受两次手术等诸多缺点一直是人们的研究热点之一。本文基于机器人集成手术系统理念,针对传统全膝关节置换手术的缺点,提出了手眼式机器人辅助外科手术(HERAS, Hand-Eye Robot Aided Surgery)模型。此模型将摄像机与刀具固定在机器人末端执行器上,利用相机标定与手眼标定技术,构建动态导航系统,能克服静态手术导航中视角不能轻易改变的缺点;应用于全膝关节置换术,根据红外探针获取的膝关节上生理标志点的位置,能得到术中切割平面的准确位置及精确的下肢对线,提高手术精度。本论文的工作包含四个方面:一是设计并实现了符合临床手术要求的基于HERAS模型的准临床手术实验系统,二是HERAS中的在线手眼标定问题,三是HERAS中的多体运动分割问题,最后是HERAS模型在膝关节手术各种实验中的具体应用与分析。综观全文,本论文的主要创新性研究成果包括以下内容:1)设计并实现了符合临床手术要求的基于HERAS模型的准临床手术实验系统。此系统根据手术要求,利用膝关节的生理特点,可进行高精度的股骨定位、胫骨定位及手术切割,建立精确的下肢对线。满足了手术安全、环境、消毒等临床医学要求,使HERAS模型在临床手术中的应用前景更加明朗;2)针对在线手眼标定中退化运动和噪音对标定精度的影响,提出了根据运动序列自身特点,自适应确定阈值的运动选择算法,提高了在线手眼标定算法的工程实用性,为手眼式机器人在手术实施过程中的安全可靠使用提供了重要保障。3)手眼式机器人进行辅助外科手术过程中,需要同时使用多个辅助定位工具并进行多目标运动跟踪。在当前视觉伺服与跟踪技术的基础上,提出了基于多体三焦点张量与直线光流的两种多体运动分割算法。利用直线特征对应进行计算,可解决使用点对应时出现的特征遮挡问题,丰富了计算机视觉领域中多体运动分割技术的理论和方法。4)利用基于HERAS模型的机器人辅助全膝关节置换术实验系统-WATO,进行了假骨模型试验、动物尸骨试验、尸体骨实验以及准临床手术实验(尸体实验),进行了详细的精度分析,对各阶段手术实验中遇到的问题提出了解决方法,为临床手术积累了数据与经验。更多还原

【Abstract】 With the rapid developments of medical theories, surgery operations have become more elaborate and complicated. State-of-art technologies, such as computer, robotics, electronic information and network communication, have been widely applied in medical areas. As a result, robot aided surgery (RAS) is increasingly adopted in modern clinic operations.Every year, thousands of patients suffered from joint diseases, such as rheumatoid arthritis or osteoarthritis, needing total knee replacement (TKR) surgery to recover their normal functions. Presently, the positioning of prosthetic components in surgeries mainly depends on clinic experiences of doctors and special surgical guiding devices. To avoid the limitations of jig-based TKR systems, robot/computer assisted surgeries are quickly developed, with the aid of which a better operation precision and surgical quality are well expected.However, the classical robot aided surgical systems of TKR has limitations in the operational precisions and twice surgeries. According to the characteristics of the TKR surgery, we designed a hand-eye robot aided surgical system and made it applicable to clinical trials. In this model, both the cameras and the cutting tool are fixed on the end-effecter of the robot. In this way, we get a dynamic navigation in stead of an inadequate static one. In total knee replacement surgeries, the information from fiducial marks helps the surgical robot to automatically determine the position of cutting planes. Using this method, an accurate mechanical axis is established.The major contributions of this thesis includes four pars: first, base on HERAS model, we designed and realize a pre-clinical experiment system, which can fulfill the requirement of clinical applications; second, we resolve the problem of online hand-eye calibration; third, we proposed new methods for motion segmentation; and at last, we applied HERAS model to TKR cadaver trials and resolve many problems in these experiments. Detailed descriptions of these contributions are as follows: 1) Design and realize a pre-clinical experiment system base on HERAS model. In this model, using the information of fiducial marks on the knee, one can establish a femur, tibia coordinate system and get an accurate mechanical axis. The system fulfills the requirements of cadaver trials and is very meaningful toward clinic applications.2) In order to guarantee the safety of surgeries, online hand-eye calibration will never be overlooked. After analyzing the traditional method of offline hand-eye calibration and online hand-eye calibration, we proposed a new concept of adaptive motion selection for online hand-eye calibration, which will not only avoid the degenerate cases, but also avoid small rotations that will lead large error in calibration.3) In the model of HERAS for TKR, we usually use several assistant guides simultaneously to track multiple motions. Based on the previous works of visual tracking and multibody motion segmentation, we proposed two new methods of segmenting multiple 3D motions from line correspondences--one is based on multibody trifocal tensor, while the other one is based on line optical flow.4) Moreover, we employed different materials, such as phantoms, animal bones, human bones and cadaver. We resolved practical problems in these experiments and made accuracy analysis. The experiences and technologies will be very helpful for the clinical surgery in the future.更多还原