论文部分内容阅读
目的:探讨基于智能术前规划的下颌骨重建手术机器人系统的可行性与精准性。方法:术前选取上海交通大学医学院附属第九人民医院就诊的115例头颅无异常的成年患者的CT影像，男57例，女58例，年龄(40.3±9.1)岁，就诊时间2010年2月至2019年5月；另选取115例肿瘤侵蚀下颌骨的成年患者的CT影像，男62例，女53例，年龄(55.6±7.2)岁，就诊时间2008年3月至2019年8月。建立智能颌骨重建手术机器人系统，主要由工作站、UR5六自由度机械臂、光学定位跟踪仪、六自由度力传感器、截骨手术工具组成。将2组患者的颌面CT图像数据用于训练残差连接的3D V-Net分割网络自动分割下颌骨，并用该网络对1例54岁需腓骨重建的下颌骨肿瘤男性患者的头颅CT数据进行自动分割，形成下颌骨模型，以手动分割结果为标准，评价自动分割的精度。使用基于机器学习的颌骨特征点还原法根据该患者的上颌骨特征点自动还原下颌骨特征点，形成重建方案，以下颌骨未缺损部分实际特征点与还原的特征点位置的误差值评价精度。利用该患者的下肢CT数据，制作5个3D打印的腓骨仿真模型。应用智能颌骨重建手术机器人系统，基于术前规划的图像，结合光学定位系统以及力传感器信号，实现机械臂与手术医生协同操作进行腓骨塑形，完成30次截骨。以CT影像下的术后腓骨段截骨面与术前规划的位置距离与角度偏差为精度标准，验证手术机器人系统精度。采用描述性方法进行统计学分析，数据以n ±n s表示。n 结果:设计的分割网络自动分割下颌骨肿瘤患者下颌骨的精度为96.581%，耗时小于30 s。该病例下颌骨特征点的误差值为(2.24±1.74) mm。下颌骨重建手术机器人能精确、快速地实施腓骨模型塑形，截骨的位置误差为(1.02±0.45) mm，角度误差为(0.96±0.42)°，术中耗时约15 min。结论:智能术前规划能精确地分割下颌骨并定位下颌骨特征点，下颌骨重建手术机器人系统能精确地实现功能性下颌骨重建。“,”Objective:To evaluate the performance and the accuracy of surgical robot for mandibular reconstruction based on intelligent surgical planning.Methods:115 CT scanning images of normal mandible (57 males, 48 females, 40.3±9.1 years old, from February 2010 to May 2019) and 115 CT scanning images of mandible with tumor (62 males, 53 females, 55.6±7.2 years old, from March 2008 to August 2019) from Shanghai Ninth People’s Hospital were selected. The surgical robot system including work station, UR robot, optical navigation system, 6 dimensional force senor and surgical instrument. A 3D V-Net for mandible preoperational segmentation from CT scans was proposed and used to segment the mandible of a 54-year-old male patient who received mandible reconstruction with fibular flaps. The machine learning algorithm was used to aid surgical planning for maxillo-mandibular defect by detecting landmarks. The accuracy was defined as the distance between corresponding landmarks on the intact mandible. The robot could locate the target according to surgical planning and perform fibula osteotomy through force-motion control. The CT scanning of limb and head from the patient (male, 54 years old) was used for phantom experiments. 30 osteotomies on 5 3D-priented resin phantom were carried out. The pre- and post-operative images were compared to calculate the accuracy. The descriptive results were in the format of Mean±SD.Results:The average accuracy of V-Net for mandible segmentation was 96.581% and the time cost was less than 30 seconds. The average error of feature points on mandible was (2.24±1.74) mm. The residual length error was (1.02±0.45) mm and angle error was (0.96±0.42) degree in robotic-assisted osteotomy according to 3 cases of phantom experiments. The surgical robot could perform osteotomy safely and steadily within 15 min.Conclusions:Intelligent surgical planning can precisely segment the mandible and determine its landmarks. Robot for mandibular reconstruction can perform fibular osteotomy precisely with the pre-operative planning.