脊椎融合手術（Spinal fusion surgery）是常用的脊椎退化性疾病的手術治療方式，其中植入椎莖螺釘的步驟風險高，需要相當高的定位準確度以及醫師的臨床經驗，避免椎莖螺釘傷害到中樞神經。而於微創手術中無法直接得知椎莖位置，因此需拍攝多張X光影像來確認以手握持的手術器械方位是否正確安全，造成手術時間拉長，病患及醫療人員吸收放射線劑量高。相較於手持手術器械進行手術，使用協作式機械手臂（Collaborative robot）進行手術有較高穩定度，在手術路徑規劃完成後，機器手臂即可自動定位在手術路徑方位上，提高手術效率與成功率，且減少醫療人員的負擔。然而機器手臂的重複準確度雖高而絕對準確度較低，手臂準確度需通過校正後才得以應用在脊椎手術。

本研究使用正交神經網路（Orthogonal neural network）訓練並預測UR5機器手臂在訓練範圍內的任意姿態方位補償值，以校正安裝於機器手臂法蘭面上延伸臂的定位準確度，並探討機器手臂應用於脊椎手術的各種難題。本研究提出兩種測試機器手臂定位誤差的實驗方法，一是於訓練範圍內輸入480筆方位指令給機器手臂當作測試目標方位，量測在校正前延伸臂末端的平均位置誤差為2.48mm，經校正後延伸臂末端的平均位置誤差為0.67mm，定位準確度提升72.9%，此測試方法顯示多筆均勻分布於訓練範圍內的手臂姿態定位誤差。另一方法為手持定位器械置於腰椎模型的椎莖上，將此定位器械的方位當作測試目標方位，在校正前機器手臂上定位器械的平均位置誤差為3.45mm，經校正後定位器械的平均位置誤差為0.56mm，器械定位準確度提升83.7%。此測試方法雖測試目標點少且集中於某特定區域，但較能模擬手術操作的真實情況。兩種誤差測試的實驗結果皆顯示本研究的機器手臂校正方法成功提升機器手臂的定位準確度，使機器手臂可應用於脊椎手術中。

Spinal fusion surgery is a common surgical treatment for degenerative spinal diseases. The step of implanting pedicle screws is so risky that it requires high positioning accuracy and the surgeon's clinical experience to avoid injuring the spinal nerves. In minimally invasive surgery, the position of the pedicle cannot be directly known. Thus, the surgeon has to hold the surgical instrument and continuously take X-ray images to confirm whether the orientation of the instrument is safe, which caused patient and medical staff danger of high radiation exposure and take much more time. Compared with hand-held surgical instruments for surgery, using collaborative robotic arm for surgery has a higher stability. However, the repeatability of the robotic arm is high but the absolute accuracy is low. Thus, the absolute accuracy of the arm must be calibrated before it can be applied under the spine surgery guidance system.

This research uses Orthogonal Neural Network to train and predict the pose compensation value of the UR5 robotic arm within the training range, correct the accuracy of the robotic arm on the extension arm and discuss the various applications of the robotic arm in spinal surgery problem. This research proposes two experimental methods to test the error of the robot arm: input catch test and probe catch test. The results of input catch test show that the average position error of the extension arm before calibration is 2.48mm, and the average position error of the extension arm after calibration is 0.67mm. The robot position accuracy is improved by 72.9%. The results of probe catch test show that the average position error of the probe before calibration is 3.45mm, and the average position error of the probe after calibration is 0.56mm. The robot position accuracy is improved by 83.7%. The experimental results of the two error tests show that the robot arm calibration method in this study has successfully improved the position accuracy of the robot arm, allowing the robot arm to be used in spinal surgery.

[1]. 夏天, “工業機器人運動學標定及誤差分析研究”, 上海交通大學碩士論文, 2009
[2]. 劉振宇, “制約機器人向先進製造系統集成若干問題的研究”, 中國科學院研究生院博士論文, 2002.
[3]. 陳忠昊, “脊椎手術用3D C-arm影像輔助機械臂導引系統之研發”, 中央大學機械工程學系碩士論文, 2020
[4]. 吳吉春, “基於C-arm影像的手術導引定位”, 中央大學機械工程學系碩士論文, 2012
[5]. R.P. Judd, A.B. Knasinski, “A technique to calibrate industrial robots with experimental verification”, IEEE Trans. Robot. Autom. 6, 1990.
[6]. I.W. Park, et al., “Laser-based kinematic calibration of robot manipulator using differential kinematics”, IEEE/ASME Trans. Mechatron. 99, 2011.
[7]. C. Gong, J. Yuan, J. Ni, “Non-geometric error identification and compensation for robotic system by inverse calibration”, Int. J. Mach. Tools Manuf. 40, 2000.
[8]. H.N. Nguyen, J. Zhou, H.J. Kang, “A calibration method for enhancing robot accuracy through integration of an extended Kalman filter algorithm and an artificial neural network”, Neurocomputing 151, 2015.
[9]. C.T. Cao, V.P. Do, B.R. Lee, “A novel indirect calibration approach for robot positioning error compensation based on Neural Network and Hand-Eye Vision”, Applied Sciences vol. 9, 2019.
[10]. L. Qian, C. Mavroidis, “Identification of the end-effector positioning errors of a high accuracy large medical robot using neural networks,” IEEE/RSJ International Conference on Intelligent Robots and Systems vol. 2, 1998.
[11]. 林芃威, “機器人校正與醫學影像導引定位應用”, 中央大學機械工程學系碩士論文, 2003.
[12]. 楊秀雄, “以正交函數為基的適應性網路及其應用”, 中央大學機械工程學系碩士論文, 1993.
[13]. 余介凡, “正交神經網路之建構”, 中央大學機械工程學系碩士論文, 1995.
[14]. Halliday, A., et al., “Dorsal thoracic and lumbar screw fixation and pedicle fixation techniques”, Spine Surgery: Techniques, Complication Avoidance, and Management. New York: Churchill-Livingstone, 1999.