基于机器学习的机器人辅助椎弓根植钉规划-中国科学院深圳先进技术.PDF

第 7 卷 第3期 集 成 技 术 Vol. 7 No. 3 2018年5月 JOURNAL OF INTEGRATION TECHNOLOGY May 2018 基于机器学习的机器人辅助椎弓根植钉规划 1,2 1,2 1 苏全健 孙 宇 齐晓志 1 (中国科学院深圳先进技术研究院 深圳 518055) 2 (哈尔滨工业大学深圳研究生院 深圳 518055) 摘  要  为了解决脊柱手术机器人辅助实施椎弓根植钉过程中的图像辅助定位问题，该文提出了一种 基于机器学习策略的椎弓根植钉规划方法。该方法利用卷积神经网络对脊柱计算机断层扫描 (CT) 图像 进行学习和训练，通过建立神经网络模型确定网络内各层间的调整参数，然后对样本图像进行特征提 取并分类，采用交叉验证法对样本数据进行训练，验证卷积神经网络模型的正确性。通过机器学习方 法对计算机断层扫描图像中适合做椎弓根植钉手术规划的图像区域进行识别，从而快速定位到植钉安 全约束区域，并通过相应的图像处理方法实现植钉操作规划。医生只需要基于安全约束区域内的植钉 规划完成最终的手术任务规划，能够显著提升手术效率。 关键词  椎弓根植钉规划；机器学习；图像分类；图像识别；卷积神经网络 中图分类号  TG 156 文献标志码  A Planning of Robot-Assisted Pedicle Screws Insertion Based on Machine Learning SU Quanjian1,2 SUN Yu1,2 QI Xiaozhi1 1( Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences, Shenzhen 518055, China ) 2( Shenzhen Graduate School, Harbin Institutes of Technology, Shenzhen 518055, China ) Abstract In the process of pedicle screw insertion surgery assisted by spinal surgery robot, the surgeon needs to spend a lot of time in searching and determining the location suitable for pedicle screw operation. In order to improve the operation efficiency, a rapid method based on machine learning was investigated to realize efficient planning of robot-assisted pedicle screw insertion. In the proposed method, the convolution neural network that commonly used in artificial intelligence research was introduced based on training set of spine computed tomography images. Firstly, the