CASIA OpenIR  > 毕业生  > 硕士学位论文
 基于学习和几何变换的2D/3D配准及在椎弓根手术中的应用 陈智强1,2 Subtype 工程硕士 Thesis Advisor 何晖光 2017-05-25 Degree Grantor 中国科学院研究生院 Place of Conferral 北京 Keyword 2d/3d配准 学习 几何 统计形状模型 实时 Abstract 在图像引导的脊椎手术中，实时高效的2D/3D配准是一项重要且具有挑战性的任务。通常的2D/3D配准一般是将三维模型投影到2维平面，然后进行2D-2D的配准。由于投影空间涉及到3个平移以及3个旋转参数，其投影空间的复杂度为O（N6）,采用计算密集的搜索策略，会使得配准很难兼具高准确性和高实时性。 本文提出了一个结合学习与几何变换的2D/3D配准方法。首先使用统计形状模型对目标脊椎进行建模，并构建了一种新的投影方式，使得六个投影参数中的四个可以使用几何的方法计算出来，因而需要学习求解的参数从6个减少到2个，这样很大程度地提高了投影和计算的效率。接下来利用回归学习的方法学习目标脊椎的形状与投影参数之间的关系，避免了计算密集的搜索过程，使得配准的效率得到了较大的提升。最终结合学到的关系和几何变换完成配准。 本方法的两个姿态参数的平均预测误差为 和 ，平均目标配准误差mTRE（mean Target Registration Error）为0.87mm，平均配准时间为0.9s。实验结果表明本方法具有很好的实时性和准确性。 本工作需要对脊椎进行建模，模型的优劣对整个方法的性能有很大的影响，并且建模需要对特征点进行标记，为了减低标记的工作量和更好的建立模型，本工作也同时开发了一个辅助的标记工具，大大提高了标记工作的效率和性能。 Other Abstract In vertebra operation, an effective 2D/3D registration is of great importance and a challenging task. Traditional 2D/3D registration projects 3D data to 2D plan and conducts a 2D-2D registration. It is difficult to obtain both high accuracy and high real-time performance, because it has  projection space complexity composed of 3 translation and 3 rotation degrees of freedom. In this thesis we proposed a method which combined learning strategy and geometric transformation. We built a shape model using statistical shape model and constructed a new projection method, which made it possible that 4 of 6 projection parameters can be calculated by geometric method. Thus, the parameters we need calculate using learning strategy reduced to 2 from 6, which significantly improved efficency of projection and calculation. Then we used regression learning method to learn a pose model between shape of vertebra and projection parameters, which avoided the compute intensive searching step and improved efficiency of calculation. Finally we fulfilled the registration by using the pose model we learned and geometric transformation. Using the proposed method, the mean predict error of pose parameters is  and . The mean Target Registration Error (mTRE) is 0.87mm. And the mean time of registration is 0.9s. The result shows our proposed method owns both high accuracy and high real-time performance. Building the model of vertebra and signing markers, which is of great importance, is needed in this work. To reduce the workload and build the model better, this work developed a tool to help sign markers, which improved the efficiency greatly. Subject Area 医学图像处理与分析 Document Type 学位论文 Identifier http://ir.ia.ac.cn/handle/173211/14698 Collection 毕业生_硕士学位论文 Affiliation 1.中国科学院自动化研究所2.中国科学院大学 Recommended CitationGB/T 7714 陈智强. 基于学习和几何变换的2D/3D配准及在椎弓根手术中的应用[D]. 北京. 中国科学院研究生院,2017.
 Files in This Item: File Name/Size DocType Version Access License 陈智强毕业论文-评阅意见改.pdf（2240KB） 学位论文 暂不开放 CC BY-NC-SA Application Full Text
 Related Services Recommend this item Bookmark Usage statistics Export to Endnote Google Scholar Similar articles in Google Scholar [陈智强]'s Articles Baidu academic Similar articles in Baidu academic [陈智强]'s Articles Bing Scholar Similar articles in Bing Scholar [陈智强]'s Articles Terms of Use No data! Social Bookmark/Share