摘要
旨在研究放疗中图像配准方法,特别是针对放疗中常用的CT、MRI,提出基于混合框架的配准方法,该方法主要包括两个方面:(1)采用掩膜(Mask)提取感兴趣区域、形态学运算等图像处理方法以及CPU多线程并行技术,大幅度提高配准速度;(2)采用由全局到局部的混合配准策略,首先利用基于仿射变换的刚性配准整体配准,以防止图像间偏差过大,在此基础上针对感兴趣区域采用B样条弹性配准,调整局部形变。通过实验表明,采用预处理及加速策略的刚性配准,在保持其精度的情况下,提速比可达10倍,测试结果已达到临床需求;此外,采用基于GPU加速的混合配准策略,其配准速度提至约4 min。
The methods for image registration in radiotherapy are investigated in the study.Aiming at CT and MRI commonly used in radiotherapy,a registration method based on hybrid framework is proposed.In the proposed method,image processing methods such as mask extraction of regions of interest and morphological operations as well as CPU multithreading parallel technology are used to greatly improve the registration speed,and a hybrid strategy of global and local registrations is adopted.Global rigid registration with an affine transformation is used to prevent the deviation between registered images,and then B-spline elastic registration is applied on regions of interest for adjusting local deformations.The experiments show that the preprocessing and acceleration strategy for rigid registration can increase the speed ratio by up to 10 times while maintaining its accuracy,and the test results reach the clinical requirements.In addition,the CT/MRI hybrid registration method based on GPU acceleration can achieve an average registration speed of 4 minutes.
作者
吴茜
皮一飞
周解平
WU Qian;PI Yifei;ZHOU Jieping(School of Humanistic Medicine,Anhui Medical University,Hefei 230032,China;Department of Physics,University of Science and Technology of China,Hefei 230026,China;Department of Radiation Oncology,Affiliated Hospital of University of Science and Technology of China,Hefei 230001,China)
出处
《中国医学物理学杂志》
CSCD
2020年第9期1148-1154,共7页
Chinese Journal of Medical Physics
基金
安徽医科大学博士科研资助基金(XJ201546)
安徽高校自然科学研究项目(KJ2019A0240)。
关键词
多模态图像配准
混合配准
互信息
B样条
预处理
multi-modality image registration
hybrid registration
mutual information
B-spline
preprocessing
