摘要
针对互信息仅考虑两幅图像相应像素的灰度信息以及B样条变换模型存在形变场奇异点的问题,提出一种基于P样条和局部互信息的非刚性医学图像配准方法。该方法以局部互信息为相似性测度,采用P样条变换模型模拟待配准图像的几何形变,使用三次插值算法对图像像素进行赋值,结合对大规模参数优化效率高的LBFGS算法优化配准参数。较传统的B样条变换模型和互信息,提出的方法除计算时间外,其他三项指标更优,均方误差下降了89.25%,归一化互信息提高了11.04%,相关系数提高了5.64%。实验结果表明,该方法有效地提高了配准的精度。
To overcome the drawbacks which mutual information took into account only the relationships between correspon- ding individual pixels and the B-spline transformation model existed the singularities in the deformation field, this paper put forward an nonrigid medical image registration method based on P-spline and regional mutual information. The method took re- gional mutual information as similarity measure, adopted P-spline transformation model to simulate geometric deformation of registration image, used cubic interpolation to assign image pixels, and employed LBFGS algorithm which had high efficiency in large scale parameters optimization to optimize the registration parameters. Comparing with mutual information and B-spline transformation model, the proposed method was better for the three indexes except for computation times, the mean square er- ror decreased by 89.25%, the normalized mutual information rose by 11.04% and the correlation coefficient increased by 5.64%. Experimental results show that the proposed scheme effectively improves the accuracy of registration.
出处
《计算机应用研究》
CSCD
北大核心
2017年第8期2538-2541,共4页
Application Research of Computers
基金
山西省自然科学基金基础研究项目(2013011017-3)
关键词
非刚性配准
医学图像
P样条
局部互信息
LBFGS优化算法
nonrigid registration
medical image
P-spline
regional mutual information
LBFGS optimization algorithm