黎曼流形上的多模态医学图像配准

Multimodal image registration of medical images on Riemannian manifold

针对传统多模态配准方法忽视图像的结构信息和像素间的空间关系,并假定灰度全局一致的前提,提出了一种在黎曼流形上的多模态医学图像配准算法。首先采用线性动态模型捕捉图像高维空间的非线性结构和局部信息;然后通过参数化动态模型构造出一种李群群元,形成黎曼流形,继而将流形嵌入到高维的再生核希尔伯特空间;最后在核空间上学习出相似性测度。仿真和临床数据实验结果表明,该算法在刚体配准和仿射配准精度上均优于传统互信息方法和基于邻域的相似性测度学习方法。

Mutual information based multimodal registration fails to consider the image structure information and spatial relationship among pixels,and assumes that there exists a global statistical relationship between anatomic individuals.This paper presented an algorithm for multimodal image registration of medical images on Riemannian manifold.Firstly,it took advantage of a linear dynamic model(LDM)to capture high-dimensional spatial nonlinear information of the image,then parameterized LDM and constituted Lie group elements,which formed Riemannian manifold.Secondly,it embeded Riemannian manifold to a high-dimensional Hilbert space.Finally,it learnt a similarity measure in Hilbert space.Results from numerical comparative experiments on both synthetic data and clinical data show that compared to the traditional mutual information algorithm and neighborhood based learning similarity measure algorithm,the proposed algorithm obtains better registration accuracy.