粒子群与改进的鲍威尔算法相结合的多分辨率三维医学图像配准

Particle swarm combined with improved Powell algorithm of multi-resolution 3D medical image registration

为克服传统基于互信息的多模医学图像配准算法容易陷入局部最优的问题,提出了一种改进的多分辨率三维医学图像配准算法。该算法通过高斯滤波将三维医学图像进行多尺度化,形成多分辨率图像金字塔,以Mattes互信息作为配准框架的相似性测度。在图像金字塔的低分辨率层使用粒子群优化算法进行全局变换参数的搜索,然后以全局变换参数作为高分辨率层配准的初始参数,并以鲍威尔优化算法进行优化,完成最终的三维医学图像配准。实验结果表明,改进的算法不仅使待配准两幅图像空间位置对齐,而且较传统互信息算法提高了配准精度,鲁棒性更强,有效地解决了基于互信息的配准算法陷入局部最优的可能。

In order to overcome the traditional mutual information-based multimodality medical image registration algorithm is easy to drop into local optimal problem,this paper presented an improved multi-resolution three-dimensional medical image registration method. In this method,original images were decomposed into multi-scale image pyramid by Gaussian filter,and Mattes mutual information was used as the similarity measure of the registration framework. It used the particle swarm optimization algorithm to search global transformation parameters in lower resolution layer of image pyramid,then took the global transformation parameters as the initial parameters of high resolution layer registration optimized by Powell optimization algorithm.The experimental results show that the improved registration method not only makes the spatial position of images align,but also is higher precision,better robustness than traditional mutual information method. And it can efficiently solve the possibility of registration algorithm based on mutual information dropping into a local optimal solution.