基于特征点互信息预配准的医学图像配准技术

Medical Image Registration Based on Feature Point Pre-registration Strategy

基于最大互信息的医学图像配准算法近几年来成为医学图像处理领域的热点。由局部极值导致的寻优困难是困扰该算法的核心问题,混合优化算法成功地解决了互信息函数的寻优问题,但延长了配准时间。文中研究了互信息函数峰值周围的局部极值特征,提出安全区域的概念。利用特征点互信息理论,并结合多灰度级和多分辨率策略,提出一种基于混合优化算法寻优和特征点互信息预配准的改进型算法。经过模拟数据和实际数据配准实验证明,该算法在保证了配准精度的同时,提高了配准的速度,稳健性更强,具有临床推广价值。

The registration algorithm based on mutual information in medical image processing community has become a hot spot. The difficulty of optimization caused by local extremum is the key problem of the algorithm. It can be solved by hybrid optimization algorithm successfully. However, time consumption is increased. In this paper, the feature of local extremum around the globe extremum is analyzed and the concept of security region is proposed for the theoretical basis of pre-registration. Combining feature point mutual information theory with strategies of multiple gray levels and resolutions, a kind of improved registration algorithm based on hybrid optimization algorithm and feature point mutual information pre-registration is developed. Experimental results show the validation, high precision and excellent accelerating capability of the algorithm.