基于混合互信息的医学图像配准

Medical image registration method based on mixed mutual information

通常的互信息测度是基于Shannon熵的,对Renyi熵进行分析,根据某些参数下的Renyi熵可以消除局部极值、而Shannon熵对于局部极值具有很强吸引域的特点,提出一种使用Renyi熵和Shannon熵的混合互信息测度,将两种测度分别用于不同的搜索阶段,首先使用全局搜索算法寻找基于Renyi熵的归一化互信息测度的局部极值,再通过局部优化方法对当前的局部最优解进行局部寻优以找到全局最优解,在局部优化阶段使用基于Shannon熵的归一化互信息测度作为目标函数。实验表明,这种配准算法比单纯使用Shannon熵能够取得更准确的配准结果,而且求解速度得到提高。

Traditionally, the similarity metric is based on Shannon＇s entropy. Through the analysis of Renyi＇s entropy, it is found that Renyi＇s entropy can remove some unwanted local optimum, smooth out difficult optimization terrain accordingly; Shannon＇s entropy has the ＂depth＂ of the basin of attraction, making the registration function easier to be optimized. So a new similarity measure based on mixed mutual information was proposed. The measures based on different entropy were used in different searching phases, and global optimization algorithm and local one were used individually. At first, the global optimization algorithm was used to find the local extrema of generalized mutual information measure based on Renyi＇s entropy. Then, the local one was used to locate the global optimal solution by searching the current local optimal ones, and the generalized mutual information measure based on Shannon＇s entropy was taken as the objective function.