Abstract: Mutual information-based medical image registration has the characteristics of high precision and rob-ustness, but the mutual information has some local extremum. Additionally, the method is inefficient when facing to the noise images, which brings great difficulties to the optimization process. In order to solve this problem, a modified differential search algorithm (MDSA) is proposed to optimize the cross cumulative residual entropy (CCRE). The MDSA ameliorates the search method and the iterative conditions, which makes the optimization process more stable and efficient. The MDSA has the advantages of simple control parameters, no dependence on initial point selection, reasonable search direction and boundary control strategy and so on, which make MDSA have very powerful exploration and exploitation capabilities. In the experiment, the original DSA is added to the comparison with MDSA. The experimental results demonstrate the MDSA is suitable for rigid medical image registration in that MDSA not only overcomes the problem of local extrema, but also improves the speed and precision of registration. It is certified that the MDSA is an effective, robust, fast-speed automatic registration algorithm.

Key words: medical image registration, differential search algorithm, cross cumulative residual entropy, computational intelligence

摘要： 基于互信息的医学图像配准具有精度高、鲁棒性强等特点，但互信息存在一定的局部极值，加上面对噪声图像时曲线往往不平滑，给优化过程带来了很大的困难。针对此问题，提出一种改进的差分搜索算法（modified differential search algorithm，MDSA），对交叉累计剩余熵（cross cumulative residual entropy，CCRE）进行寻优。该MDSA对原始差分搜索算法模型的搜索范围和迭代条件进行了改进，使得寻优过程更加稳定、高效。改进后的MDSA具有控制参数简单，不依赖于初始点选择，合理的搜索方向和边界控制策略等优势，有着优秀的全局和局部寻优能力。将该方法应用于医学图像刚体配准，结果证明MDSA相比差分搜索算法，能够有效地克服互信息函数存在的局部极值，提高了配准的成功率，具有较高的配准精度和较快的配准  速度。

关键词: 医学图像配准, 差分搜索算法, 交叉累计剩余熵, 智能计算