中文摘要：

以互信息为相似性测度,采用B样条变换对多模态医学图像进行非刚性配准时,由于噪声及图像插值等原因造成的互信息局部极值使得传统优化方法不能搜索到最佳配准参数。为此,使用粒子群智能优化方法作为搜索策略,以降低对图像预处理的要求,进一步提高基于互信息的非刚性配准的鲁棒性。为了克服粒子群算法受初始值选取等因素的影响易陷于局部最优的缺点,使用LBFGS优化得到的结果构造初始粒子群,采用多目标优化方法结合交叉变异策略加以改进,使得算法在解空间搜索的遍历性得到改善,优化结果更接近全局最优。MR-T2与MR-PD图像的配准实验证明,该方法提高了基于互信息的B样条非刚性配准的鲁棒性,配准率达到94%;CT与PET图像的配准实验表明该方法相比惯性权重粒子群算法提高了配准精度,互信息增加了0.026;另外,CT与CBCT图像的配准实验也验证了本方法的有效性。

英文摘要：

When mutual information is used as similarity measure in B-spline based non-rigid registration for multimodal medical images, it is difficult to find the best registration parameters using traditional optimization as search strategy due to local maxima of mutual information caused by image noise and interpolation. We proposed to apply particle swarm optimization （PSO） to improve the robustness of non-rigid registration. LBFGS optimization was utilized to make initial particle swarm to avoid local optimum. Then, the multi-objective optimization, combined with cross mutation strategy, was used to achieve global optimum because of the improvement in traversal for solution space. Registration experiments with MR-T2 and MR-PD images demonstrated that the robustness of mutual information based non-rigid registration was improved and the registration ratio was up to 94%. Experiments with CT and PET images indicated that the mutual information increased by 0. 026 over inertia weight PSO. Moreover, experiments with CT and CBCT images also verified the effectiveness of the proposed method.