摘要： 提出了一种新的基于互信息(Mutual Information, MI)的多步骤优化的配准方法。计算输入图像的梯度值，减少了图像的内在信息而使轮廓更为清晰。设计了多步骤的配准框架，优化了配准的收敛过程，使用完整的图像进行有限次的传统配准方法的微调，以实现高精度。为了验证该方法的有效性，分别使用单模、多模和时间序列的方法对临床医学数据进行了实验，与传统的MI配准方法相比，基于互信息的多步骤优化的配准方法具有更高的有效性和精确度。

关键词: 互信息, 医学图像配准, 梯度图像, 多步骤优化

Abstract: This paper proposes an optimized multi-stage registration approach based on mutual information(MI). It calculates the gradient information of an input image as the reference image, which reduces the most inner details of the reference image but emphases its contour information. This pre-processing is proposed to resist the expenses of the normal MI registration. Then it designs a multistage transform in processing framework, which optimizes the convergence during the registration. An adjustment using the traditional MI with two complete images in limited iterations is employed. To demonstrate the effectiveness of this optimized multistage method, three case studies by using mono-, multi-modality and time series clinic datasets in the experiment is implemented. Compared with the common MI method, it is proved to be more efficient in better accuracy.

Key words: Mutual information, Medical image registration, Gradient information, Multistage