引入高斯函数的互信息法多模态图像配准

Mutual Information-based Multimodal Image Registration Using Gaussian Function

目的:最大互信息作为相似度测量在医学图像配准中已被广泛应用。在计算图像互信息时,为了避免引入新的灰度值一般采用部分体积插值统计两幅图像的联合直方图。但用该方法计算中,当图像平移整数点时,统计联合直方图会出现缺陷,使目标函数出现局部极值,从而造成误配准。方法:将高斯函数引入到直方图统计中,选取适当的邻域,用高斯函数计算邻域内各点像素对联合直方图的贡献。利用高斯函数的平滑性,避免了在互信息计算过程中统计图像联合直方图时出现误差。使用Powell优化方法,寻找最佳的优化参数,实现图像的最佳配准。结果:采用CT-PET数据进行实验,该方法平滑了目标函数,有效地消除了局部极值,提高了多模态图像配准的精确性,并且,对噪音图像配准也产生很好的效果。结论:该方法适用于多模态医学图像配准,克服了传统互信息计算时的不足,提高了配准的正确率和精确度。

Objective：Maximization of mutual information is a popular similarity measure for medical image registration.In the calculation of mutual information image,people often use partial volume interpolation to update the joint histogram for each pixel pair to avoid the introduction of the new gray value.However,this method may cause some local extreme values when the image is translated by integer point,leading to many errors in image registration.Methods：This paper proposed to use the new algorithm which calculated the joint histogram with the Gaussian function.The smoothness of the Gaussian function can avoid statistical errors of the image joint histogram.The best optimization parameters were find using the Powell optimization method.Results：Experimental results which are complete using CT-PET experimental data indicate the proposed algorithm effectively eliminate the local extremum and improve the accuracy of medical image registration.Moreover,this algorithm is also applicable to noise image registration.Conclusions：The method meets multimodal image registration,overcomes the lack of traditional method,improves the accuracy of results.