基于自由形变的3D非线性医学图像配准

Three dimensional nonrigid medical image registration using free form deformation

目的针对北京大学医学物理和工程北京市重点实验室研发的小动物SPECT/CT成像仪,实现三维上的图像配准以校正成像间产生的组织运动伪影。方法把被成像组织的运动用整体的仿射变换和局部的非线性变换两部分进行表达,并用基于体素的归一化互信息作为配准的相似性测度,建立综合图像相似性测度和空间变换函数平滑性要求的代价方程,用近似牛顿算法(LBFGS)在全空间搜索最优化变换参数使得代价方程取得极值,从而确定该相似性测度下的空间变换函数。结果用模拟数据对该方法进行验证的结果表明,配准后图像间的相关系数值有所增加,误配基本被消除。结论采用此方法能够较好地描述局部组织的非线性运动,消除呼吸运动伪影,实现三维图像的配准。

Objective A 3D medical imaging registration method was implemented to eliminate motion artifacts for a dual modality small animal SPECT/CT which developed in Beijing Key Laboratory of Medical Physics Engineering.Methods An affine transformation model was used to describe the global motion while the local motion was fitted by a non-linear B-spline transformation.The voxel-based normalized mutual information was used as the similarity measurement.Then the equation of the cost function was constructed,which combined the similarity measurement and the smoothness requirement of the transformation.Registration was achieved by minimizing the cost function mentioned above using LBFGS algorithm searching for transform parameters in global.Results This algorithm was verified by using simulated 3D data set.The misalignments was removed and the cross correlation between images increased after this registration.Conclusion This nonrigid registration method is useful to eliminate the movement artifacts caused by respiratory.