基于力学分解原理的断层图像配准算法

Tomography image co-registration algorithm based on principle of mechanics decomposition

为了通过配准的手段得到信息融合的医学图像,利用力学的分解原理将待配准图像的轮廓信息解释为一系列边界力的合成,将待配准图像的力图投影分量描述成具有角度延迟关系且与平移位置无关的两组周期信号,并采用两组信号间的最小均方误差作为配准的目标函数,利用改进的粒子群优化算法求解得到两幅图像的空间变换参数.通过在颅脑(MR)图像上进行的配准实验表明,在寻优方法一致的情况下采用力学图像的目标函数比采用其他目标函数大大缩短配准时间,且配准误差达到了亚像素级以下.

In order to obtain medical images with fused information by means of co-registration, the frame of the image to be co-registered was explained as composed of a set of forces in the contour based on the principle of mechanics decomposition, and then the mechanics projection weight of the image was described as two rows of periodic signals that have angle delay to each other and have no relations to their location shift. By choosing the minimal mean square error of the two rows of signals as target function, the transform parameters of the two images were obtained through the modified particle swarm optimization algorithm. The co-registration result in brain MR image shows that the mechanics image target function is less time consuming than other target functions when the optimization method is identical and the registration error is limited within one pixel.