基于点集与互信息的肺部CT图像三维弹性配准算法

3D deformable registration algorithm of CT lung images based on point set and mutual information

针对基于互信息的肺部CT图像三维弹性配准算法精度低、耗时多的问题,提出了一种基于混合配准框架的配准算法.该方法将点集与互信息相结合,先采用点集配准算法获得点集位移向量,再求出变换函数,最后以基于互信息的方法进行细化配准.采用4组三维肺部CT图像,以标志点距离误差来验证算法精度.试验结果表明,所提方法能够快速精确地完成三维肺部CT图像弹性配准;相对于仅基于互信息的配准算法,耗时平均减少70%,配准精度平均提高5%.该方法可以用于4D肺CT图像的快速配准.

To solve the disadvantages of time-consuming and low accuracy in 3D deformable registration algorithm of CT lung images based on mutual information,a new registration method was proposed based on hybrid framework.The mutual information registration was combined with point set registration in the algorithm.Point set registration was performed to obtain the point set displacement vectors,and the transformation function was determined.The resulted transformation function was refined based on mutual information.The registration algorithm for lung regions was performed on 4 sets of 3D CT images,and the registration accuracy was evaluated by landmarks.The results of 3D CT lung images registration show that the hybrid registration algorithm is superior to the mutual information registration method with 70% time reduction and 5% registration accuracy improving on average.The proposed method is fit for 4D CT lung images rapid registration.