A parallel algorithm for medical images registration based on B-splines

Cubic B-splines is widely applied in non-rigid registration because of its approximation performance and fast computational characteristics. However, a small scale non-rigid deformation is needed to characterize by a large number of control points. Moreover, an iterative optimization strategy of the non-rigid registration algorithm and the normalized mutual information (NMI) cost a great quantity calculation. So, the process of the non-rigid registration is slowed by calculations of NMI in a iterative optimization strategy. In this paper, a parallel optimization algorithm based on cubic B-splines functions is proposed to parallelize the optimization algorithm of the nonrigid registration and the calculations of normalize mutual information. In practice, a fast algorithm of cubic B-splines is used and the control points are only distributed on the targets. Experiments show that the use of the fast algorithm and the parallel optimization strategy improves the non-rigid registration process of medical images.