Improved Local Optimization for Adaptive Bases Non-rigid Image Registration

Non-rigid image registration is an important preprocessing step in many medical image applications. It is a very computation intensive process. Adaptive bases method with local optimization is one approach used to speed up the commonly used free form deformation with B-spline transformation model. In this paper, an improved local optimization for the adaptive bases non-rigid image registration is proposed. By using smaller support size for basis function in the misregistration region identification process, the registration speed is significantly increased. The registration accuracy is also improved by using higher grid point density for local optimization in the identified misregistration region. The proposed algorithm not only preserves the advantage of computing local deformation on disjoint regions which trying to solve the problem globally, but also further reduces the running time. It can achieve notable improvement over the traditional methods. Performance of the proposed algorithm is evaluated using simulated and clinical brain dataset with standard measures such as RMSE and mutual information based similarity measure (NMI). Experimental results show that significant improvements on speed and accuracy are achieved as compared to the conventional approaches.