Statistical Deformation Model Based Non-Rigid Multimodal Medical Image Registration

There may exist the complex non-rigid deformation among multimodal medical images. To correct such deformations, the nonlinear transformation models with a high degree of freedom must be used. Solving the high-dimensional parameters of the nonlinear transformation directly will not only increase registration time but also affect registration accuracy. To solve this problem, a registration method was proposed based on statistical deformation model in this paper. Firstly, a statistical deformation model was established to statistically learn the non-rigid deformation among a large number of multimodal images, and to greatly reduce the number of parameters in the transformation model, to improve image registration efficiency and accuracy. Experimental results show that, compared with the registration method based on traditional free-form deformation model, the efficiency of the proposed statistical deformation model based registration method can be improved by 52%, and the target registration error can be reduced by 0.503 2 pixels.