Abdominal images non-rigid registration using local-affine diffeomorphic demons

Abdominal image non-rigid registration is a particularly challenging task due to the presence of multiple organs, many of which move independently, contributing to independent deformations. Local-affine registration methods can handle multiple independent movements by assigning prior definition of each affine component and its spatial extent which is less suitable for multiple soft-tissue structures as in the abdomen. Instead, we propose to use the local-affine assumption as a prior constraint within the dense deformation field computation. Our method use the dense correspondences field computed using the optical-flow equations to estimate the local-affine transformations that best represent the deformation associated with each voxel with Gaussian regularization to ensure the smoothness of the deformation field. Experimental results from both synthetic and 400 controlled experiments on abdominal CT images and Diffusion Weighted MRI images demonstrate that our method yields a smoother deformation field with superior registration accuracy compared to the demons and diffeomorphic demons algorithms.