Collapsed-cone based deformation field regularization for nonrigid image registration

Incorporating biomedical information into nonrigid image registration is an important approach to improve the registration quality and provide realistic results. However, previous tissue-dependent deformation field filtering incur a relatively high computation cost in order to obtain results of improved quality. In this paper, we propose a collapsed-cone based adaptive filtering method to reduce the computational overhead of regularization. The filter is designed to change its filtering parameters dynamically at each voxel according to the tissue characteristics and the deformation of the surrounding voxels. The proposed filter is integrated into the demons deformable registration method to evaluate its effectiveness and performance. The evaluation is performed on a set of 3D computed tomography (CT) images and the result quality is compared with the output of those without applying the tissue-dependent filter. The results show that our proposed method can preserve the global features better. Based on the measure of sum of squared differences (SSD), the proposed method is also found converging faster and leading to lower SSD.