Hybrid elastic registration using constrained free-form deformation

In this paper we present a deformation registration technique that utilizes rigid registration for bony structures while allowing elastic transformation of soft tissue to more accurately register the entire image volume. The technique is applied to the registration of computed tomography images of the thoracic and abdominal regions. First, bony structures are segmented from the CT data. The landmarks are automatically obtained from the segmented data using boundary-detection techniques and random sampling from the boundary candidates. Each bone structure is then individually registered to the corresponding structure using Least Square Method. The whole transformation is obtained by using Thin Plate Spline interpolation. The resulting set of rigid transformations allows for accurate registration of the parts of the CT data representing the vertebrate but not the adjacent soft tissue. To align the soft tissue, a smoothly varying deformation to model different deformability for some structures is computed by means of B-Spline Free Form Deformations algorithm and a normalized mutual information image similarity measure. The advantage of these approaches is that they take into account rigid structures and the deformations applied to the images are continuous and smooth. We applied the proposed methodology to the clinical images. The results have been very positively evaluated by four medical experts.