Several recent studies demonstrate the potential of using tubular structures such as vessels as a basis for image registration. In this paper, we present a novel technique for the deformable registration of tubular structures. Our approach aligns tubular models, e.g. vessels of an organ, with an image by combining both rigid and non-rigid transformations in a hierarchical manner. The physical structure and properties of the vessels are taken into account to drive the registration process. Our model-to-image registration method shows sub-voxel accuracy as well as robustness to noise and a convergence time of less than one minute.
Julien Jomier, Stephen R. Aylward