Landmark Constrained Non-rigid Image Registration with Anisotropic Tolerances

The registration of medical images containing soft tissue like inner organs, muscles, fat , etc., is challenging due to complex deformations between different image acquisitions. Despite different approaches to get smooth transformations the number of feasible transformations is still huge and ambiguous local image contents may lead to unwanted results. The incorporation of additional user knowledge is a promising way to restrict the number of possible non-rigid transformations and to increase the probability to find a clinically reasonable solution. A small number of pre-operatively and interactively defined landmarks is a straight forward example for such expert knowledge. Typically, when vessels appear in the image data, a natural way is to determine landmarks as vessel branchings. Here, we present a generalization that allows also the usage of corresponding vessel segments. Therefor, we introduce a registration scheme that can handle anisotropic localization uncertainties. The contribution of this work is a consistent modeling of a combined intensity and landmark registration approach as an inequality constrained optimization problem. This guarantees that each reference landmark lies within an error ellipsoid around the corresponding template landmark at the end of the registration process. First results are presented for the registration of preoperative CT images to intra-operative 3D ultrasound data of the liver as an important issue in an intra-operative navigation system.