Alumni

Current research / employment

Medical image registration tries to compute a mapping from one image's frame of reference (FOR) to another's, such that both images are well aligned. Even when the mapping is assumed to be rigid this can be a quite challenging task to accomplish between different image modalities, and is further complicated by the image quality, that is subject to noise and artifacts. In nonrigid image registration these problems are further compounded by the additional degrees of freedom in the transformation. It is in general unclear how nonrigid the transformation for a specific problem should be and whether a resulting transform is correct in a physiological or medical sense.

The aim of my work is to incorporate prior knowledge about morphological correspondences or known patient anatomies into the nonrigid registration, thus constraining the possible deformations. My work is based on a nonrigid, nonparametric image registration approach, for which different formulations and optimization strategies are employed. The main focus however is the mentioned integration of prior knowledge into the registration formulation. One approach explored in this context generates a model for the deformations that can be observed in a set of gold standard registrations. This deformation model then reflects the general morphological variability found in the data. When an unknown dataset is registered the registration algorithm can be prevented from generating results not consistent with the model. This is employed for example in an atlas registration approach for PET-MR attenuation correction.

Figure: From top left to bottom right: atlas CT, patient MR, ground truth patient CT, atlas CT registered to patient MR (pseudo-CT).


Publications D. Hahn, V. Daum, J. Hornegger, Automatic parameter selection for multimodal image registration, IEEE Transactions on Medical Imaging, 29(5), 1140-1155, (2010)

M. Spiegel, D. Hahn, V. Daum, J. Wasza, J. Hornegger, Segmentation of kidneys using a new active shape model generation technique based on non-rigid image registration, Computerized Medical Imaging and Graphics, 33(1), 29-39, (2009)