Medical image registration

Image registration has been a area of research at the University Medical Center Utrecht for many years. This presentation aims to give a brief overview of the work on image registration, which will be followed by a more in-depth treatment of selected topics. Interpolation of intensities is a vital step during image registration. It has been found that interpolation methods can cause regular patterns of local extrema in the registration search space. These patterns are a result of image noise and interpolation characteristics. The patterns occur when images of identical voxel size are to be registered. A second topic is the conservation of rigid structures in nonrigid image registration. In applications it is usually a problem that the best match found is one that deforms rigid structures. We have developed a regularization term to enforce a rigid transformation locally. The term can be combined with a similarity measure of choice. The method is demonstrated for mutual information based registration with a deformation modelled by B-splines. The final topic is a clinical study on the relation between diabetes mellitus type 2 and so-called white matter lesions (areas of damaged tissue in the white matter of the brain). Such lesions are thought to be correlated with decreased cognitive function. Existing methods use a rough classification of the lesions according to size and location. We have applied nonlinear registration of patient and control data to create a map of lesion distribution throughout the brain for each group. This allows a more detailed investigation of lesion volume and location.