| In clinical practice regularly images of persons are being compared. Since with each imaging session people have a different position in the scanner, it is possible to compare the images only after registration, i.e. that one image is transformed in order to obtain correspondence of the content of the images. If there are anatomic differences in the images (e.g. due to tumour growth or because the images stem from different individuals), an elastic transformation has to be applied, in which the anatomy in one picture is deformed until this corresponds with the anatomy in the other picture. The problem in elastic transformation is mostly not to attain correspondence between two pictures, but to make the underlying deformation realistic. The majority of existing methods tries to prevent unrealistic deformation by forbidding large or sharp deformations. In some cases a physical model of material deformation is used to steer the deformation. In this project we wish to improve elastic registration through model controlled deformation. First these may be physical models. As a second type of model statistical deformation models will be used. To this end a mean deformation and the main variations in deformation will be derived from a collection of sample deformations. This enables to weigh the statistical probability of a deformation in the registration. The method to be developed will be applied in clinical questions like the registration of images of groups of individuals. With such studies differences in e.g. anatomy or disease pattern may be analysed. |
