Orientation Constraints in Point-Based Elastic Image Registration

We present a new approach to incorporate orientation attributes at landmarks in an elastic medical image registration scheme. The approach is based on a minimizing functional and can cope with images of arbitrary dimensions. Constraints due to orientations are incorporated through scalar products between the transformed orientation vectors in the source image and vectors which are perpendicular to the orientation vectors in the target image. For synthetic data sets we show that the structure of rigid objects in an otherwise elastically deformed image can be preserved. The application of our approach to tomographic data sets shows that improvements w.r.t. accuracy can be achieved provided a sufficient number of landmarks is well-distributed over the image.