Transformation of and similarity measures for diffusion tensor MRIs

This paper discusses registration of diffusion tensor (DT) MRIs (magnetic resonance images). An existing method for registration of 3D intensity images, the multi-resolution elastic matching algorithm, [1-3], is adapted to work with this new data type. A method is described for extracting the appropriate reorientation of DTs to accompany warps of the image from the local displacement field. Furthermore, the effectiveness of a number of similarity measures for matching DT images are tested. Preliminary results over a single pair of DT images of the human brain indicate that the best results are obtained by use of a tensor component difference measure.