Strategies for Data Reorientation during Non-rigid Warps of Diffusion Tensor Images

This paper describes work on the registration of diffusion tensor images of the human brain. An existing registration algorithm, the multi-resolution, elastic matching algorithm, [1-3], has been adapted for this purpose. One problem with the application of such a method to this new data type is that transformations of the image affect the DT values at each voxel, as the orientation can change with respect to the surrounding anatomical structures. Three methods for the estimation of an appropriate reorientation of the data from the local displacement field, which describes the image transformation, are presented and tested. Results indicate that the best matches are obtained from a reorientation strategy that takes into account the effects of local shearing on the data as well as the rigid rotational component of the displacement. The methods presented here may be useful for the computation of region based similarity measures of single valued intensity images, which also vary with local image orientation.