This paper describes the implementation of a resampling filter for Diffusion Tensor Images (DTI) in the Insight ToolKit (ITK). ITK already contains a filter for resampling scalar and vector images as well as several transformation and interpolation classes. However, due to the directional nature of DT images, using the existing classes would result in losing the structural information of the image.
We developed a new resampling filter, specific to DTI, that preserves the structure by applying a rotation directly on the tensors while performing the transformation of the image. New transformation and interpolator classes have also been implemented to handle tensors correctly. The new transformation classes are based on algorithms described by D.C. Alexander et al. Finally, three filters have been written to correct symmetric semi-definite matrices that would no longer be positive after the resampling process and project them into the tensors’ space. In addition, a software based on the new classes has been developed and is provided with this article.