Deformable registration of DT-MRI data based on transformation invariant tensor characteristics

Conventional deformable registration methods are mostly driven by the interface between different brain structures. In recent years, Diffusion Tensor Magnetic Resonance Imaging (DT-MRI) has been used in conjunction with other conventional Magnetic Resonance (MR) pulse sequences to provide more structural information during the registration process. Due to the geometric nature of the DT-MRI information, tensors need to be reoriented during warping, which results in a significant increase in algorithmic complexity and computation time. This work is our first attempt to circumvent this problem by using for registration tensors characteristics that are invariant to rigid transformations, instead of the tensors themselves. The invariants used in the experiments reported here are the tensor eigenvalues.