Diffusion Tensor Image Registration Using Uncertainty Information

Introduction: Population and longitudinal analyses using Diffusion Tensor Imaging (DTI) data have become feasible over the past decade with advanced sequences and sophisticated mathematical tools. These studies make use of some form of elastic tensor field registration framework to derive a population average brain and the deviation modes. These registration algorithms need to employ a specialized tensor similarity metric [1,2] and an tensor interpolation method [3]. Previously proposed metrics directly use tensor-derived information and disregard the diffusion weighted data once tensor fitting has been performed. In this work, we propose a new, analytical tensor similarity metric that not only inherently considers tensor directionality and shape but that also uses the entire experimental design information to infer the uncertainty in the computed tensors.