Anatomical priors for global probabilistic diffusion tractography

Abstract

We investigate the use of anatomical priors in a Bayesian framework for diffusion tractography. We compare priors that utilize different types of information on the white-matter pathways to be reconstructed. This information includes manually labeled paths from a set of training subjects and anatomical segmentation labels obtained from T1-weighted MR images of the same subjects. Our results indicate that the use of prior information increases robustness to end-point ROI size and yields solutions that agree with expert-drawn manual labels, obviating the need for manual intervention on any new test subjects.