Pew-Thian Yap1, John H. Gilmore2, Weili Lin1, Dinggang Shen1
1Radiology & BRIC, University of North Carolina, Chapel Hill, NC, United States; 2Psychiatry, University of North Carolina, Chapel Hill, NC, United States
Tractography in an atlas space allow the reconstructed trajectories to form a common geometry onto which diffusion properties from the individual images can be projected for tract-based comparison. We show, however, that conventional average-atlas-based approach, when applied to HARDI data, causes significant deviation of the estimated local orientations from the 'true' orientations, inevitably jeopardizing subsequent trajectory reconstruction. In our approach, local fiber orientation information is estimated by harnessing orientation information simultaneously from all images in a population. We model the orientation statistics at each voxel location by employing the bipolar Watson distribution, which will capture the mean orientations of the fiber bundles and also the related degrees of orientation dispersion. This distribution information, when fed into a stochastic tractography algorithm allows reconstruction of fiber trajectories which are consistent across images in the population.