Improved white matter fiber tracking using stochastic labeling.

Diffusion tensor imaging (DTI) promises a robust means of visualizing in vivo white matter fibers in individual subjects, and of inferring direct connectivity between distant points in the brain. By following the primary eigenvector of the diffusion tensor, trajectories may be defined that trace the path of the underlying fiber tract. However, fiber tracking is prone to cumulative error from acquisition noise and partial volume, which limits the repeatability of such techniques. An image-processing method based on stochastic labeling, by which the noisy primary eigenvectors may be reconfigured according to anatomically reasonable assumptions, is described. The method's potential to improve fiber tracking is first demonstrated on numerical test data. It is then applied to real data acquired from healthy volunteers. Trajectories defined within the corpus callosum and the pyramidal tracts are rendered using 3D graphic imaging software, and the results are compared before and after processing. Fiber tracking was shown to produce anatomically plausible results, and typical errors were largely resolved by the method. Further, the sensitivity of trajectories to their start point was greatly reduced after processing. The use of stochastic labeling may therefore improve the reliability of experiments using white matter fiber tracking.