Automated Registration, Segmentation and Labeling of White Matter Fiber Tracts

Qing Xu¹, Adam W. Anderson¹, John C. Gore¹, Zhaohua Ding¹

Brain white Matter (WM) fibers are widely reconstructed by using magnetic resonance diffusion tensor imaging (DTI). A fiber-to-bundle registration algorithm is proposed to coregister a whole-brain target fiber set with an already reconstructed reference bundle set. Using the resulting transformation, the fibers from different target datasets can all be warped into the reference space for comparison and group analysis. Moreover, their bundle type can be automatically labeled. Experiments with eight human brain DTIs have shown significant similarity of the reference with the post-registered target bundles, and also high consistency of the resulting bundles with a manually segmented ground truth.