ICA Based Multi-Fiber Tractography

Manbir Singh¹, Chi-Wah Wong¹

ICA was used to estimate orientations of up to 3 fibers per voxel and conduct streamline whole-brain tractography. Simulation studies suggest an accuracy of about 10 deg for 2 fibers, around 15 deg for 3 fibers, and that ICA is about 100-times faster than multi-tensor compartmental models. Experimental data suggest the residual error between ICA estimated orientations and measured diffusion profiles is 100-times smaller than multi-tensor models. Comparison of ICA and PCA to recover the fornix and cingulum tracts from streamline whole-brain tractography in humans using identical ROIs show better tract continuity and branching with ICA than PCA.