A Full Bi-Tensor Neural Tractography Algorithm using the Unscented Kalman Filter

Stefan Lienhard¹, James Malcolm², Carl-Frederik Westin³, Yogesh Rathi²

We introduce a tractography method by extending an existing framework which models the signal with Gaussian tensors. At each fiber point an unscented Kalman filter finds the most consistent direction as a mixture of previous estimates and of the local model. In the existing framework the diffusion tensors second and third eigenvalues are identical. We extend the tensor representation so that the diffusion tensor can be an arbitrary ellipsoid. Synthetic experiments show better angular resolution at fiber crossings. Tests on in vivo data show that our new model finds fibers in areas where the simpler model stops.