Probabilistic fiber tracking using a modified Lasso bootstrap method

Chuyang Ye, Jeffrey Glaister, Jerry L. Prince

Abstract: Diffusion MRI (dMRI) provides a noninvasive tool for investigating white matter tracts. Probabilistic fiber tracking has been proposed to represent the fiber structures as 3D streamlines while taking the uncertainty introduced by noise into account. In this paper, we propose a probabilistic fiber tracking method based on bootstrapping a multi-tensor model with a fixed tensor basis. The fiber orientation (FO) estimation is formulated as a Lasso problem. Then by resampling the residuals calculated using a modified Lasso estimator to create synthetic diffusion signals, a distribution of FOs is estimated. Probabilistic fiber tracking can then be performed by sampling from the FO distribution. Experiments were performed on a digital crossing phantom and brain dMRI for validation.