Rebooting diffusion MRI uncertainty distributions in the presence of outliers with ROBOOT

Characterizing uncertainty distributions in diffusion MRI derived metrics such as fractional anisotropy or kurtosis anisotropy requires non-parametric approaches, since the correct form of the distribution is rarely known a priori. Previously suggested wild bootstrapping methods, however, have not considered the impact of outliers in the data. In this work, we updated the existing wild bootstrap methodology to consider outliers detected by a robust model estimator, adopting a strategy similar to the rejection of the outliers prior to the model estimate. Additionally, we used simulations based on real human data to demonstrate the benefits of our pipeline for recovering uncertainty distributions.