Estimation of uncertainty in constrained spherical deconvolution fiber orientations

Constrained spherical deconvolution (CSD) is a new reconstruction technique that extracts white matter fiber orientations from diffusion weighted MRI data of the brain. However, since these orientations are estimated from noisy data, they are subject to errors, which propagate during fiber tractography. Therefore, it is important to estimate the uncertainty associated with the fiber orientations. In this work, we investigate the performance of a statistical method called the bootstrap, when estimating confidence intervals for CSD fiber orientations. The bootstrap is a nonparametric statistical technique based on data resampling. We used Monte Carlo simulations to measure both its accuracy and precision when applied to CSD. Also, we evaluated an alternative method called the bootknife, which aims to increase the precision of the bootstrap.