Model-Based Residual Bootstrap of Constrained Spherical Deconvolution for Probabilistic Segmentation and Tractography

Hamied Ahmad Haroon¹, David M. Morris¹, Karl V. Embleton^1,2, Geoff J. Parker¹

Here we describe the application of model-based residual bootstrapping to the analysis of HARDI data using constrained spherical deconvolution. We demonstrate that the method is able to provide estimates of the probability of finding different fiber configurations within the brain. These distributions of fiber orientations may then be used directly as PDFs across each configuration for probabilistic tractography. This method provides a means by which the microstructural complexity of tissue, as reflected in the HARDI diffusion signal, may be characterised, naturally accounting for the underlying tissue microscopic complexity, macroscopic partial volume, and data noise levels.