Atlas of the short fiber bundles inferred from the Human Connectome Project diffusion MRI dataset

Using the outstanding diffusion MRI dataset of the Human Connectome Project, we inferred the largest atlas to date of reproducible short fiber bundles (less than 8cm) of the human brain. This atlas results from a framework aligning subjects using a diffeomorphic procedure forcing the match of the main cortical sulci. This is a key feature to align correctly the U-bundles, which circumvent the cortical folds. This sulcus-based alignment includes the nonlinear ICBM T1-weighted MRI average in order to drive all the subject’s tractograms (the set of dMRI-based virtual fibers) to the standard ICBM space. Then, a two level clustering (intrasubject and intersubject) is performed on the tractograms to define the most reproducible bundles across subjects. A postregistration with freesurfer space leads to label each bundle according to the nomenclature of the Desikan-Killiany atlas. We deliver probabilistic maps of each bundle in the ICBM space and an aggregated map providing a parcellation of superficial white matter defined from a maximum probability principle.