In this article we present a novel approach for
diffusion MRI global tractography. Our formulation models
the signal in each voxel as a linear combination of
fiber-tract basis func- tions, which consist of a
comprehensive set of plausible fiber tracts that are
locally compatible with the measured MR signal. This
large dictionary of candidate fibers is directly
estimated from the data and, subsequently, efficient
convex optimization techniques are used for recovering
the smallest subset globally best fitting the measured
signal. Experimen- tal results conducted on a realistic
phantom demonstrate that our approach significantly
reduces the computational cost of global tractography
while still attaining a reconstruction quality at least
as good as the state-of-the-art global methods.