Fast high spatio-angular resolution estimation of the neuronal fiber orientations in the brain with diffusion MRI
MetadataShow full item record
The human brain is a very complex organ. The white matter tissue composing the brain is made of threadlike structures, called axons, which are responsible for the transmission of the impulses between different areas of the brain. Axons are extremely fine and cannot be visualized by any in vivo imaging technique, leaving a lot to explore about which brain regions are connected and how information is carried through these structures. Diffusion Magnetic Resonance Imaging (dMRI) is a unique technique that allows to investigate the inner structures of the brain in vivo and in a non-invasive way. High spatio-angular resolution dMRI techniques have been shown to provide accurate fiber reconstructions even in the presence of complex fiber configurations. However, high resolution techniques are characterized by long acquisition times, which hamper their application into the clinical practice. In this manuscript we present a novel method to recover the fiber orientation distribution (FOD) of the bundles of axons at high spatio-angular resolution via practical kq-space under-sampling that enables both acceleration and super-resolution. The quality of the recovered fibers is preserved by making use of advanced anatomical priors for the FOD reconstruction. Prior knowledge of the spatial distribution of the white matter, the gray matter and the cerebrospinal fluid is taken into account for the recovery of the FOD coefficients. In addition, the simultaneous voxelwise sparsity and spatial smoothness of fiber orientations is accounted for by means of a structured sparsity prior. A convex minimization problem is formulated and solved via an accelerated stochastic Forward-Backward algorithm. Simulations show that the proposed method outperforms state-of-the-art kq-space approaches in terms of reconstruction quality. Real data analysis suggests that accurate FOD mapping can be achieved from severe kq-space under-sampling regimes, potentially enabling the application of high spatio-angular resolution dMRI into the clinical practice.