A Whole-Brain Reconstruction Approach for FOD Modeling from Multi-Shell Diffusion MRI

Abstract

With the advance of connectome imaging techniques, there is a great need of robust methods for modeling the distribution of fiber orientations from multi-shell diffusion imaging. Existing tools for fiber orientation distribution (FOD) reconstruction, however, predominantly solves this problem on a voxel-by-voxel basis, disregarding the spatial regularity in brain anatomy. In this work, we propose a novel computational framework for the joint reconstruction of FODs over the whole brain volume. Our framework takes into account compartment modeling from multi-shell imaging data and uses an operator splitting scheme to decouple the whole-brain reconstruction problem into a series of local computations. Within this framework, we can investigate both isotropic and anisotropic regularizations. In the experiments, we conduct extensive simulations to compare the performance of both types of regularizations and show that anisotropic regularization produces more robust results across various fiber configurations. We also apply our method to in vivo data from 80 HCP subjects and evaluate the impact of FOD modeling methods on the reconstruction of the challenging fiber bundles from the locus coeruleus (LC) nuclei. Our results indicate that the proposed whole-brain approach for FOD modeling leads to more robust LC fiber bundle reconstruction than results from voxel-wise modeling.

Y. Shi—This work was in part supported by the National Institute of Health (NIH) under Grant R01EB022744, P41EB015922, U01EY025864, U01AG051218, P50AG05142.