A Minimum Cost Approach to Connectivity from Orientation Distribution Functions via Efficient Multi-directional Graph Propagation

Regional connectivity measurements derived from diffusion imaging datasets are of considerable interest in the neuroimaging community for better understanding white matter connectivity. Current connectivity measurements are usually either based on fiber tractography applied in Monte-Carlo fashion, or variations of the Hamilton-Jacobi approach, or are graph-based. We propose a novel, graph-based algorithm that provides a fully deterministic, efficient and stable connectivity measure. This method handles crossing fibers and deals well with multiple seed regions. The computation is based on a multi-directional graph propagation algorithm applied to sampled orientation distribution functions computed directly from the original diffusion imaging data. While a maximum probability approach is possible, here we focus on a minimum cost formulation. We present results on synthetic and real datasets to illustrate the potential of our method towards subject-specific connectivity measurements performed in an efficient, stable and reproducible manner. Such individual connectivity measurements would be well suited for neuroimaging studies.