On evaluating the accuracy and biological plausibility of diffusion MRI tractograms

One of the major limitations of diffusion MRI tractography is that the tractograms, i.e. set of fiber tracts, recovered by existing algorithms are not truly quantitative. Hence, the structural connectivity between different brain regions, a.k.a. connectomics, is nowadays quantified by counting the number of recovered pathways or averaging some scalar maps along them; in both cases, these estimates provide only indirect measures of the true underlying neuronal connectivity. A number of methods have recently started to appear to address this limitation; in particular, COMMIT and LiFE have been developed upon the recently proposed framework that showed how to formulate tractography as an efficient system of linear equations, opening de facto the door for the practical possibility to evaluate and compare the accuracy of the tractograms. These two models follow different strategies to describe the signal in each voxel. On one hand, COMMIT uses a forward-model that takes into account that the diffusion MR signal can originate from distinct water pools, e.g. intra- and extra-cellular. On the other hand, LiFE models the signal as consisting only of contributions arising from the tracts passing through each voxel (i.e. restricted diffusion). The extra-cellular space around the axons (i.e. hindered diffusion) and any partial volume that can occur with non white-matter (WM) tissue (i.e. isotropic diffusion) are not directly considered, but are “removed” with a de-meaning procedure. However, as shown by several independent studies, the relative contribution of these compartments is not homogeneous in the WM and can change considerably. The schematic representation in the top-right figure depicts such a situation: the callosal fibers projecting from the corpus callosum (CC) and the corticospinal tract (CST) consist both of tightly-packed axons (yellow circles) that progressively fan-out and eventually cross. Differences in the axonal packing density are compensated by variations in the spacing surrounding the axons themselves, i.e. extra-cellular space. This consideration is implicitly or explicitly assumed in most state-of-the-art techniques for voxelwise microstructure imaging and independent histological studies also corroborate this hypothesis. Furthermore, to be sensitive with diffusion MRI to the tissue microstructure, multiple b-values have been proven necessary. In this work we investigated the importance of using (i) proper multi-compartment models and (ii) adequate multiple b-value acquisitions in order to be able to evaluate the accuracy and the biological plausibility of the tractograms using these global approaches.