Can hubs of the human connectome be identified consistently with diffusion MRI?

Abstract

Recent years have seen a surge in the use of diffusion MRI to map connectomes in humans, paralleled by a similar increase in pre-processing and analysis choices. Yet these different steps and their effects are rarely compared systematically. Here, in a healthy young adult population (n = 294), we characterized the impact of a range of analysis pipelines on one widely studied property of the human connectome; its degree distribution. We evaluated the effects of 40 pipelines (comparing common choices of parcellation, streamline seeding, tractography algorithm, and streamline propagation constraint) and 44 group-representative connectome reconstruction schemes on highly connected hub regions. We found that hub location is highly variable between pipelines. The choice of parcellation has a major influence on hub architecture, and hub connectivity is highly correlated with regional surface area in most of the assessed pipelines (ρ > 0.70 in 69% of the pipelines), particularly when using weighted networks. Overall, our results demonstrate the need for prudent decision-making when processing diffusion MRI data, and for carefully considering how different pre-processing choices can influence connectome organization.

Geometric constraints on human brain function

“…Nonetheless, the topological centrality, metabolic cost, and tight genetic control of such connections [66][67][68] suggest that they provide important functional and evolutionary advantages beyond wave-like dynamics 69 . The limited resolution and sensitivity to preprocessing pipelines 70,71 of dMRI and fMRI data complicate attempts to uncover the role of long-range connections, but high-quality animal tract-tracing and electrophysiological data may be helpful in this regard. The close coupling between geometry and dynamics is apparent in neocortical and non-neocortical structures alike, suggesting that the functional organization of regions outside the neocortex is also dominated by local anatomical connectivity and wave dynamics, as found in recent experiments 46,72,73 .…”

“…The superior performance of geometric eigenmodes offers an immediate practical benefit, since the modes can be estimated using only a mesh representation of the structure of interest, which can easily be derived using well-established, automated processing pipelines for T1-weighted anatomical images 82 . In contrast, connectome eigenmodes require a graph-based model of macroscopic inter-regional connectivity generated via complex data processing pipelines 71,83 ; the definition of graph nodes, which is a topic of contention 84 ; and the application of a thresholding procedure to remove putatively spurious connections, which our own analysis shows can affect the findings (Fig. S9).…”

A simulated annealing algorithm for randomizing weighted networks

“…Second, while the datasets covered a broad range of possible processing choices, they did not allow to delineate their individual effects on null model performance. As tools for multiverse analysis in connectomics are developed to facilitate the isolation of specific processing choices across a range of pipelines [40], future work can increasingly interrogate how they affect connectomes and downstream network analysis, including null network generation.…”

Structurally informed resting-state effective connectivity recapitulates cortical hierarchy

“…Tractography exhibits well-documented biases whereby a) short-range connections are overrepresented, overshadowing longer-distance tracts; and b) larger brain regions appear disproportionately connected (i.e., connectivity is positively correlated with regional surface area). 59 To account for these issues, for both the test-retest and validation samples, we generated distance-based consensus structural connectivity matrices per the approach described in ref. 60 .…”