Connectome-Based Smoothing (CBS) for structural connectomes
Date:
Presentation at the “Maths in the Brain” Conference organized by the Turner Institute for Brain and Mental Health & Monash Biomedical Imaging.
Abstract:
Structural connectomes computed from diffusion MRI tractography are increasingly mapped at higher spatial resolutions comprising thousands of network nodes. High-resolution connectomes refer to those mapped at the resolution of vertices forming the cortical surface mesh. These connectomes are particularly susceptible to image registration misalignment, tractography artifacts, and noise, all of which can lead to reductions in connectome accuracy and test-retest reliability. We’ve implemented a connectome analogue of image smoothing to increase the robustness of mapped high-resolution structural connectomes against noise and registration errors. Connectome-Based Smoothing (CBS) involves jointly applying a carefully chosen bivariate smoothing kernel to the two endpoints of all tractography streamlines, yielding a spatially smoothed connectivity matrix. CBS proposed a computationally efficient method using a matrix congruence transformation on sparse connectome representations to smooth connectivity matrices. We evaluated a range of different smoothing kernels on CBS performance. We find that CBS substantially improves the identifiability, sensitivity and test-retest reliability of high-resolution connectivity maps, though at a cost of increasing storage burden. We show that the choice of smoothing kernel can balance the trade-off between connectome reliability and identifiability. Additionally, for connectomes mapped at the lower resolution of brain atlases, we show that CBS marginally improves the statistical power to detect associations between structural connectivity and cognitive performance, particularly for connectomes mapped using probabilistic tractography. CBS was also found to enable more replicable statistical inferences compared to connectomes without any smoothing. We conclude that spatial smoothing is particularly important for the reliability of high-resolution structural connectomes, but can also provide benefits to connectomes mapped at a lower parcellation resolution. Given that CBS is a fast operation relative to the time required to perform whole-brain tractography, we anticipate that spatial smoothing will be integrated as a common step in established connectome mapping workflows.