Optimizing network neuroscience computation of individual differences in human spontaneous brain activity for test-retest reliability

Abstract

A rapidly emerging application of network neuroscience in neuroimaging studies has provided useful tools to understand individual differences in complex brain function. However, the variability of methodologies applied across studies - with respect to node definition, edge construction, and graph measurements- makes it difficult to directly compare findings and also challenging for end users to select the optimal strategies for mapping individual differences in brain networks. Here, we aim to provide a benchmark for best practices by systematically comparing the reliability of human brain network measurements of individual differences under different analytical strategies using the test-retest design of the resting-state functional magnetic resonance imaging from the Human Connectome Project. The results uncovered four essential principles to guide reliable network neuroscience of individual differences: 1) use a whole brain parcellation to define network nodes, including subcortical and cerebellar regions, 2) construct functional connectome using spontaneous brain activity in multiple slow bands, 3) optimize topological economy of networks at individual level, 4) characterise information flow with metrics of integration and segregation.

Age-related reorganization of functional network architecture in semantic cognition

“…Like most graph theoretical measures, the calculation of global efficiency requires a sparse graph to represent a biologically plausible network of functional connectivity 40 . Following recent work on the most reliable and representative construction of brain networks 33,40,41 , we calculated individual OMST based on the weighted functional connectivity matrices and used the OMST to assess global efficiency. A linear mixed-effects model indicated higher global efficiency for young adults ( x 2 = 21.86, p < 0.001; Figure 4d; Table S8).…”

“…While threshold-based filtering methods like proportional or absolute thresholding are commonly applied in network neuroscience, they are driven by an arbitrary choice of the respective threshold and suffer from low reliability (Luppi and Stamatakis, 2021). To avoid these pitfalls and based on recent research on the reliability of graph construction in neuroscience (Jiang et al, 2021; Luppi and Stamatakis, 2021), we calculated the orthogonalized minimum spanning tree (OMST; Dimitriadis et al, 2017) on the weighted functional connectivity matrices. Apart from its high reliability, the OMST has several advantages compared to commonly applied threshold-based methods of graph construction: It adheres to the intrinsic topological structure of the brain network by resulting in a fully connected, weighted graph and offers a data-driven method of individualized network construction accounting for each individual’s optimal state of economic wiring in terms of cost and efficiency.…”

“…While threshold-based filtering methods like proportional or absolute thresholding are commonly applied in network neuroscience, they are driven by an arbitrary choice of the respective threshold and suffer from low reliability 40 . To avoid these pitfalls and based on recent research on the reliability of graph construction in neuroscience 40,41 , we calculated the orthogonalized minimum spanning tree (OMST) 33 on the weighted functional connectivity matrices.…”

Benchmarking functional connectivity by the structure and geometry of the human brain

“…Despite the fact that we adopted a robust dataset and included a replication section, methodological choices including MRI acquisition scheme, processing pipeline, network reconstruction, and group consensus algorithm may still be susceptible to false positives and negatives (Jiang et al, 2021;Korhonen, Zanin, & Papo, 2021;Maier-Hein et al, 2017;Sarwar, Ramamohanarao, & Zalesky, 2021). In particular, the deterministic tractography procedure yields relatively sparse connectomes and future work should consider the effect of connectome reconstruction and sparsity on the definition of polysynaptic FCs.…”

A robust core architecture of functional brain networks supports topological resilience and cognitive performance in aging

“…Construction of functional brain networks using OMST-based methods tends to increase the amount of between-subject variability when applied to functional brain networks (Jiang, Betzel et al 2021). However, our analyses comparing different parameter selections showed that for the network features we studied, the strongest relationships with episodic memory occurred when we reduced the variability in initial network construction (via thresholding edges by their frequencies) ( Figure S7 ).…”