Reproducibility of structural brain connectivity and network metrics using probabilistic diffusion tractography.

Area of Science:

• Neuroscience
• Medical Imaging
• Network Science

Background:

• Structural connectivity networks, derived from diffusion tractography, are characterized by network metrics.
• Evaluating the reproducibility of these networks and their metrics is crucial for reliable analysis.

Purpose of the Study:

• To assess the short-term test-retest reproducibility of structural brain networks and their metrics.
• To investigate the impact of various connectivity thresholds and sparsity levels on reproducibility.

Main Methods:

• Structural networks and metrics were evaluated in 30 subjects using within- and between-subject coefficients of variance (CVws, CVbs) and intra-class correlation (ICC).
• Analyses were performed across various connectivity thresholds and sparsity conditions (same vs. different sparsity between groups).

Main Results:

• A connectivity threshold of 0.01 excluded approximately 80% of edges, yielding CVws=73.2±37.7%, CVbs=119.3±44.0%, and ICC=0.62±0.19.
• The remaining 20% of edges showed better reproducibility (CVws<45%, CVbs<90%, ICC=0.75±0.12).
• A 1% difference in sparsity introduced additional within-subject variations in network metrics.

Conclusions:

• Applying connectivity thresholds to structural networks is necessary to eliminate spurious connections for accurate network analysis.
• A connectivity threshold above 0.01 can be used without significantly affecting short-term reproducibility when sparsity is consistent.
• Integrating various connectivity thresholds provides reliable network metric estimation when sparsity levels differ between subjects.