Negative edges and soft thresholding in complex network analysis of resting state functional connectivity data
Adam J Schwarz1, John McGonigle
1Department of Psychological and Brain Sciences, Indiana University, 1101 E. 10th Street, Bloomington, IN 47405, USA. adamschw@indiana.edu
Neuroimage
|January 4, 2011
Summary
Investigating brain functional connectivity networks reveals that soft thresholding, unlike hard thresholding, preserves network properties across different correlation ranges. This method enhances reproducibility and avoids issues from global signal removal.
Area of Science:
- Neuroscience
- Network Science
- Data Analysis
Background:
- Functional connectivity analysis often uses hard thresholding on correlation matrices, typically focusing on the right tail of the distribution.
- This approach may overlook important network properties in other correlation ranges, particularly the left (negative) tail, and is sensitive to confound signal removal strategies.
Purpose of the Study:
- To investigate network properties across the full correlation histogram, including the left tail.
- To assess the impact of different confound signal removal strategies on network topology.
- To develop and evaluate a soft-thresholding approach for more robust functional connectivity analysis.
Main Methods:
- Analysis of functional connectivity networks using edges constrained to specific ranges (left and right tails) of the correlation histogram.
- Comparison of network properties (modularity, clustering, assortativity) with and without global signal correction and deconvolution of specific confound signals (white matter, CSF, motion).
- Development and application of a soft-thresholding method using a power law adjacency function to weight all correlation values.
Main Results:
- In the absence of global signal correction, left-tail networks showed reduced modularity and clustering, with negative assortativity.
- Deconvolution of specific confound signals improved within-subject reproducibility of global network parameters.
- Soft-thresholding with a power law function (β=12) produced reproducible, modular, small-world, and scale-free-like networks, aligning with hard-thresholded values while avoiding fragmentation.
Conclusions:
- The choice of thresholding method and confound removal strategy significantly impacts functional connectivity network topology, especially in the left tail of the correlation histogram.
- Soft-thresholding offers a more robust and reproducible approach to analyzing functional connectivity, preserving network properties across the full correlation spectrum.
- Soft-thresholding mitigates issues associated with hard thresholding, such as network fragmentation and loss of information from weaker connections.
