In the past decade, resting-state functional magnetic resonance imaging (rs-fMRI) and graph-based measures have been widely used to quantitatively characterize the architectures of brain functional networks in healthy individuals and in patients with abnormalities related to psychopathic and neurological disorders. To accurately evaluate the topological organization of brain functional networks, the definition of the nodes and edges for the construction of functional networks is critical. Furthermore, both types of brain functional networks (binarized networks and weighted networks) are widely used to analyze topological organization. However, how to best select the network type is still debated. Consequently, we investigated the test-retest reliability of brain functional networks with binarized and weighted edges using two independent datasets and four strategies for defining nodes. We revealed fair to good reliability for a majority of network topological attributes and overall higher reliabilities for weighted networks than for binarized networks. For regional nodal efficiency, weighted networks also showed higher reliability across nodes. Thus, our findings imply that weighted networks contain more information, leading to more stable results. In addition, we found that the reliability of brain functional networks was influenced by the node definition strategy and that more precise of nodal definition were associated with higher reliability. The time effect of reliability was restricted, as no differences between long-term and short-term reliability were observed. In conclusion, our results suggest that weighted networks have better reliability because they reflect more topological information, implying broader applications of weighted networks related to normal and disordered function of the human brain.
Keywords: Binarized and weighted edges; Graph-based measures; Resting-state fMRI; Test-retest reliability.