Academic journal article Cognitive, Affective and Behavioral Neuroscience

Reproducibility of Functional Network Metrics and Network Structure: A Comparison of Task-Related BOLD, Resting ASL with BOLD Contrast, and Resting Cerebral Blood Flow

Academic journal article Cognitive, Affective and Behavioral Neuroscience

Reproducibility of Functional Network Metrics and Network Structure: A Comparison of Task-Related BOLD, Resting ASL with BOLD Contrast, and Resting Cerebral Blood Flow

Article excerpt

Abstract Network analysis is an emerging approach to functional connectivity in which the brain is construed as a graph and its connectivity and information processing estimated by mathematical characterizations of graphs. There has been little to no work examining the reproducibility of network metrics derived from different types of functional magnetic resonance imaging (fMRI) data (e.g., resting vs. task related, or pulse sequences other than standard blood oxygen level dependent [BOLD] data) or of measures of network structure at levels other than summary statistics. Here, we take up these questions, comparing the reproducibility of graphs derived from resting arterial spin-labeling perfusion fMRI with those derived from BOLD scans collected while the participant was performing a task. We also examine the reproducibility of the anatomical connectivity implied by the graph by investigating test-retest consistency of the graphs' edges. We compare two measures of graphedge consistency both within versus between subjects and across data types. We find a dissociation in the reproducibility of network metrics, with metrics from resting data most reproducible at lower frequencies and metrics from taskrelated data most reproducible at higher frequencies; that same dissociation is not recapitulated, however, in network structure, for which the task-related data are most consistent at all frequencies. Implications for the practice of network analysis are discussed.

Keywords fMRI . Network analysis . Functional connectivity . Reproducibility . Perfusion

(ProQuest: ... denotes formula omitted.)

Introduction

An emerging approach to neuroimaging data analysis is the construal of the brain's connections as a graph, in the mathematical sense of a set of nodes connected by edges, and the application of mathematical characterizations of graphs to the brain. For present purposes, we refer to this cognitive neuroscience method as network analysis, although network methods are widely used in other fields of the natural and social sciences. A four-dimensional fMRI data set is generally converted into a graph by parceling it into regions, computing the correlation matrix of the time courses in those regions, and applying some transformation to those correlations so that some regions are connected to one another and others are not. (Nonbinary approaches-i.e., examinations of graphs with edges that are signed and/or weighted-are possible but rare as yet.) The mathematical characterization of a graph can be regional-that is, applicable to a single node-or network-wide. Node metrics generally aim to characterize the centrality of a node, in the sense of the strength of its connectivity to other nodes. Network-wide metrics are somewhat more diverse. Here, we review six network metrics that are popular subjects of study in this emerging field.

Transitivity, efficiency,andmodularity are, respectively, measures of clustering, efficiency of information transfer (derived from the mean path length between any given pair of nodes), and community structure. Gamma and lambda are measures of clustering and path length normalized relative to random graphs; their ratio quantifies a graph's small- worldness. Small-world networks are interesting in that they are highly clustered, yet few (<10) edges intervene between any given node and any other (Watts & Strogatz, 1998); such networks are robust to both targeted attack and random error (Achard, Salvador, Whitcher, Suckling, & Bullmore, 2006). Several studies have reported that the brains of healthy adults appear to be organized in small-world networks (Achard et al., 2006; Eguiluz, Chialvo, Cecchi, Baliki, & Apkarian, 2005). Small-worldness and other network metrics have been reported as markers of individual traits, including age, intelligence, and language onset, and clinical status, including Alzheimer's disease, mild cognitive impairment, schizophrenia, traumatic brain injury, epilepsy, and ADHD (for reviews, see Bullmore & Sporns, 2009, 2012; He & Evans, 2010; Wang, Zuo, & He, 2010; see also Beckage, Smith, & Hills, 2011; Sato, Takahashi, Hoexter, Massirer, & Fujita, 2013). …

Search by... Author
Show... All Results Primary Sources Peer-reviewed

Oops!

An unknown error has occurred. Please click the button below to reload the page. If the problem persists, please try again in a little while.