Network scaling effects in graph analytic studies of human resting-state FMRI data - PubMed (original) (raw)

Network scaling effects in graph analytic studies of human resting-state FMRI data

Alex Fornito et al. Front Syst Neurosci. 2010.

Abstract

Graph analysis has become an increasingly popular tool for characterizing topological properties of brain connectivity networks. Within this approach, the brain is modeled as a graph comprising N nodes connected by M edges. In functional magnetic resonance imaging (fMRI) studies, the nodes typically represent brain regions and the edges some measure of interaction between them. These nodes are commonly defined using a variety of regional parcellation templates, which can vary both in the volume sampled by each region, and the number of regions parcellated. Here, we sought to investigate how such variations in parcellation templates affect key graph analytic measures of functional brain organization using resting-state fMRI in 30 healthy volunteers. Seven different parcellation resolutions (84, 91, 230, 438, 890, 1314, and 4320 regions) were investigated. We found that gross inferences regarding network topology, such as whether the brain is small-world or scale-free, were robust to the template used, but that both absolute values of, and individual differences in, specific parameters such as path length, clustering, small-worldness, and degree distribution descriptors varied considerably across the resolutions studied. These findings underscore the need to consider the effect that a specific parcellation approach has on graph analytic findings in human fMRI studies, and indicate that results obtained using different templates may not be directly comparable.

Keywords: BOLD; complex; cortex; spontaneous.

PubMed Disclaimer

Figures

Figure 1

Figure 1

Overview of image processing steps used in generating graph-based representations of whole-brain connectivity network. Far left: seven distinct parcellation templates were generated, which divided the brain into (from top to bottom) 84, 91, 230, 438, 890, 1314, or 4230 regions-of-interest. Middle-left: the templates were applied to each subject's fMRI volumes and the mean time series of each region were extracted and decomposed into four frequency intervals using a wavelet transform. Middle-right: spontaneous oscillations subtended by the wavelet frequency interval 0.04–0.08 Hz were further corrected for physiological noise signals and temporal correlations between each possible pair of regional corrected time series were calculated to generate a functional connectivity matrix for each template for each participant. These matrices were then thresholded and binarized across a range of connection costs (examples shown are for 10, 20, 30, and 40% costs). Far right: The thresholded, binarized adajacency matrices were used to generate graph-based representations of network connectivity, such that each region was represented as a node and each supra-threshold correlation as a connecting edge. These graphs were used as a basis for calculating graph metrics.

Figure 2

Figure 2

Functional connectivity histograms obtained at each parcellation scale. Each line corresponds to the sample average histogram of correlation values contained in each participant's unthresholded functional connectivity matrix.

Figure 3

Figure 3

Boxplots of subject-specific correlations between mean regional functional connectivity and regional volumes at each parcellation scale. Boxes represent the inter-quartile range, heavy horizontal lines the median. Whiskers represent the 5th and 95th percentiles. Circles represent values beyond these percentiles.

Figure 4

Figure 4

Mean values and individual differences in network connectedness as a function of parcellation scale. Left: Sample mean size of largest connected network component for each parcellation scale across all costs examined (*p < 0.05). Right: correlation matrix of inter-scale associations for the cost at which each individual's network became connected.

Figure 5

Figure 5

Mean values of global network path length (left) and clustering coefficient (right) at each cost. *p < 0.05.

Figure 6

Figure 6

Mean values of global network path length (top left), λER (top middle), λRW(top right), clustering (bottom left), γER (bottom middle) and γRW (bottom right) at costs of 10, 20, 30, and 40% for each parcellation scale. *p < 0.05.

Figure 7

Figure 7

Mean small-worldness (σ) values computing using ER- or RW-normalization (left and right respectively) for each parcellation scale at costs of 10, 20, 30, and 40%. *p < 0.05.

Figure 8

Figure 8

Inter-scale correlations in global network properties at costs of 10, 20, 30, and 40%. (A) Clustering; (B) γRW; (C) path length; (D) λRW; (E) σER; and (F) σRW.

Figure 9

Figure 9

Log–log nodal rank-degree plots for each parcellation scale at costs of 10, 20, 30, and 40%.

Figure 10

Figure 10

Mean values of the power-law exponent (slope, left) and exponential cutoff (right) of the degree distributions at each parcellation scale for costs of 10, 20, 30 and 40%. *p < 0.05.

Figure 11

Figure 11

Inter-scale correlations in degree distribution power-law exponents (top row) and exponential cut-off values (bottom row) for costs of 10, 20, 30, and 40% (left to right columns).

Figure 12

Figure 12

Medial and lateral cortical surface renderings illustrating regional variations in sample mean path length (left) and clustering (right) at each parcellation scale. Path length values have been inverted so that lower values are represented by “hotter” colors.

Similar articles

Cited by

References

    1. Achard S., Bassett D. S., Meyer-Lindenberg A., Bullmore E. (2008). Fractal connectivity of long-memory networks. Phys. Rev. E. Stat. Nonlin. Soft Matter Phys. 77, 036104.10.1103/PhysRevE.77.036104 - DOI - PubMed
    1. Achard S., Bullmore E. (2007). Efficiency and cost of economical brain functional networks. PLoS Comput. Biol. 3, e17.10.1371/journal.pcbi.0030017 - DOI - PMC - PubMed
    1. Achard S., Salvador R., Whitcher B., Suckling J., Bullmore E. (2006). A resilient, low-frequency, small-world human brain functional network with highly connected association cortical hubs. J. Neurosci. 26, 63–7210.1523/JNEUROSCI.3874-05.2006 - DOI - PMC - PubMed
    1. Albert N. B., Robertson E. M., Miall R. C. (2009). The resting human brain and motor learning. Curr. Biol. 19, 1023–102710.1016/j.cub.2009.04.028 - DOI - PMC - PubMed
    1. Albert R., Barabasi A. L. (2002). Statistical mechanics of complex networks. Rev Modern Phys. 74, 47–9710.1103/RevModPhys.74.47 - DOI

LinkOut - more resources