Limits...
A data-driven method to reduce the impact of region size on degree metrics in voxel-wise functional brain networks.

Liu C, Tian X - Front Neurol (2014)

Bottom Line: In this study, we designed a data-driven method to reduce this impact of the region size in degree metrics by (1) disregarding all self-connections among voxels within the same region and (2) regulating connections from voxels of other regions by the sizes of those regions.In addition, our modified method improved test-retest reliability of degree metrics as well as the sensitivity in group-statistic comparisons.As a promising new tool, our method may reveal network properties that better represent true brain architecture without compromising its data-driven advantage.

View Article: PubMed Central - PubMed

Affiliation: Queensland Brain Institute, The University of Queensland , St Lucia, QLD , Australia.

ABSTRACT
Degree, which is the number of connections incident upon a node, measures the relative importance of the node within a network. By computing degree metrics in voxel-wise functional brain networks, many studies performed high-resolution mapping of brain network hubs using resting-state functional magnetic resonance imaging. Despite its extensive applications, defining nodes as voxels without considering the different sizes of brain regions may result in a network where the degree cannot accurately represent the importance of nodes. In this study, we designed a data-driven method to reduce this impact of the region size in degree metrics by (1) disregarding all self-connections among voxels within the same region and (2) regulating connections from voxels of other regions by the sizes of those regions. The modified method that we proposed allowed direct evaluation of the impact of the region size, showing that traditional degree metrics overestimated the degree of previous identified hubs in humans, including the visual cortex, precuneus/posterior cingulate cortex, and posterior parietal cortex, and underestimated the degree of regions including the insular cortex, anterior cingulate cortex, parahippocampus, sensory and motor cortex, and supplementary motor area. However, the locations of prominent hubs were stable even after correcting the impact. These findings were robust under different connectivity thresholds, degree metrics, data-preprocessing procedures, and datasets. In addition, our modified method improved test-retest reliability of degree metrics as well as the sensitivity in group-statistic comparisons. As a promising new tool, our method may reveal network properties that better represent true brain architecture without compromising its data-driven advantage.

No MeSH data available.


Comparison between temporal and spatial region growing is shown. (A) Illustrations of temporal and spatial region growing represented by their type I cluster-size maps. (B) Paired two-sample t-test between temporal and spatial region growing (p < 0.05, FDR corrected). (C) Similarity between temporal and spatial region growing represented by intra-class correlation.
© Copyright Policy - open-access
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4195315&req=5

Figure 5: Comparison between temporal and spatial region growing is shown. (A) Illustrations of temporal and spatial region growing represented by their type I cluster-size maps. (B) Paired two-sample t-test between temporal and spatial region growing (p < 0.05, FDR corrected). (C) Similarity between temporal and spatial region growing represented by intra-class correlation.

Mentions: Figure 5 shows the region-growing results of data without GSR, represented by the standardized cluster size of each voxel (type I cluster-size maps). Results with GSR were similar. Temporal region growing and spatial region growing produce similar results as revealed by high ICC (mean ± SD = 0.69 ± 0.17) between them (Figure 5C). Both results indicate the heterogeneity of functional region sizes. The visual cortex and precuneus/posterior cingulate cortex have prominent larger cluster sizes than the other brain regions. However, temporal and spatial region growing are not identical. Temporal region growing results in significantly larger clusters in the lateral prefrontal cortex, parietal cortex, and precuneus, and smaller clusters in supplementary motor cortex and insular cortex than spatial region growing (Figure 5B, p < 0.05, FDR corrected). These differences reflected the nature of temporal and spatial correlation. The information used to calculate temporal correlation is derived from preprocessed time series, whereas the information for spatial correlation involves whole-brain functional connectivity patterns. Voxels located in association brain regions, such as prefrontal cortex, parietal cortex, and cingulate cortex/precuneus, may have similar preprocessed time series (due to real local connections and blurring from data preprocessing), but may have distinct whole-brain functional connectivity patterns, because they are involved in different functional systems. Thus, compared with temporal correlation, the spatial correlation can be a better feature to distinguish voxels in association brain regions, and may result in smaller region-growing clusters in these regions. In contrast, whole-brain functional connectivity patterns of voxels related to sensory and motor function may be less important. The temporal and spatial region-growing results were intersected and resulted in combined region-growing results for degree analysis. The combination of two region growing allows more accurate estimation of functional region sizes because temporal and spatial correlations may reflect different aspects of functional connectivity features.


A data-driven method to reduce the impact of region size on degree metrics in voxel-wise functional brain networks.

Liu C, Tian X - Front Neurol (2014)

Comparison between temporal and spatial region growing is shown. (A) Illustrations of temporal and spatial region growing represented by their type I cluster-size maps. (B) Paired two-sample t-test between temporal and spatial region growing (p < 0.05, FDR corrected). (C) Similarity between temporal and spatial region growing represented by intra-class correlation.
© Copyright Policy - open-access
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4195315&req=5

Figure 5: Comparison between temporal and spatial region growing is shown. (A) Illustrations of temporal and spatial region growing represented by their type I cluster-size maps. (B) Paired two-sample t-test between temporal and spatial region growing (p < 0.05, FDR corrected). (C) Similarity between temporal and spatial region growing represented by intra-class correlation.
Mentions: Figure 5 shows the region-growing results of data without GSR, represented by the standardized cluster size of each voxel (type I cluster-size maps). Results with GSR were similar. Temporal region growing and spatial region growing produce similar results as revealed by high ICC (mean ± SD = 0.69 ± 0.17) between them (Figure 5C). Both results indicate the heterogeneity of functional region sizes. The visual cortex and precuneus/posterior cingulate cortex have prominent larger cluster sizes than the other brain regions. However, temporal and spatial region growing are not identical. Temporal region growing results in significantly larger clusters in the lateral prefrontal cortex, parietal cortex, and precuneus, and smaller clusters in supplementary motor cortex and insular cortex than spatial region growing (Figure 5B, p < 0.05, FDR corrected). These differences reflected the nature of temporal and spatial correlation. The information used to calculate temporal correlation is derived from preprocessed time series, whereas the information for spatial correlation involves whole-brain functional connectivity patterns. Voxels located in association brain regions, such as prefrontal cortex, parietal cortex, and cingulate cortex/precuneus, may have similar preprocessed time series (due to real local connections and blurring from data preprocessing), but may have distinct whole-brain functional connectivity patterns, because they are involved in different functional systems. Thus, compared with temporal correlation, the spatial correlation can be a better feature to distinguish voxels in association brain regions, and may result in smaller region-growing clusters in these regions. In contrast, whole-brain functional connectivity patterns of voxels related to sensory and motor function may be less important. The temporal and spatial region-growing results were intersected and resulted in combined region-growing results for degree analysis. The combination of two region growing allows more accurate estimation of functional region sizes because temporal and spatial correlations may reflect different aspects of functional connectivity features.

Bottom Line: In this study, we designed a data-driven method to reduce this impact of the region size in degree metrics by (1) disregarding all self-connections among voxels within the same region and (2) regulating connections from voxels of other regions by the sizes of those regions.In addition, our modified method improved test-retest reliability of degree metrics as well as the sensitivity in group-statistic comparisons.As a promising new tool, our method may reveal network properties that better represent true brain architecture without compromising its data-driven advantage.

View Article: PubMed Central - PubMed

Affiliation: Queensland Brain Institute, The University of Queensland , St Lucia, QLD , Australia.

ABSTRACT
Degree, which is the number of connections incident upon a node, measures the relative importance of the node within a network. By computing degree metrics in voxel-wise functional brain networks, many studies performed high-resolution mapping of brain network hubs using resting-state functional magnetic resonance imaging. Despite its extensive applications, defining nodes as voxels without considering the different sizes of brain regions may result in a network where the degree cannot accurately represent the importance of nodes. In this study, we designed a data-driven method to reduce this impact of the region size in degree metrics by (1) disregarding all self-connections among voxels within the same region and (2) regulating connections from voxels of other regions by the sizes of those regions. The modified method that we proposed allowed direct evaluation of the impact of the region size, showing that traditional degree metrics overestimated the degree of previous identified hubs in humans, including the visual cortex, precuneus/posterior cingulate cortex, and posterior parietal cortex, and underestimated the degree of regions including the insular cortex, anterior cingulate cortex, parahippocampus, sensory and motor cortex, and supplementary motor area. However, the locations of prominent hubs were stable even after correcting the impact. These findings were robust under different connectivity thresholds, degree metrics, data-preprocessing procedures, and datasets. In addition, our modified method improved test-retest reliability of degree metrics as well as the sensitivity in group-statistic comparisons. As a promising new tool, our method may reveal network properties that better represent true brain architecture without compromising its data-driven advantage.

No MeSH data available.