Figure 1.
Flowchart for the construction of a functional brain network using R-fNIRS data.
(1) Schematic arrangement of the probe array (12 sources, red, and 24 detectors, blue, which configurate 46 measurement channels over the whole head, as indicated by digits from 1 to 46). (2) Extraction of the time course from R-fNIRS data from each measurement channel (i.e., network node). (3) Calculation of the correlation matrix for all pairs of channels or nodes. (4) Thresholding of the correlation matrix into a binary adjacency matrix. (5) Visualization of the binary adjacency matrix as a graph.
Figure 2.
Characteristics of oxy-Hb-based group-level correlation network.
(A) The correlation distribution and (B) its connectivity pattern. Only the topmost ranked 10% connection with correlation larger than 0.67 (blue line in A) are showed in B.
Figure 3.
Small-world properties and network efficiency of oxy-Hb-based functional networks as a function of sparsity threshold.
(A) Clustering coefficient, Cp; (B) characteristic path length, Lp; (C) normalized clustering coefficients,, and normalized characteristic path lengths,
; (D) local efficiency, Eloc; (E) global efficiency, Eglob; and (F) normalized local efficiency,
, and normalized global efficiency,
. For a wide range of sparsity thresholds, the real brain networks have larger values of Cp and Eloc as than random networks; however, the values of Lp and Eglob are approximately equal, resulting in γ>1 and ∼1 as well as
>1 and ∼1. These findings imply that R-fNIRS-based functional brain networks have prominent small-world features and are efficient in information processing. Error bars (A, B, D and E) correspond to standard errors of the mean for 1000 comparable random null networks. The gray areas indicate the sparsity range over which the parameters derived from real brain network are significantly (P<0.05) different from those derived from comparable random networks.
Figure 4.
Hierarchy and modularity of oxy-Hb-based functional networks as a function of sparsity threshold.
(A) Hierarchy coefficients,, and (B) modularity, Q, of the real (red) and random (green) networks as functions of sparsity threshold. Error bars correspond to standard errors of the mean for 1000 comparable random null networks. The gray areas indicate the sparsity range over which the parameters derived from real brain network are significantly (P<0.05) different from those derived from comparable random networks. These results demonstrate a significant non-random hierarchical and modular organization of the fNIRS-based functional human brain networks.
Figure 5.
Modular architectures of oxy-Hb-based functional network at several selective sparsity thresholds (0.1, 0.2 and 0.3).
Five, three and two functional modules are separately identified in the functional brain network. The size of each node denotes its relative nodal degree value in the brain network.
Figure 6.
Hubs (red) in oxy-Hb-based functional network.
The node sizes indicate their relative nodal centralities normalized to the corresponding mean for all nodes in the network. A region was considered a hub if its normalized nodal centrality was at least 1 standard deviation greater than the mean of all the nodes in the network.
Figure 7.
Reproducibility of oxy-Hb-based functional network properties across subjects.
The mean correlation matrices derived from subgroup 1 (A) and subgroup 2 (B) are highly similar to each other (C). The regional nodal metrics also exhibited strong correlations between the two subgroups (D) over a wide range of sparsity thresholds (degree: r = 0.56±0.14; efficiency: r = 0.57±0.16; and betweenness: r = 0.57±0.20). These results suggest a high reproducibility of fNIRS-based brain network metrics across subjects.
Figure 8.
Reproducibility of oxy-Hb-based functional network properties over time.
The mean correlation matrices derived from sub-dataset 1 (A) and sub-dataset 1 (B) are highly similar to each other (C). The regional nodal metrics also exhibited strong correlations between the two sub-datasets (D) over a wide range of sparsity thresholds (degree: r = 0.79±0.06; efficiency: r = 0.81±0.08; and betweenness: r = 0.68±0.14). These results suggest a high reproducibility of fNIRS-based brain network metrics over time.
Figure 9.
Small-world properties, network efficiency, Hierarchy (), and modularity (Q) of deoxy-Hb-based functional networks as a function of sparsity threshold.
Error bars correspond to standard deviation of the mean for 1000 comparable random null networks (blue lines). The gray areas indicate the sparsity range over which the parameters derived from real brain network (red lines) are significantly (P<0.05) different from those derived from comparable random networks. These results demonstrate efficient small-world properties, significant non-random hierarchical and modular organization of deoxy-Hb-based functional brain networks.
Figure 10.
Small-world properties, network efficiency, Hierarchy (), and modularity (Q) of total-Hb-based functional networks as a function of sparsity threshold.
Error bars correspond to standard deviation of the mean for 1000 comparable random null networks (blue lines). The gray areas indicate the sparsity range over which the parameters derived from real brain network (red lines) are significantly (P<0.05) different from those derived from comparable random networks. Again, these results demonstrate efficient small-world properties, significant non-random hierarchical and modular organization of total-Hb-based functional brain networks.
Figure 11.
The differences of network properties derived from oxy-Hb, deoxy-Hb and total-Hb signals.
Bars show the mean areas under curves (AUC) of (A) small-world parameters (Cp, Lp, and
), (B) network efficiency (Eloc and Eglob), (C) hierarchy coefficient (
) and modularity (Q). Error bars correspond to standard deviation of the mean across participants. The asterisk indicates P<0.05 and double asterisk indicates P<0.01.
Figure 12.
Small-world properties, network efficiency, hierarchy (), and modularity (Q) of individual oxy-Hb-based functional networks as a function of sparsity threshold.
The gray areas indicate the sparsity range over which the parameters derived from all individual brain networks are significantly (P<0.05) different from those derived from comparable random networks.