Small-world properties in the human resting-state brain: a graph theoretical approach
Conference Paper/Poster - May 25, 2018
Graph theoretical analysis of structural and functional systems gain insight into the organization of physiological and pathological human brain networks. However, most studies calculated network measures based on unweighted, undirected graphs whereby information may get lost due to thresholding. Aim was to generate a weighted undirected graph based on functional connectivity patterns of a resting-state human brain network and to investigate small-worldness, communities and main hubs.
Resting state f-MRI scans of 24 right handed healthy volunteers (12 male, 12 female, mean age 61.8) were included in this analysis. Volunteers were instructed to keep their eyes closed and to focus on their breathing during the scan. Scans were preprocessed with SPM12 and normalized to MNI space. BOLD time series were extracted for each subject from a 333 regions of interest defined functional atlas by Gordon et al. Pearsons correlation between signals was calculated which resulted in a weighted undirected graph for each volunteer, consisting of 333 nodes and 55278 edges. Small-World-Propensity and the betweenness centrality was calculated. Community structure detection using the Louvain-Algorithm (alpha-level 0.9) is based on the mean adjacency matrix of the graph.
Small-World-Propensity (mean 0.57 (±SD 0.12)) indicates small world properties in the resting-state brain network. Community detection revealed six modules reflecting functional and anatomical regions: (1) sensori-motor network, (2) dorsal-attention network, (3) visual network, (4) temporal network in the left hemisphere, (5) frontal network, and (6) temporal network in the right hemisphere. Regarding the classification of functional networks in the atlas of Gordon et al., nodes with high betweenness centrality (main hubs) were part of dorsal/ventral attention, sensori-motor, auditory and default mode network.
Resting state human brain networks have small world properties, indicating a high efficiency in information transfer and robustness to random error. We have demonstrated the existence of community structures using the Louvain-Algorithm, however more graph measures like cluster coefficient, triangles and nodal degree are necessary to identify connector hubs between communities in this complex network. Regarding pathological human brain networks, e.g. after stroke, graph analysis may contribute to get insight into pathoph¬¬¬ysiological processes and neuroplasticity.