Network analysis of resting state EEG in the developing young brain: structure comes with maturation

Abstract

During childhood, brain structure and function changes substantially. Recently, graph theory has been introduced to model connectivity in the brain. Small-world networks, such as the brain, combine optimal properties of both ordered and random networks, i.e., high clustering and short path lengths. We used graph theoretical concepts to examine changes in functional brain networks during normal development in young children. Resting-state eyes-closed electroencephalography (EEG) was recorded (14 channels) from 227 children twice at 5 and 7 years of age. Synchronization likelihood (SL) was calculated in three different frequency bands and between each pair of electrodes to obtain SL-weighted graphs. Mean normalized clustering index, average path length and weight dispersion were calculated to characterize network organization. Repeated measures analysis of variance tested for time and gender effects. For all frequency bands mean SL decreased from 5 to 7 years. Clustering coefficient increased in the alpha band. Path length increased in all frequency bands. Mean normalized weight dispersion decreased in beta band. Girls showed higher synchronization for all frequency bands and a higher mean clustering in alpha and beta bands. The overall decrease in functional connectivity (SL) might reflect pruning of unused synapses and preservation of strong connections resulting in more cost-effective networks. Accordingly, we found increases in average clustering and path length and decreased weight dispersion indicating that normal brain maturation is characterized by a shift from random to more organized small-world functional networks. This developmental process is influenced by gender differences early in development.

Figure 1

Relative power spectra at 5 and 7 years of age recorded from 14 EEG channels at the following scalp locations: prefrontal (Fp1; Fp2), frontal (F3; F4; F7; F8), central (C3; C4), parietal (P3; P4), temporal (T5; T6), and occipital (O1; O2). In all power spectra, the vertical markers on the X‐scale correspond to 4 Hz steps in the spectrum starting at 0 Hz. Y‐scale values are arbitrary due to computation of relative power spectra, ranging from 0 to 0.06 in all channels.

Figure 2

Schematic representation of graph analysis applied to EEG recordings of brain activity. The first step (A) consists of filtering of the EEG signal in the frequency band of interest. Synchronization likelihood (SL) was calculated as a measure of generalized synchronization between all possible pairs of EEG channels (B), resulting in a synchronization diagram (C) with the likelihood of synchronization between channels indicated with black and white scale. Next, the synchronization matrix was converted to weighted graphs (D) with links of varying thickness that represent SL between nodes (channels). From these graphs, measures such as the clustering coefficient (Cw) and path length (Lw) were computed. For comparisons, networks were randomized by shuffling the cells of the SL matrix, resulting in randomized graphs (E). From random graphs graph, parameters were calculated and averaged. Finally, the ratio of the graphs parameters of the original networks and the mean of the graph parameters for the randomized networks was computed (F).

Figure 3

Mean SL over all epochs for boys and girls at 5 and 7 years of age in three frequency bands. The variance in SL was significantly lower in children at 7 years of age compared to that at 5 years of age in theta [F(1,225) = 30.116, P < 0.001], alpha [F(1,225) = 8.330, P = 0.004] and beta [F(1,225) = 29.367, P < 0.001] band. Boys had significant lower SL in theta [F(1,225) = 14.616, P < 0.001], alpha [F(1,225) = 8.025, P = 0.005] and beta [F(1,225) = 16.796, P < 0.001] band. The beta frequency band showed a significant interaction effect between time and gender [F(1,225) = 5.116, P = 0.025].

Figure 4

Mean normalized clustering index (Cw/Cw − s) for boys and girls at 5 and 7 years of age in three frequency bands. The mean clustering index was significant higher in children of 7 years of age compared to children at 5 years of age in the alpha band (F = 7.087, P = 0.008). Girls showed higher clustering in the alpha (F = 10.966, P = 0.001) and beta (F = 9.207, P = 0.003) bands.

Figure 5

Mean normalized path length (Lw/Lw − s) over all epochs for children at 5 and 7 years of age in three frequency bands. The mean normalized path length was significant higher in children at 7 years of age compared to children at 5 years of age in theta (F = 28.297, P < 0.001), alpha (F = 30.989, P < 0.001) and beta (F = 55.416, P < 0.001) bands.

Figure 6

Mean normalized weight dispersion over all epochs for boys and girls at 5 and 7 years of age in three frequency bands. Weight dispersion was significantly lower at 7 years of age compared to that at 5 years of age in the theta [F(1,225) = 8.188, P = 0.005], alpha [F(1,225) = 8.468, P = 0.004] and beta [F(1,225) = 34.756, P < 0.001] band. Girls had significant lower weight dispersion in the beta band [F(1,225) = 7.153, P = 0.008].The alpha frequency band showed a significant interaction effect between time and gender [F(1,225) = 5.252, P = 0.023].