Joe Bathelt1, Helen O’Reilly1,2 , Jonathan D Clayden1, J Helen Cross1, Michelle de Haan1
Functional brain network organisation of children between 2 and
5 years derived from reconstructed activity of cortical sources of
high-density EEG recordings
1 University College London Institute of Child Health
2 University of Cambridge Autism Research Centre
Materials & Methods
The sample consisted of 47 typically-developing children (22 female) between the ages of 1.5 and 6
EEG was recorded using EGI Geodesic Dense Array Sensor Nets with 128 electrodes while
children watched a video clip of calming scenes for 2 minutes in a dimly lit room. The children were
video-recorded to assess the degree of attentiveness.
The EEG recordings were segmented into 1s epochs. The segmented recordings were visually
inspected. EEG channels that contained a high-level of noise or flat lines were rejected. Channels
with a peak-to-peak amplitude larger than 100µV were excluded for individual segments.
Boundary-element models (BEM) were calculated from average MRIs taken from the
Neurodevelopmental MRI database using FreeSurfer software v5.1.0 and MNE suite v2.7.0. A
cortical surface tessellation with 9750 vertices was used. BEMs were calculated for average MRIs
for children from 2 years to 6 years in 0.5 year intervals.
The cortical reconstructions were parcellated into regions of interest (ROI) using Freesurfer
according to the Desikan-Killiany atlas. The connectivity was analysed in three frequency bands.
The recordings were low-pass and high-pass filtered using digital 6th order Butterworth filters to
isolate the frequency bands of interest. Frequencies between 1 and 12, 12 and 25 and 25-40Hz
The time series of cortical ROIs derived from the EEG segments were correlated in pairwise
correlations. The average connectivity for a 1s segment was calculated for individual participants
and thresholded with the average connectivity.
Correlations between age and graph theory
measures Analysis of co-variance (ANCOVA)
shows a main effect of age and frequency band
(Age: F(5,85)= 3.465, p<0.01; Frequency band:
F(10,172)= 2.496, p<0.01). There was a
significant effect of age on all graph measures
(F1(1,30)= 7.63, p<0.01; F2(1,30)= 9.64, p<0.01;
F3(1,30)=6.6, p<0.05; F4(1,30)=4.84, p<0.05).
Frequency band had a significant effect on
clustering coefficient and characteristic path
length (F1(2,30)= 9.43, p<0.001; F2(2,30)=
10.91, p<0.001). The lines show least-square
fitted lines for each frequency band to make
linear associations easier to visually assess.
There is increasing interest in applying connectivity analysis to brain measures, but most
studies have relied on fMRI, which substantially limits the participant groups and numbers
that can be studied.
In this study, we used source reconstruction with age-matched templates to task-free high-
density EEG recordings in typically-developing children between 2 and 6 years of age. Graph
theory was then applied to the association strengths of 68 cortical regions of interest based
the Desikan-Killiany atlas. We found linear increases of mean node degree, mean clustering
coefﬁcient and maximum betweenness centrality between 2 years and 6 years of age. The
correlation of the network measures with age indicate network development towards more
closely integrated networks, a result similar to those reports using other imaging modalities.
Correlation between age and summed edge
weight of key regions Top Left: Interhemispheric
connections,;Top right: 1-12Hz; Bottom left:
12-25Hz; Bottom right: 25-40Hz
There were no significant correlations between
age and interhemispheric connection strength for
all frequency bands. There was also no
significant correlation between age and edge
weight within the subnetworks identified with
principal network analysis. The lines show least-
square fitted lines for each frequency band to
make linear associations easier to visually
The study aimed to characterise the association between network characteristics and participant age
in children between 2 and 5 years based on source reconstruction with age-matched average
templates (Sanchez et al., 2010) of high-density EEG recordings.
We applied Graph Theory measures to the networks derived from correlations between cortical
regions of interest. The results indicate network development from segregated towards more closely
integrated networks. These findings are in line with the developmental tendencies described for other
imaging modalities (Fair et al., 2008). Different frequency bands displayed similar network
architecture and development. Connection weight within core networks identified through eigenvalue
decomposition (Clayden et al., 2013) was not significantly correlated with age. This suggests that the
structural core of the network remains stable over development.
In conclusion, the current study demonstrates that EEG source reconstruction based on age-
matched average MRI templates can be applied to investigate the development cortical networks in
children. Future studies will be able to exploit the temporal resolution of EEG more fully and
establish relationship between structural and functional networks. Fair, D. A., Cohen, A. L., Dosenbach, N. U. F., Church, J. A., Miezin, F. M., Barch, D. M., et al. (2008).
The maturing architecture of the brain's default network. Proceedings of the National Academy of
Sciences of the United States of America, 105(10), 4028–4032
Sanchez, C. E., Richards, J. E., & Almli, C. R. (2010). Age-specific MRI brain templates for healthy brain
development from 4 to 24 years.
Clayden, J. D., Dayan, M., & Clark, C. A. (2013). Principal networks. PLoS ONE
Network architecture for different age groups in three frequency bands Each node
represents one region of interest in the cortical parcellation. The connection strength between the
nodes is determined by the Pearson correlation coefficient between the reconstructed time series
of the region of interest. The spatial layout was generated through an energy-based algorithm
(ForceAtlas 2). The connection matrices were averaged for all participants within each age group.
The average network were thresholded to the average connectivity within the connection matrix.
Networks are more segregated in younger children,
different frequency bands show similar developmental tendencies
Graph measures correlate with
Connection weight within Principle
Networks does not correlate with age
•Investigation of the relationships between cortical areas is a research area of high interest
•fMRI research in young children is limited, because of practical and ethical concerns
•High-density EEG is easy to apply and is well tolerated in young children
•Previous EEG channel-level analyses of connectivity are strongly influenced by volume conduction
•Channel-level analyses do not provide information about the cortical areas involved
•Source reconstruction with age-matched average templates to allow investigation of the functional
relationships among cortical areas based on high-density EEG recordings in young children with
improved spatial resolution compared to channel-level analyses and source analysis based on
We want to thank Prof. John Richards, University of
South Carolina, for granting us access to the
Neurodevelopmental MRI Database
Graph theory measures
Mean node degree: number of nodes that each node is on average connected with. The higher the
node degree, the higher the local connectivity.
Mean clustering coefficient: number of nodes that each node is connected to divided by the
number of possible connections; indicates the tendency of local
vs global connection
Characteristic path length: typical number of nodes that need
to be traversed from any one node to any other; indicates
efficiency of information transfer
Maximum Betweenness-Centrality: importance of the most
important node in the network; indicates how centralised the