Characterizing edge communities within and between neural networks in fetal brains
Poster No:
1817
Submission Type:
Abstract Submission
Authors:
Lanxin Ji1, Amyn Majbri1, Iris Menu1, Richard Betzel2, Olaf Sporns2, Moriah Thomason1
Institutions:
1NYU School of Medicine, New York, NY, 2Indiana University, Bloomington, IN
First Author:
Co-Author(s):
Introduction:
Traditional brain network analysis largely relies on node-centric functional connectivity (nFC), in which links between regions are examined using static correlation analysis. However, this method overlooks interactions between edges, missing potentially meaningful features. In this study, we employed an edge-centric network model [1, 2], by generating interpretable time series for each edge, in fetal brains. These edge time series enable the estimation of edge-centric functional correlation (eFC), tracking the evolution of communication patterns and assessing simultaneous occurrences in the brain. Unlike nFC, which measures the extent of activity fluctuations between brain regions, eFC unravels co-fluctuations across time, providing moment-by-moment accounts and assessing the similarity between pairs of co-fluctuation time series.
Methods:
Imaging data were obtained from 137 fetuses (57 females) aged 25 to 39 weeks gestation (mean = 31.53 ± 3.69) participating in the Perinatal Imaging of Neural Connectivity study. Functional MRI were acquired using a 3 T Siemens Verio 70 cm open-bore system with an abdominal 4-channel Siemens Flex coil. Two sets of multi echo fMRI data were attained with the following scanning parameters: dataset a) TR = 2000ms; TE = 18, 31.07, 44.14ms (3 echoes); flip angle: 83 degrees; voxel size: 3.5 x 3.5 x 3.5 mm3; dataset b) TR = 2000ms; TE = 18, 34, 50ms (3 echoes); flip-angle: 83 degrees; voxel-size: 3.487 x 3.487 x 3.5 mm3.
Preprocessing began with automatic fetal brain segmentation using deep learning [3], motion estimation, and censoring with FSL [4]. Participants with fewer than 105 low-motion frames were excluded. Subsequent preprocessing steps included optimal combination across echoes, normalization to standard space (GA = 32 weeks), smoothing, Independent Component Analysis (ICA) [5] and CompCor denoising [6], and regression of motion confounds. Gray matter was divided into 197 functional parcels using Pyclustering package, and regional time series were extracted from concatenated data of all subjects.
Calculation of edge-centric functional connectivity (eFC) involved the following steps: 1) z-scoring the time series, 2) calculating the element-wise product of z-scored time series for all pairs of brain regions, and 3) determining the element-wise product between pairs of edge time series, resulting in an eFC matrix. K-means clustering (k = 10) was then applied to eFC, partitioning the eFC network into communities of co-fluctuating edges. The group-representative edge communities, community similarity, and normalized entropy of each network were examined. Edge community similarity represents how similar of the community labels of two regions. Normalized entropy per network is a measure of community overlap. Low entropy indicate that a brain regions' edges are concentrated among a small number of communities, whereas higher values indicate that edges are uniformly distributed over communities.
Preprocessing began with automatic fetal brain segmentation using deep learning [3], motion estimation, and censoring with FSL [4]. Participants with fewer than 105 low-motion frames were excluded. Subsequent preprocessing steps included optimal combination across echoes, normalization to standard space (GA = 32 weeks), smoothing, Independent Component Analysis (ICA) [5] and CompCor denoising [6], and regression of motion confounds. Gray matter was divided into 197 functional parcels using Pyclustering package, and regional time series were extracted from concatenated data of all subjects.
Calculation of edge-centric functional connectivity (eFC) involved the following steps: 1) z-scoring the time series, 2) calculating the element-wise product of z-scored time series for all pairs of brain regions, and 3) determining the element-wise product between pairs of edge time series, resulting in an eFC matrix. K-means clustering (k = 10) was then applied to eFC, partitioning the eFC network into communities of co-fluctuating edges. The group-representative edge communities, community similarity, and normalized entropy of each network were examined. Edge community similarity represents how similar of the community labels of two regions. Normalized entropy per network is a measure of community overlap. Low entropy indicate that a brain regions' edges are concentrated among a small number of communities, whereas higher values indicate that edges are uniformly distributed over communities.
Results:
We demonstrate that clustering eFC yields communities of edges in fetal brains can naturally divide the brain into overlapping clusters. Regions in subcortical networks (network 12 and 13) exhibiting the greatest levels of norm entropy, indicating that they concentrate to a small number of communities, while the orbital medial prefrontal cortex (network 4) contribute to varied edge communities with lowest entropy.
Conclusions:
This is the first study examining the edge-centric functional communities in fetal brains. In future work, we will characterize eFC in fetuses on the individual level, and will examine the developmental and sex effect. An ultimate goal will be leveraging this approach to identify novel biomarkers of disease, characterizing individual variation and mapping the architecture of highly resolved neural circuits.
Lifespan Development:
Normal Brain Development: Fetus to Adolescence
Modeling and Analysis Methods:
Connectivity (eg. functional, effective, structural) 2
fMRI Connectivity and Network Modeling 1
Keywords:
Computational Neuroscience
Data analysis
FUNCTIONAL MRI
1|2Indicates the priority used for review
·Figure 1. (A) Group-representative edge communities; (B) Community templates. (C) Edge community. (D) ROIs and networks. (E) Normalized entropy per network.
Provide references using author date format
2. Jo, Y., et al., The diversity and multiplexity of edge communities within and between brain systems. Cell reports, 2021. 37(7).
3. Rutherford, S., et al., Automated Brain Masking of Fetal Functional MRI with Open Data. Neuroinformatics, 2021: p. 1-13.
4. Jenkinson, M., et al., Fsl. Neuroimage, 2012. 62(2): p. 782-790.
5. Griffanti, L., et al., ICA-based artefact removal and accelerated fMRI acquisition for improved resting state network imaging. Neuroimage, 2014. 95: p. 232-247.
6. Behzadi, Y., et al., A component based noise correction method (CompCor) for BOLD and perfusion based fMRI. Neuroimage, 2007. 37(1): p. 90-101.