Modularity of Human Brain Asymmetric Network Revealed by 40,000 Healthy Individuals from sMRI
Poster No:
1587
Submission Type:
Abstract Submission
Authors:
Sifeng Wang1, Suyu Zhong1
Institutions:
1Beijing University of Posts and Telecommunications, Beijing, Beijing
First Author:
Co-Author:
Introduction:
Brain asymmetry has been a predominant perspective to unravel the mapping from structural morphology to functions and complex behaviors of human. Great efforts have been made depicting the wiring between brain structure and function where network analysis approaches are widely used to measure similarity and discrepancy between structural and functional networks[4]. Following the footsteps, we construct highly modulated networks of brain asymmetry index derived from regional morphology including regional area and grey matter volume. Especially, a consistent module decomposition across gender subgroups is observed in the regional grey matter volume network, demonstrating the potential of asymmetrical networks to bridge brain structure and function.
Methods:
Brain imaging phenotypes of 42800 (Female: 22584) and 43063 (Female: 22684) cross-sectional T1-weighted MRIs in UK Biobank dataset[5] were selected to construct asymmetrical networks on regional grey matter volume (rGMV, 65 paired regions) and white surface area (rWSA, 74 paired regions) respectively. Age of subjects were ranged from 44 to 86.
Brain asymmetric networks were defined based on correlation matrices across all cerebral regional Asymmetry Index (AI) in corresponding parcellations. A linear regression was performed to alleviate impact of age and gender before calculating AI following a simple definition: AI(X)=(L(X)-R(X))/2(L(X)+R(X))[6]. Empirical interregional correlations were then calculated on normalized AI[2]. Pearson's R with FDR correction are applied to reject insignificant connections. Additional gender-specific correlations were constructed independently on gender subsets following same steps as above, except that linear regression was only performed on age.
Regional pairs survived from significance test in each dataset formed a weighted undirected graph. That is, brain regions were treated as nodes and correlations as edges. Modularity Q is optimized with greedy algorithm by maintaining a matrix of ∆Q proposed by Clauset[1]. We used Clauset's greedy-based algorithm encapsulated in Gretna toolbox[7] (v2.0.0) to perform modularity analysis of our lateralization networks. Significance of network modularity were then tested with 1,000 randomly generated networks with the same number of nodes and degree distribution as the brain network to compare with.
Brain asymmetric networks were defined based on correlation matrices across all cerebral regional Asymmetry Index (AI) in corresponding parcellations. A linear regression was performed to alleviate impact of age and gender before calculating AI following a simple definition: AI(X)=(L(X)-R(X))/2(L(X)+R(X))[6]. Empirical interregional correlations were then calculated on normalized AI[2]. Pearson's R with FDR correction are applied to reject insignificant connections. Additional gender-specific correlations were constructed independently on gender subsets following same steps as above, except that linear regression was only performed on age.
Regional pairs survived from significance test in each dataset formed a weighted undirected graph. That is, brain regions were treated as nodes and correlations as edges. Modularity Q is optimized with greedy algorithm by maintaining a matrix of ∆Q proposed by Clauset[1]. We used Clauset's greedy-based algorithm encapsulated in Gretna toolbox[7] (v2.0.0) to perform modularity analysis of our lateralization networks. Significance of network modularity were then tested with 1,000 randomly generated networks with the same number of nodes and degree distribution as the brain network to compare with.
Results:
rGMV asymmetrical networks showed strong modularity in overall dataset (Q=0.27, z=7.95, p<0.001) and gender-specific subsets (Male: Q=0.28, z=12.56, p<0.001, Female: Q=0.28, z=12.24, p<0.001), with a consistent five-module composition across gender (Fig. 1). Five nonoverlapping modules were characterized by subcortical necli and cerebellum subdivisions, temporal regions, frontal regions, and parieto-occipital regions respectively.
rWSA asymmetrical networks exhibited even stronger modularity compared to rGMV in both overall dataset (Q=0.32, z=13.97, p<0.001) and gender-specific subsets (Male: Q=0.34, z=23.01, p<0.001, Female: Q=0.33, z=19.84, p<0.001). However, slight permutations in module composition were observed that causes significant difference in modularity analysis results (Number of modules. Overall: 4, Male: 4, Female: 5). The majorities in modules were central and parietal regions, frontal regions and occipito-temporal regions respectively (Fig. 2).
rWSA asymmetrical networks exhibited even stronger modularity compared to rGMV in both overall dataset (Q=0.32, z=13.97, p<0.001) and gender-specific subsets (Male: Q=0.34, z=23.01, p<0.001, Female: Q=0.33, z=19.84, p<0.001). However, slight permutations in module composition were observed that causes significant difference in modularity analysis results (Number of modules. Overall: 4, Male: 4, Female: 5). The majorities in modules were central and parietal regions, frontal regions and occipito-temporal regions respectively (Fig. 2).
·Asymmetry Index Network of rGMV. (A) Female subgroup network. (B) Overall network. (C) Male subgroup network. (D) Overall network with region names. A consistent five-module was observed in rGMV Asymm
·Asymmetry Index Network of rWSA. (A) Female subgroup network. (B) Overall network. (C) Male subgroup network. (D) Overall network with region names. A four-module partitioning of overall and male-spec
Conclusions:
We provided a brand new perspective of analyzing human brain connectivity simply by constructing a region-wise cerebral network of Asymmetry Index. Asymmetrical networks derived from regional grey matter volume and white surface area demonstrate outstanding modularity properties. Networks of regional grey matter volume were perfectly aligned across gender, while some of the modules of in white surface area network altered across gender. Future studies are needed to testify genetic and behavioral associations of individual asymmetrical networks.
Modeling and Analysis Methods:
Connectivity (eg. functional, effective, structural) 1
Neuroanatomy, Physiology, Metabolism and Neurotransmission:
Anatomy and Functional Systems 2
Keywords:
Computational Neuroscience
Cortex
Data analysis
Design and Analysis
Modeling
MRI
STRUCTURAL MRI
Other - Asymmetrical Connectivity
1|2Indicates the priority used for review
Provide references using author date format
[2] He, Y., Chen, Z.J. and Evans, A.C. (2007) ‘Small-World Anatomical Networks in the Human Brain Revealed by Cortical Thickness from MRI’, Cerebral Cortex, 17(10), pp. 2407–2419.
[3] Newman, M.E.J. (2006) ‘Modularity and community structure in networks’, Proceedings of the National Academy of Sciences, 103(23), pp. 8577–8582.
[4] Park, H.-J. and Friston, K. (2013) ‘Structural and Functional Brain Networks: From Connections to Cognition’, Science, 342(6158), p. 1238411.
[5] Sudlow, C. et al. (2015) ‘UK Biobank: An Open Access Resource for Identifying the Causes of a Wide Range of Complex Diseases of Middle and Old Age’, PLOS Medicine, 12(3), p. e1001779.
[6] Toga, A.W. and Thompson, P.M. (2003) ‘Mapping brain asymmetry’, Nature Reviews Neuroscience, 4(1), pp. 37–48.
[7] Wang, J. et al. (2015) ‘GRETNA: a graph theoretical network analysis toolbox for imaging connectomics’, Frontiers in Human Neuroscience, 9.