Graph theory to study complex networks in the brain
Item Usage Stats
MetadataShow full item record
Please cite this item using this persistent URLhttp://hdl.handle.net/11693/46935
The brain is a large-scale, intricate web of neurons, known as the connectome. By representing the brain as a network i.e. a set of nodes connected by edges, one can study its organization by using concepts from graph theory to evaluate various measures. We have developed BRAPH - BRain Analysis using graPH theory, a MatLab, object-oriented freeware that facilitates the connectivity analysis of brain networks. BRAPH provides user-friendly interfaces that guide the user through the various steps of the connectivity analysis, such as, calculating adjacency matrices, evaluating global and local measures, performing group comparisons by non-parametric permutations and assessing the communities in a network. To demonstrate its capabilities, we performed connectivity analyses of structural and functional data in two separate studies. Furthermore, using graph theory, we showed that structural magnetic resonance imaging (MRI) undirected networks of stable mild cognitive impairment (sMCI) subjects, late MCI converters (lMCIc), early MCI converters (eMCIc), and Alzheimer’s Disease (AD) patients show abnormal organization. This is indicated, at global level, by decreases in clustering and transitivity accompanied by increases in path length and modularity and, at nodal level, by changes in nodal clustering and closeness centrality in patient groups when compared to controls. In samples that do not exhibit differences in the undirected analysis, we propose the usage of directed networks to assess any topological changes due to a neurodegenerative disease. We demonstrate that such changes can be identified in Alzheimer’s and Parkinson’s patients by using directed networks built by delayed correlation coefficients. Finally, we put forward a method that improves the reconstruction of the brain connectome by utilizing the delays in the dynamic behavior of the neurons. We show that this delayed correlation method correctly identifies 70% to 80% of the real connections in simulated networks and performs well in the identification of their global and nodal properties.