330
V. Vuksanovi´ c
through structural white matter connections and through coherent activity [30, 58],
creating complex groups of interconnected functional units. The connectivity architecture of these units exhibits an extraordinary level of complexity, whose properties
can be analysed across multiple scales—spatial, temporal or topological. To address
these different levels of complexity, significant attention over the past decade has
focused on mapping the large-scale networks of the human brain extracted from
brain scans using Magnetic Resonance Imaging (MRI) [42, 63, 64]. The aim is to
provide a picture of the brain and its connections at the system level.
A common simplified form for brain networks maps is a graph, in which brain
regions (nodes) are linked to one another by network connections (edges) [13, 14].
The definition of a node or an edge is of critical importance to the relevance of the
resulting brain network models [4, 17, 84]. Inspired by neuroanatomy, the definition
of nodes and edges is commonly inferred from diffusion, structural or functional
MRIs [17]. For example, nodes are defined by Brodmann areas [37], gross anatomical
landmarks [25, 69], or increased functional activation [34]. Likewise, the number of
streamlines identified between MRI voxels via diffusion of water along the axons
[35], coherent/synchronized activity between voxels time series [55] or correlated
morphological characteristics [1] are defined as network edges.
Another level of brain network complexity is the arrangement of nodes and edges,
which defines network topology. Evidence has accumulated that large-scale brain
networks are characterized by modular topology. This means that they contain communities – groups of nodes that are more densely connected to members of their own
group than to members of other groups [65]. Modular architecture, with anatomically segregated and functionally specialised communities, is potentially naturally
selected because it reduces metabolic costs [56]. From the graph theory perspective,
these networks are preferred since they reduce the wiring cost (the average length
and number of connections), which enables more efficient information processing
[65]. Moreover, recent findings demonstrate that functional networks are enabled
not only by critical modular interactions between brain areas, but also by swiftly
reconfiguring patterns of these interactions [16, 43, 68]. Whether the subject is at
rest [41], or performing either cognitively demanding or simplistic task, the patterns
of functional connections between brain areas change, revealing mutli-layered community structures in time-varying brain activity. Time-varying dynamics of these
networks accompany neurological disorders [45], brain injury [47], and psychiatric
disease [18, 82].
The estimation of brain structural and functional connections is confounded by
experimental limitations of MRI techniques. For example, limitations of diffusion
MRIs to accurately reconstruct crossing-fibers within white matter is well documented. More importantly, diffusion MRI, which is predominately used as a surrogate for structural brain connectivity (i.e., physical links between the nodes based
on white-matter fiber tracking), lacks tools for reconstruction of axonal connections
within gray matter [57]. Given that functional connectivity maps gray matter networks, there is a growing interest in anatomical MRI, (i.e., 3D T1-weighted images)
and gross morphological features that can be extracted from both gray and white
V. Vuksanovi´ c
through structural white matter connections and through coherent activity [30, 58],
creating complex groups of interconnected functional units. The connectivity architecture of these units exhibits an extraordinary level of complexity, whose properties
can be analysed across multiple scales—spatial, temporal or topological. To address
these different levels of complexity, significant attention over the past decade has
focused on mapping the large-scale networks of the human brain extracted from
brain scans using Magnetic Resonance Imaging (MRI) [42, 63, 64]. The aim is to
provide a picture of the brain and its connections at the system level.
A common simplified form for brain networks maps is a graph, in which brain
regions (nodes) are linked to one another by network connections (edges) [13, 14].
The definition of a node or an edge is of critical importance to the relevance of the
resulting brain network models [4, 17, 84]. Inspired by neuroanatomy, the definition
of nodes and edges is commonly inferred from diffusion, structural or functional
MRIs [17]. For example, nodes are defined by Brodmann areas [37], gross anatomical
landmarks [25, 69], or increased functional activation [34]. Likewise, the number of
streamlines identified between MRI voxels via diffusion of water along the axons
[35], coherent/synchronized activity between voxels time series [55] or correlated
morphological characteristics [1] are defined as network edges.
Another level of brain network complexity is the arrangement of nodes and edges,
which defines network topology. Evidence has accumulated that large-scale brain
networks are characterized by modular topology. This means that they contain communities – groups of nodes that are more densely connected to members of their own
group than to members of other groups [65]. Modular architecture, with anatomically segregated and functionally specialised communities, is potentially naturally
selected because it reduces metabolic costs [56]. From the graph theory perspective,
these networks are preferred since they reduce the wiring cost (the average length
and number of connections), which enables more efficient information processing
[65]. Moreover, recent findings demonstrate that functional networks are enabled
not only by critical modular interactions between brain areas, but also by swiftly
reconfiguring patterns of these interactions [16, 43, 68]. Whether the subject is at
rest [41], or performing either cognitively demanding or simplistic task, the patterns
of functional connections between brain areas change, revealing mutli-layered community structures in time-varying brain activity. Time-varying dynamics of these
networks accompany neurological disorders [45], brain injury [47], and psychiatric
disease [18, 82].
The estimation of brain structural and functional connections is confounded by
experimental limitations of MRI techniques. For example, limitations of diffusion
MRIs to accurately reconstruct crossing-fibers within white matter is well documented. More importantly, diffusion MRI, which is predominately used as a surrogate for structural brain connectivity (i.e., physical links between the nodes based
on white-matter fiber tracking), lacks tools for reconstruction of axonal connections
within gray matter [57]. Given that functional connectivity maps gray matter networks, there is a growing interest in anatomical MRI, (i.e., 3D T1-weighted images)
and gross morphological features that can be extracted from both gray and white