CONNECTOME EIGENMODES AND THEIR APPLICATIONS

The 21st century has marked an explosion in our understanding of the organization and dynamics of the human brain, largely thanks to the wide adoption of non-invasive magnetic resonance imaging (MRI) and exponentially increasing computational resources to analyze large amounts of data. The burgeoning field of network neuroscience seeks to document the properties of brain networks and relate them to the computations that underlie cognition. Still, a comprehensive understanding of the relationship between the brain’s connectivity and the cognition, behavior, and experience it supports is missing. The transfer of analysis techniques from network science/ graph theory and physics in combination with large, high-quality MR datasets offers a lens through which brain networks can be more richly characterized. This work treats brain networks as graphs and uses a physically motivated framework to compute the global organizational axes of brain networks. By computing graph Laplacian eigenmodes of large and complex brain networks, I both reduce the dimensionality of these networks and recover their most salient intrinsic geometric features. Analogous to how the Fourier basis provides a concise and informative basis to describe signals and geometry in 1 dimension, brain network eigenmodes characterize the steady-state vibrational modes of brain network graphs. First, this paradigm is deployed on structural brain networks estimated via diffusion tractography, and derived eigenmodes are treated as a feature space that situates brain regions according to their structural connectivity profiles. Parcellations, or divisions of the cortex into discrete regions, are then computed based on this feature space using hierarchical clustering. Computed parcellations bear a marked similarity to parcellations derived from functional connectivity data. Next, I formulate a novel method to decode the functional dynamics of brain networks, introduced in a framework called “Dynamic Functional Connectome Harmonics.” Here, I compute time-varying functional connectivity graphs and, from them, compute a time series of functional eigenmodes. These eigenmodes are demonstrated to be stable with respect to the timescale at which they are computed, shedding light on the mechanisms by which they are generated. Dynamic functional connectome harmonics are then used to formulate an elegant and succinct definition of cortical flexibility which is shown to yield results similar to prior methods with lower computational cost and fewer free parameters. Finally, I study the evolution of the hierarchical organization of functional brain networks across the human lifespan from just days after birth to 100 years of age using connectome eigenmodes. I establish a normative growth chart of the principal organizational axes of functional connectivity and demonstrate that they undergo nonlinear, multifaceted developmental trajectories. Each such axis corresponds to a well-studied information processing hierarchy crucial to the texture of human cognition. I document the establishment, refinement, expansion, and eventual contraction of these hierarches across the human lifespan from cradle to grave.