RELATING STRUCTURAL CONNECTIVITY TO BRAIN FUNCTION USING DEEP LEARNING, GRAPH THEORY, COMPLEXITY, AND DISEASE
MetadataShow full item record
In the field of neuroscience, researchers are tasked with the enormous question of how and why this critical organ containing vast numbers of neurons and synapses is able to produce the complex range of behavioural outputs that make up the human experience. In particular, the relationship between structure and function in the human brain is a core question in network neuroscience, and a topic of paramount importance to our ability to meaningfully describe and predict individual functional and behavioural outcomes. In this thesis, I will review the literature investigating important structure-function relationships in the brain, before describing the variety of novel methods and applications that I have used to address the problem of relating structural connectivity to functional connectivity and activity. First of all, I demonstrate that a graph neural network deep learning model is able to use structural connectivity to predict the functional connectivity and centrality (i.e., how connected and important a region is to the network) accounting for more variance than any other previously applied methods in the literature. Next, I examine the relationship between graph theory structural measures of centrality and the functional complexity of the regional activity, and find that regions in the structural network with high centrality that are able to facilitate the integration of information from many sources produce more complex functional activation as measured using the Hurst exponent. Then, by applying graph theory comparative analyses of structural connectivity and functional analysis of language-related activation to patients with left hemisphere temporal lobe epilepsy (TLE), right hemisphere TLE, and control groups, I show that both structural connectivity and functional activity favour the opposite hemisphere to the locus of TLE, suggesting that both structure and function may adapt in tandem as a response to disordered brain activity. Finally, I examine explicit graph theory models of information transfer in the brain to determine which of the diffusion model and shortest path routing model are better able to account for functional connectivity from the underlying structural connectivity, and show that the diffusion model seems to be the primary driver of this relationship. Taken together, this program of research has led the field of network neuroscience in the direction of both setting a clear benchmark for prediction thanks to the novel deep learning model, while also taking important steps to clearly elucidate the explicit mathematical and theoretical core of how structure informs function in the complex network that is the human brain.
DegreeDoctor of Philosophy (Ph.D.)
CommitteePrime, Steven; Farthing, Jon; Gould, Layla; Rayan, Steven
temporal lobe epilepsy