Probabilistic modelling of functional modes in the human brain

Thesis

It is well established that it is possible to observe spontaneous, highly structured fluctuations in human brain activity from functional magnetic resonance imaging (fMRI) data when the subject is 'at rest'. This activity can be decomposed into groups of spatially distributed brain regions that are consistently temporally co- activated, and this thesis is concerned with developing new methods to identify these functional modes.

We introduce a probabilistic model that allows us to infer functional modes without making restrictive assumptions about the spatio-temporal interactions between them, and furthermore, the model also accounts for the variability of these modes over subjects. Both of these facets of the model represent advances compared to current mode-identification techniques. We use a computationally efficient variational Bayesian approach to make inferences from this model, and this allows us to draw upon the enormous amount of information available from the types of large-scale fMRI data collection initiatives that are becoming the norm.

We demonstrate, using data from over 450 subjects collected as part of the Human Connectome Project, that we can reliably infer a set of modes that captures an enormous degree of spatial variability over subjects and that makes a novel set of predictions about the temporal relationships that modes have with one another. Finally, we highlight the importance of functional registration and bring to light a surprising link between temporal non-stationarities and the fMRI global signal.

Authors: Harrison, S

• Type of award: DPhil
• Level of award: Doctoral
• Awarding institution: University of Oxford