UC Irvine

Dynamic Functional Connectivity (DFC) investigates how the interactions among brain regions vary over the course of an fMRI experiment. A challenge in DFC inference is the between-subject heterogeneity in both brain region interactions and the evolution of these interactions over time. In this dissertation, I discuss three proposed state-space models to address these challenges. (1) I propose a multi-subject DFC model where transitions between functional connectivity (FC) states are modeled by a non-homogenous Hidden Markov Model: a state-space model where the transition distribution are informed by subject-specific physiological readings. I discuss the results of this model applied to fMRI data where pupil dilation is simultaneously tracked, suggesting FC stability is linked to attention and arousal. (2) I propose an extension to the single subject Psychophysiological Interaction (PPI) model, which seeks to estimate the effect modification of FC by a stimulus/task. The standard PPI model assumes static FC when the subject is at rest. The proposed model allows for DFC in absence of a stimuli, reducing bias in estimating PPI effects in DFC settings. This model is applied to a predictive learning task, where increased unpredictability of a sequence of images is estimated to be associated to tighter coupling in Anterior Cingulate and Caudate regions. (3) Lastly, I propose a multi-subject bi-clustering model that dynamically clusters brain regions based on their activation profile, then clusters subjects based on their regional clustering behavior over time. This model is applied to a \textit{handgrip} task-fMRI dataset.