Nonparametric Bayesian Inference of Dynamic Functional Connectivity
Keywords
No Thumbnail Available
Authors
Issue Date
2017-09-07
Language
en
Document type
Journal Title
Journal ISSN
Volume Title
Publisher
Title
ISSN
Volume
Issue
Startpage
Endpage
DOI
Abstract
Neuroimaging techniques reveal the human brain exhibiting a great repertoire of activity
patterns—not only while perceiving external stimuli or acting in an experimental
paradigm, but also at times of rest. In connectomic research, the time-average functional
interactions of large-scale brain network components is commonly summarized in
the measure of functional connectivity. Recently, there has been a surge of interest in
considering functional connectivity at rest no longer as time-invariant, but as a characterization
of brain networks displaying qualitatively differentiable dynamics over time.
Previous research on dynamical functional connectivity provides a glimpse into the rich
world of macroscopic brain dynamics, yet most of this work relies on heuristics or ad-hoc
assumptions, which are not a priori given and often part of ongoing research themselves.
To circumvent such issues, this thesis investigates functional connectivity in functional
magnetic resonance imaging recordings with the aid of a hierarchical model in a Bayesian
framework, leveraging recent developments in nonparametric modeling. The parameters
of interest are inferred via an efficient Gibbs sampling scheme. The nonparametric model
is evaluated and benchmarked against a sliding-window based K-means clustering on synthetic
resting state data as well as on empirical motor-task and resting state recordings.
The outcome of the analysis is that the proposed model generally infers more states of dynamic
functional connectivity than do actually exist, deteriorating its otherwise superior
predictive validity. Moreover, it is demonstrated that blind deconvolution of the functional
magnetic resonance imaging data may be beneficial for the inference of dynamic
functional connectivity. The findings reported in this thesis highlight the potential and
challenges of nonparametric models in exploratory research of the human connectome.
Description
Citation
Supervisor
Faculty
Faculteit der Sociale Wetenschappen