Title: “Modeling Temporal Dynamics in Functional Brain Connectivity”
Thursday, 18 October, at 14:00, The Technical University of Denmark, Building 324, room 040
Supervisors:
Principal supervisor: Associate Professor Morten Mørup
Co supervisor: Associate Professor Mikkel N. Schmidt
Co supervisor: Associate Professor Kristoffer H. Madsen
Examiners:
(Chairman) Associate Professor Finn Årup Nielsen, DTU Compute
Professor Tom Heskes, Radboud University Nijmegen, Netherlands
Adjunct Professor Tom Eichele, Haukeland University Hospital, Bergen
Chairperson at defence:
Associate Professor Tobias Andersen, DTU Compute
Summary:
Research dealing with the mapping of cognitive functions to areas in the human brain has flourished since the invention of functional magnetic resonance imaging (fMRI) in the early 1990s. From initial discovery of patterns of activations more and more studies today focus on quantifying the strength of associations between brain regions. This is accomplished using the temporal similarity of the fMRI signal, denoted functional connectivity (FC). Recent studies have suggested that FC fluctuates within a scanning session, termed dynamic functional connectivity, even without an explicit task being performed (resting-state). These fluctuations can be divided into a number of reoccurring connectivity profiles, denoted functional connectivity states. In this rather new area of research, there is currently a lot of debate about the validity of these states including the procedures used for their estimation. This thesis investigates the influence of modeling assumptions for the extraction of dynamic functional connectivity (dFC) states using Bayesian probabilistic machine learning methods. Treating dFC in a fully probabilistic way allows for quantification of model uncertainty and model generalization making a quantitative comparison possible between different dFC models. This has been illustrated in three research papers in which different modeling assumptions of the dynamic changes were compared. In two additional research papers we further investigated dFC models’ abilities to explain external subject-specific variables. We found that there was a low association between the dFC temporal features extracted from healthy subjects resting-state fMRI data and higher order cognitive traits. Finally, we showed in a clinical population containing patients diagnosed with schizophrenia (SZ) and healthy controls (HC) that a model using a single FC state from resting-state fMRI was adequate to predict the disease status of a subject. Our findings suggest that quantitative comparisons are important in order to understand the influence and impact of modeling assumptions for the characterization of the brain’s (dynamic) functional organization.
READ MORE in DTU Orbit
A copy of the PhD thesis is available for reading at the department