Data-driven Techniques for the Study of Brain Dynamics and Identification of Subgroups: Application to Multi-subject Resting-state fMRI Data
Loading...
Links to Files
Permanent Link
Author/Creator
Author/Creator ORCID
Date
2020-01-20
Type of Work
Department
Computer Science and Electrical Engineering
Program
Engineering, Electrical
Citation of Original Publication
Rights
Distribution Rights granted to UMBC by the author.
This item may be protected under Title 17 of the U.S. Copyright Law. It is made available by UMBC for non-commercial research and education. For permission to publish or reproduce, please see http://aok.lib.umbc.edu/specoll/repro.php or contact Special Collections at speccoll(at)umbc.edu
This item may be protected under Title 17 of the U.S. Copyright Law. It is made available by UMBC for non-commercial research and education. For permission to publish or reproduce, please see http://aok.lib.umbc.edu/specoll/repro.php or contact Special Collections at speccoll(at)umbc.edu
Subjects
Abstract
Functional magnetic resonance imaging (fMRI) captures the blood-oxygen-level-dependent (BOLD) response and has been a valuable tool for understanding human brain function. Data-driven techniques have proven to be very effective in fMRI analysis and identified unique biomedical patterns in neurological disorders such as in schizophrenia (SZ), atrophy, degenerative dementias, Alzheimer's disease, and many others. There are a number of data-driven techniques developed for fMRI analysis to extract functional networks and their associated properties, but there are still a number of important challenges. First, there is a need to better understand the properties of different approaches to be able to select the best method for a given scenario. Then, a big challenge has been capturing the subject variability while simultaneously performing analysis on multi-subject fMRI. This motivates the development of a method that is able to preserve variabilities across datasets, so that one can identify statistically significant groupings of subjects, another major research thrust in medical imaging. Finally, besides investigating static functional patterns, extracting dynamic features to study brain dynamics is becoming important due to the evidence that brain functional patterns exhibit changes during the scanning period of fMRI. Both BOLD activity and functional network connectivity (FNC) are shown to be related to mental and cognitive processes. However most previous dynamic studies only conduct a dynamic FNC (dFNC) analysis and few studies have evaluated the inter-relationships of these two domains of function. It is desirable to incorporate dynamic BOLD activity (dBA) to gain insight into the activity-connectivity co-evolution in the study of brain dynamics. In this dissertations, we address these challenges by working within the source separation umbrella for fMRI analysis. We first demonstrate that jointly incorporating multiple types of diversity is more desirable by proposing the use of objective global metrics to assess the performance of different data-driven algorithms---independent component analysis (ICA), dictionary learning, and sparse version of ICA---that each make use of different types of diversity. Independent vector analysis (IVA) extends ICA to multiple datasets by additionally making use of dependence across datasets, and hence can preserve the correlation structure across datasets but suffers from the dimensionality issue. We develop a new method, IVA for common subspace analysis (IVA-CS) for subspace analysis of multi-subject fMRI by leveraging the strengths of IVA and addressing the dimensionality issue. We show that IVA-CS is able to extract meaningful common and distinct subspaces as well as group-specific neuroimaging features that allow for the identification of significant subgroups of SZ subjects. In order to enable a study of brain dynamics in terms of both dBA and dFNC, we propose a novel use of adaptively constrained IVA (acIVA) to capture activity variabilities and efficiently quantify the spatial property of dBA (sdBA). We first address the challenge in dFNC analysis by proposing a goal-driven scheme to successfully select an optimal value for the number of dFNC states. The efficient quantification of sdBA enables a careful investigation of the association between temporal property of dBA (tdBA) and sdBA, and the activity-connectivity co-evolution of sdBA and dFNC computed using the spatial maps (sdFNC). The application to multi-subject resting-state fMRI data detects significant tdBA-sdBA patterns and activity-connectivity co-evolution patterns. Moreover, we identify significant subgroups of SZs using tdBA-sdBA association and sdBA-sdFNC co-evolution, demonstrating the effectiveness of dynamic features for studying heterogeneity of disorders.