Title:  Approaches to Extract, Characterize, and Interpret Dynamic Functional Network Connectivity in fMRI Data
Committee:Dr. Vince Calhoun, ECE, Chair, Advisor 
Dr. Eva Dyer, BME
Dr. Omer Inan, ECE
Dr. David Anderson, ECE
Dr. Robyn Miller, GSU
Dr. David Salat, Harvard

Abstract: The objective of this project is to develop new approaches for analyzing dynamic functional network connectivity (dFNC) and investigate the link between dFNC with cognitive score and symptom severity in different neurological disorders, including schizophrenia, major depressive disorder, and Alzheimer’s disease. Knowing that the brain is highly dynamic during the resting-state fMRI, even in the absence of external inputs, dFNC got much attention in recent years. However, there are still some gaps in the field. These gaps include the lack of an analytic pipeline analyzing big dFNC data, a pipeline uncovering hidden dynamics masked by the highly influential networks, a comprehensive toolbox extracting dFNC features, and a lack of understanding of the clinical benefit of dFNC results. In this Ph.D. proposal, we aim to address these potential gaps. This Ph.D. dissertation contributed to the field by developing new frameworks (methodological contributions) and identifying new biomarkers in brain disorders (clinical contributions). We proposed multiple frameworks to analyze static (sFNC) and dynamic functional network connectivity (dFNC) for methodological contributions. We developed a framework called iSparse k-means to analyze big dFNC data. We showed that this framework analyzes dFNC data 27 times faster than the conventional framework, but it does not need huge computational power. We also developed an analytic pipeline (toolbox), called “DyConX”, to estimate transient states and extract temporal features from dFNC. Also, we introduced some additional summary metrics to characterize dFNC. We validated these new features with the new toolbox in the largest dFNC analysis ever. We developed a new dFNC pipeline to uncover some hidden network dynamics masked by highly influential brain networks. Next, we proposed integrating our pipeline with an interpretable machine learning method to investigate the use of dynamic features to be useful as predictors (or biomarkers). For the clinical contribution, we identified new dFNC biomarkers in Alzheimer's disease, schizophrenia, and major depressive disorder. We interpreted how dFNC information is linked with symptom severity in these neurological and neuropsychiatric disorders and what the clinical could be benefits of this dFNC information.