A Study of Dynamic Functional Brain Connectivity Using Functional Magnetic Resonance Imaging (Fmri): Method and Applications

This dissertation, "A Study of Dynamic Functional Brain Connectivity Using Functional Magnetic Resonance Imaging (fMRI): Method and Applications" by Zening, Fu, 傅泽宁, was obtained from The University of Hong Kong (Pokfulam, Hong Kong) and is being sold pursuant to Creative Commons: Attribution 3.0 Hong Kong License. The content of this dissertation has not been altered in any way. We have altered the formatting in order to facilitate the ease of printing and reading of the dissertation. All rights not granted by the above license are retained by the author. Abstract: Identifying the statistical interdependence (functional connectivity, FC) between brain regions using functional magnetic resonance imaging (fMRI)is an important approach towards understanding how brain system is organized. Most fMRI studies assumed temporal stationarity of FC, so that the dynamic fluctuations of FC were overlooked. Emerging evidence has shown that FC fluctuates significantly across time and such fluctuations are physiologically relevant. The objectives of this work were (1) to develop novel methods for estimating dynamic FC from non-stationary fMRI signals, and (2) to apply new methods on real-life fMRI datasets for exploring dynamic patterns of FC in tasks and at rest. In particular, new methods were introduced to tackle two key issues in dynamic FC estimation: how to adaptively select window size to estimate dynamic FC and how to estimate dynamic FC networks with sparse architecture and sparse evolution. Firstly, a local polynomial regression (LPR) method was introduced to estimate time-varying covariance (TVCOV) for the inference of dynamic FC. The asymptotic analysis of this covariance estimator was performed and then a data-driven method, intersection of confidence intervals (ICI), was adopted to adaptively determine the window size. Simulation results showed that the LPR-ICI method could achieve robust and reliable performance in estimating TVCOV, making it a powerful tool for studying the dynamic FC from fMRI signals. Secondly, the LPR-ICI method was applied to a visual task fMRI dataset for studying the changes of FC in a block-designed visual checkerboard experiment. Reliable task-related FC changes were identified among activated visual regions during the task block. The results suggested that characterizing the task-related FC dynamics might provide greater insight into condition shifts and coordination between brain regions. Thirdly, the LPR-ICI method was applied to a resting-state fMRI dataset for exploring FC dynamics across the whole brain and investigating their relationships with dynamics of local brain activities. Converging results demonstrated that resting-state FC exhibited remarkable different dynamic patterns across the brain and these dynamic patterns were significantly correlated with the dynamic patterns of brain activities. These findings suggested that the brain might bean adaptive network, in which brain activities and their FC coevolve across time. Lastly, a novel dual l0-penalized (DLP) time-varying in verse covariance estimation method was introduced for estimating sparse dynamic FC networks. This DLP method was able to estimate dynamic networks with sparse architecture and sparse evolution by minimizing a log-likelihood function regularized by two l0-penalties (to enforce sparse architecture and sparse evolution, respectively).A coordinate descent algorithm was developed for searching the local minimizers of the objective function. Extensive simulation results showed that the DLP method could achieve better performance than conventionall1-penalized methods. In summary, two newly-developed methods (LPR-ICI and DLP) could be effective tools for studying dynamic brain FC and our results have advanced the knowledge of how brain regions dynamically coordinate. This study was also clinically relevant, as the quantification of altered FC dynamics in clinical populations of neuropsyc