CMStatistics 2020: Start Registration
View Submission - CMStatistics
Title: Time-varying dynamic network model for extracting the dynamic resting-state functional connectivity Authors:  Fei Jiang - The University of California, San Francisco (United States) [presenting]
Abstract: Dynamic resting-state functional connectivity (RSFC) is believed to reflect the intrinsic organization and network structure of brain regions. The existing methods to extract dynamic RSFCs do not adapt to different datasets. Furthermore, they are not suitable for multi-modality studies. Moreover, it is difficult to justify that the resulting dynamic RSFCs are the intrinsic features that generate the brain signals. To overcome these deficiencies, we develop a time-varying dynamic network (TVDN) framework to extract the resting-state functional connectivity from neuroimaging data. TVDN has a fully automatic parameter turning mechanism, and hence it is adaptive to different datasets. Furthermore, TVDN is easily generalizable to handle the multi-modality data. Moreover, TVDN describes the relation between RSFC and brain signals. Hence, it is easy to evaluate the method by examining whether the resulting features can reconstruct the observations. We develop the TVDN model and the estimation procedures. Furthermore, we conduct comprehensive simulations to evaluate TVDN under hypothetical settings. Finally, we apply the TVDN on both fMRI and MEG data and compare the results with the existing approaches. The results show that the TVDN is more robust to detect brain state switching in the resting state. In addition, the resulting dynamic RSFCs directly link to the signal frequency and growth/decay constant and can uncover the noiseless brain signals.