CMStatistics 2017: Start Registration
View Submission - CMStatistics
B1409
Title: Estimating dynamic connectivity states in fMRI Authors:  Chee Ming Ting - King Abdullah University of Science and Technology (Saudi Arabia) [presenting]
Hernando Ombao - King Abdullah University of Science and Technology (KAUST) (Saudi Arabia)
Abstract: Sliding-window analysis or time-varying coefficient models which are unable to capture both smooth and abrupt changes simultaneously in dynamic brain connectivity. Emerging evidence also suggests state-related changes in brain connectivity where dependence structure alternates between a finite number of latent states or regimes. Another challenge is to infer full-brain networks with large number of nodes. A novel approach is proposed based on a regime-switching factor model. The dynamic connectivity states in high-dimensional brain signals is characterized via a few common latent factors. By assuming the factor dynamics to follow Markov-switching vector autoregressive (VAR) process, regime-switching in high-dimensional directed dependence between the observations is allowed. This model enables efficient, data-adaptive estimation of change-points of effective brain connectivity regimes and the massive dependencies associated with each regime. We introduce a two-step estimation procedure: (1) Extract the factors by PCA and (2) Estimate the switching models in a state-space form by maximum likelihood method using the Kalman filter the EM algorithm. We also used the wavelet-based functional mixed models to analyze dynamic connectivity states that are varied across multiple subjects or multiple signal replicates. The method is applied in multi-subject resting-state fMRI data where mental activities are unconstrained, and in multi-trial EEGs evoked by repetitive auditory stimuli.