|
We develop a novel approach to modeling connectivity in high dimensional brain signals. In the approach, we model the cortical activity using linear mixture of latent factor activities that follows a vector autoregressive (VAR) process. The frequency-specific connectivity on the cortical surface can be characterized by the latent factor activity and its loading matrix.
The primary motivation for this work is modeling connectivity among regions on the cortical surface using multi-channel scalp electroencephalograms (EEG). Modeling connectivity between brain regions is difficult under high dimensionality of the anatomical parcellation on the cortical surface.
We present a modeling procedure that addresses a number of challenges in high dimensional brain signals. In the first step, we estimate the sources using imaging method with anatomical constraints. In the second step, to estimate temporal dependency between regions on the cortex, we fit a latent process with a vector autoregressive (VAR) structure. From the VAR parameters, we produce different measures of cortical connectivity. We apply this new approach to modeling EEG data during resting state and task.
|
