|
Abstract:
|
We present approaches to modeling and estimating brain connectivity in real time, focusing on brain signals with high temporal resolution, e.g., electroencephalograms (EEG) and local field potentials (LFP). Here, connectivity will be characterized in terms of dependence between the oscillatory behavior of pairs of signals. The measures will be considered are coherence, partial coherence and partial directed coherence. As these EEG and LFP data are being recorded, connectivity will be computed using recursive least-squares based algorithms. We will explore a time-localized version of the Kalman smoother to update connectivity estimates (rather than updating retrospectively after all the data has been recorded). One of the challenges to this project is to fully implement these measures under the high-dimensional setting. Here, we will explore some biologically-guided dimension reduction techniques and propose multi-state models under which we can formally define non-linear dependence and rigorously conduct statistical inference. The proposed methods and software will be applied to human electroencephalograms recorded during resting state and working memory experiment.
|
