Date of Award

Winter 2013

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Biomedical Engineering

First Advisor

Alan W. L. Chiu

Abstract

Multi-electrode recording is a key technology that allows the brain mechanisms of decision making, cognition, and their breakdown in diseases to be studied from a network perspective. As the hypotheses concerning the role of neural interactions in cognitive paradigms become increasingly more elaborate, the ability to evaluate the direction of neural interactions in neural networks holds the key to distinguishing their functional significance.

Granger Causality (GC) is used to detect the directional influence of signals between multiple locations. To extract the nonlinear directional flow, GC was completed through a nonlinear predictive approach using radial basis functions (RBF). Furthermore, to obtain the directional relations in the frequency domain, which is lost at the expense of a relatively accurate overall GC estimate, we investigated how the nonlinear GC influences in different frequency bands can be extracted by the proper linearization process. When the error between the nonlinear fitting signal and the linear fitting signal falls below a specific threshold, the frequency components can be approximated. Also, the advantage of this method is that this frequency decomposition model does not rely on the formation of a nonlinear process. We applied the proposed strategy to a brain computer interface (BCI) application to decode the different intended arm reaching movement (left, right and forward) using 128 surface electroencephalography (EEG) electrodes. A threshold method was set up through a spatial statistical process where only the strongest 20% of causality pathways is shown, where the directions of causal influence of active brain regions were found to be unique with respect to the intended direction. The left and right motor intention directions were found to be highly separable in the theta rhythm (3-8 Hz) only.

Share

COinS