Abstract

Neural modeling has long been a tool for understanding the brain's function, in both healthy subjects and those experiencing neural pathologies. These models have captured structures in the brain ranging in spatial scale from the subcellular level to the whole-brain level, and processes ranging in temporal scale from milliseconds to multiple hours. Currently, however, there are very few dynamical systems models of whole-brain activity, and the modeling methodologies that do exist rely on neural data modalities which are impractical for use in clinical settings. At the same time, a wealth of data is being generated by patients undergoing clinical monitoring. Therefore, there is an opportunity for engineering modeling techniques that make use of this data and capture the latent brain dynamics of the person being monitored. A key challenge in creating these modeling techniques is the fact that neural data is nonstationary, varying with time and physiologic state. Indeed, the current state-of-the-art techniques for dynamical systems modeling of the brain are formulated for stationary settings, and cannot accurately capture dynamics which change over time. In this dissertation, I develop data-driven methods for creating individualized models of neural activity which can vary temporally according to neurophysiologic states latent to the neural dynamics. I begin by adapting a current method for modeling whole-brain dynamics for use with electroencephalography (EEG) data, in order to gain mechanistic insights into changes in brain dynamics during events of clinical significance. I then extend this method to capture changing dynamic regimes when the switches in dynamic regime are labeled. This is accomplished by modeling, in essence, the process of neuromodulation. Here, instead of a static connectivity matrix typical of neural network models, I deploy a constant base connectivity matrix multiplied by a switched modulation matrix. Finally, I combine this modeling methodology with a method for blind identification of switches in neural state, in order to describe and infer the changes in dynamic regime present in non-stationary neural data, as well as the dynamics of each individual regime. The methods presented in this dissertation represent an important step forward in modeling and understanding non-stationary brain dynamics on an individual level.

Committee Chair

ShiNung Ching

Committee Members

Andrew Clark; Geoff Goodhill; Rejean Guerriero; Yiannis Kantaros

Degree

Doctor of Philosophy (PhD)

Author's Department

Electrical & Systems Engineering

Author's School

McKelvey School of Engineering

Document Type

Dissertation

Date of Award

8-18-2025

Language

English (en)

Download

Share

COinS