Name: Amrit Kashyap 

BME PhD Thesis Defense: Understanding Whole Brain Activity through Brain Network Models 

Date: Nov 4th 2020 

Time: 12-1pm 

Link: https://gatech.bluejeans.com/8078952945  

Meeting Info: No Password  

Advisor: 
 

Dr. Shella Keilholz 

Department of Biomedical Engineering 

Georgia Institute of Technology and Emory University 

Committee Members: 

Dr. Lena Ting 

Department of Biomedical Engineering 

Georgia Institute of Technology and Emory University 

Dr. Christopher Rozell 

Department of Electrical Engineering 

Georgia Institute of Technology 

Dr. Madeleine Hackney 

Department of Medicine 

Emory University 

Dr. Bruce Crosson 

Department of Neurology 

Emory University 

Dr. Sergey Plis 

School of Computer Science 

Georgia State University 

Title: Understanding Whole Brain Activity through Brain Network Models 

Abstract:  

The central nervous system coordinates many neural subpopulations connected via  macroscale white matter architecture and surface cortical connections in order to produce complex behavior depending on environmental cues. The activity occurs over many different scales, from the information transfer between individual neurons at the synapse level, to macroscale coordination of neural populations used to maximize information transfer between specialized brain regions. The whole brain activity measured through functional Magnetic Resonance Imaging (fMRI), allows us to observe how these large neural populations over time. Researchers have recently developed a set of Brain Network models (BNMs), that have simulated brain activity using the macroscale white matter structure and different models for activity in local neural populations. These simulations have been able to reproduce properties of brain signals measured via fMRI especially those averaged over long periods of time. This has generated a lot of interest, because these models can be constructed from individual estimates of the structural network and are perhaps a step towards an individualized models of brain activity and would have useful clinical implications. To find a good BNM to fit the individual fMRI data, however, is a difficult problem as BNMs represent a large family of mathematical models. Moreover, a large set of BNMs have reproduced time averaged metrics that have been used so far to compare the models with the fMRI data. In this thesis, we extend previous work on BNM research by establishing new dynamic metrics that would allow us to better differentiate between BNMs simulations on how well they reproduce measured fMRI dynamics (Chapter 2). In Chapter 3, we directly compare transient short-term trajectories by synchronizing the outputs of a BNM in relation to observed fMRI timeseries using a novel Machine Learning Algorithm, Neural Ordinary Differential Equations (ODE). Finally, we show that the Neural ODE can be used as its own stand-alone generative model and is able to simulate the most realistic fMRI signals so far (Chapter 4). In short, we demonstrate that we have made progress in developing and quantifying BNMs and advanced the research of more realistic whole brain simulations.