Investigating the Brain States Behind FMRI Temporal Dynamics Using Frame-based Analysis Methods and Variational Autoencoders
Author(s)
Zhang, Xiaodi
Advisor(s)
Editor(s)
Collections
Supplementary to:
Permanent Link
Abstract
Functional MRI (fMRI) plays an important role in studying the brain functionality. In the past decade various resting state fMRI analysis methods have been proposed to study the dynamics in the functional connectivity of the brain, including sliding window correlation, point process analysis, co-activation patterns, hidden Markov model and quasi-periodic patterns. These methods provide valuable insights into the dynamic aspects of functional connectivity, which significantly contribute to the field of resting state fMRI (rs-fMRI), or fMRI in general.
However, there are two problems in the current methods that impede our further understanding of how brain operates. First the blood oxygenation level-dependent (BOLD) signal used in fMRI studies is only an indirect measurement of neural activity through neurovascular coupling. The lack of ground truth makes it difficult to validate the dynamics originating from fluctuations of neural activity instead of noise. Secondly the brain activity has often been modeled as a discretized hopping among several brain states, whereas in reality it is likely to be a continuous process.
Here in this thesis work we have developed two methods to address these issues. In order to address the first problem. we used concurrent local field potential (LFP) and fMRI measurements on rodents to provide a direct measurement of neural activity. This method provides electrophysiological evidence that validates the existence of time-varying neural activity. To address the second problem, we developed a variational autoencoder to model the continuous trajectories of brain activity in healthy human brain. Our results suggest that there are a relatively small number of approximately orthogonal whole-brain spatiotemporal patterns that capture the most prominent features of rs-fMRI data. These spatiotemporal patterns show stereotypical temporal dynamics along different brain regions, which provide insight into how activity flows across the brain in concordance with known network structures and functional connectivity gradients.
Sponsor
Date
2021-04-07
Extent
Resource Type
Text
Resource Subtype
Dissertation