Please use this identifier to cite or link to this item:
|Title:||Nonlinear Analyses of Functional MRI Time-Series in Brain-Based Disorders|
|Authors:||Mujica-Parodi, Lilianne R|
Department of Biomedical Engineering.
|Abstract:||Functional magnetic resonance imaging (fMRI) is a non-invasive neuroimaging modality used to measure brain activity in vivo, capable of providing three-dimensional coverage of the brain at a high spatial resolution. While fMRI has advanced our understanding of the brain, it has had limited applications in medical practice due to 1) analytic approaches predominantly focused on either contrasting signal amplitudes from clearly defined conditions in simplistic task stimuli (to localize “activations”) or on computing time-course cross-correlations between pairs of brain regions to infer the strength of functional “connectivity” between them in resting-state fMRI studies, and 2) low signal-to-noise ratio/difficulty separating signal due to relevant neuronal fluctuations from signal due to noise. Consequently, research findings typically need to be averaged over many trials and subjects. On the other hand, clinical diagnostics are necessarily based upon a single subject, thus requiring high quality data and methods sensitive to abnormalities in network dynamics. To address the first issue, work presented here is driven by the hypothesis that the most sensitive biomarker of dysregulation may not be the amplitude of activation or strength of connections between two regions, but rather the complexity of the signal, reflecting the underlying (deviations in) dynamics. We introduce and apply an entropic measure of regulation and feedback (the autocorrelation function) to identify focal regions in patients with medication-resistant epilepsy. Precise localization of foci is crucial for successful surgery. To address the second issue, we introduce a new quantitative measure to accurately assess the integrity of fMRI time-series – signal fluctuation sensitivity (SFS). We show that SFS correlates with time-series integrity and that higher SFS is associated with enhanced sensitivity to detection of known local and long-range connections in resting-state (task-free) fMRI. We further show that this measure reliably identifies task-induced activations in three different tasks employing highly complex naturalistic stimuli, which still represent a challenge from the data-analysis perspective. Finally, we incorporate high quality fMRI data with machine learning to build models capable of predicting subjects’ dynamic state of mind from fMRI signals of relevant brain networks at an individual-subject level.|
|Appears in Collections:||Stony Brook Theses and Dissertations Collection|
Files in This Item:
|Nedic_grad.sunysb_0771E_13461.pdf||15.06 MB||Adobe PDF||View/Open|
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.