Dissertations and Theses (Open Access)

Author ORCID Identifier

https://orcid.org/0000-0003-3543-1719

Date of Graduation

5-2026

Document Type

Dissertation (PhD)

Program Affiliation

Medical Physics

Degree Name

Doctor of Philosophy (PhD)

Advisor/Committee Chair

Ho-Ling Anthony Liu

Committee Member

John D. Hazle

Committee Member

Vinodh A. Kumar

Committee Member

Sujit S. Prabhu

Committee Member

R. Jason Stafford

Committee Member

Peng Wei

Abstract

Functional Magnetic Resonance Imaging (fMRI) has established itself over the past three decades as the dominant functional neuroimaging modality, seeing widespread use in cognitive, systems and clinical neuroscience. Primarily, fMRI records regional differences in the endogenous blood oxygenation level-dependent (BOLD) contrast associated with hemodynamic changes driven by neuronal activity. BOLD fMRI has provided numerous insights into human brain function and functional organization, and a wide range of clinical applications have been proposed, spanning neurobehavioural health, neurodegeneration, to guiding neurological surgery and interventions. However, its main clinical application remains largely confined to presurgical functional mapping. This limitation arises because commonly used, well-established analytical methods primarily emphasize group-level inference, whereas clinical contexts expect inference and interpretation at the level of the individual patients. Accordingly, patients with brain tumors provide a compelling circumstance, as personalized analytical approaches must demonstrate robustness in the face of pronounced deviations in functional anatomy relative to healthy individuals. Converging evidence suggests that standard-of-care task-based (tb‑) fMRI is limited in approximately one‑third of patients with brain tumors in eloquent areas. While resting-state (rs‑) fMRI offers an alternative source of functional mapping, prevailing heuristics incorporated in rs-fMRI analysis do not emphasize fidelity to patient‑specific inference. Therefore, in this work, we focus on single-subject data-driven heuristics to develop methods for personalized rs-fMRI analysis. To this end, we proposed a probabilistic atlas–based approach that accounts for intersubject variability. This was accomplished with, (i) the construction of a probabilistic language network atlas from tb-fMRI of patients with brain tumors; and (ii) the conception of probabilistic template matching that leverages either the generalized weighted goodness-of-fit or information-theoretic Jensen-Shannon distance. We found that probabilistic template matching enables better detections of the language network from independent component analysis of rs-fMRI. Next, we developed geometry agnostic seed-based correlation (SBC) analysis that uses an iterative data-driven approach combining unsupervised manifold learning with distance-based clustering to generate seeds without geometrical constraints. Jensen-Shannon distance-based probabilistic template matching was used for detection of the language and motor rs-fMRI SBC networks obtained from geometry-agnostic seeds. Through this process, heuristics associated with one-size-fits-all SBC analysis such as spherical seeds or atlas-defined regions of interest can be circumvented. Finally, we proposed an extension to the latter approach for developing personalized functional parcellations, which we found to improve the association between memory and functional network segregation in brain tumor patients.

Keywords

Functional MRI, Resting-state Functional MRI, Language, Brain, Brain Tumors

Download

Available for download on Thursday, May 06, 2027

Share

COinS