Advisor(s)

Yu-Hua Dean Fang

Committee Member(s)

Anna Sorace
Erik Roberson
Jonathan McConathy
Kristina Visscher

Document Type

Dissertation

Date of Award

1-27-2026

Degree Name

Doctor of Philosophy (PhD)

School

Joint Health Sciences (Interdisciplinary)

Department

Biomedical Engineering

Abstract

Medical imaging plays a central role in modern clinical decision-making, yet its effective integration with artificial intelligence (AI) remains challenged by real-world clinical constraints. This dissertation investigates how medical imaging-based AI frameworks can be strategically designed to support clinical decision-making for two distinct but clinically critical scenarios: acute pandemic care and chronic neurodegenerative disease management. In acute pandemic settings, such as the COVID-19 crisis, healthcare systems face an urgent need for rapid risk stratification under conditions of data scarcity and operational stress including surges in patient volume and limited medical resources. This work explores strategies for developing deep learning models during early-stage outbreaks, where large, well-annotated datasets are often unavailable. By leveraging transfer learning and biologically grounded label harmonization, the proposed framework demonstrates the feasibility of extracting prognostic information from limited chest X-ray data, supporting the rapid development of automated triage tools in high-demand acute clinical environments. The focus of this dissertation subsequently shifts to chronic neurodegeneration, specifically Alzheimer’s disease (AD), where affordable and accessible biological staging of tau pathology remains a challenge. While tau positron emission tomography (PET) provides in vivo assessment of tau staging, its high cost and limited availability restrict widespread clinical use. This work identified structural MRI–derived anatomical patterns that reflect the spatial progression of tau-associated neurodegeneration, providing a biological rationale for surrogate modeling. Building on this observation, a feature-based machine learning model is developed to integrate MRI-derived structural features with clinical variables, enabling cost-effective and accessible tau stratification for AD treatment planning. By aligning model design with data availability, biological contexts, and clinical workflows, this dissertation emphasizes the importance of developing medical AI approaches that are responsive to real-world clinical constraints. Together, these studies illustrate how different AI paradigms—deep learning for exploratory feature discovery in acute settings and feature-based machine learning for interpretable modeling in chronic disease—can be tailored to address distinct clinical needs. This work contributes to the development of scalable, clinically relevant medical imaging AI systems with potential to enhance diagnostic efficiency, risk stratification, and treatment planning.

ProQuest Publication Number

32442018

ISBN

9798273349889

Download

Share

COinS