Advisor(s)
Yu-Hua Fang
Committee Member(s)
Anna Sorace
David Standaert
Jonathan McConathy
Mark Bolding
School
Joint Health Sciences (Interdisciplinary)
Document Type
Dissertation
Department (new version)
Biomedical Engineering
Date of Award
9-11-2025
Abstract
The evolving landscape of Alzheimer's disease (AD) research and clinical practice has increased the need for precise, efficient, and accessible quantification of positron emission tomography (PET) biomarkers. However, utilizing the full potential of PET imaging is often complicated by several methodological challenges, including subject motion, computationally intensive analysis pipelines, and reliance on magnetic resonance imaging (MRI) for optimal standard PET quantification. The work in this dissertation aims to contribute to this field by developing and validating a series of novel image processing methods designed to improve upon existing quantification workflows. This work first explored the challenge of subject motion in simultaneous PET/MRI scans. A tracer characteristic-based co-registration (TCBC) method was proposed, which demonstrated improved accuracy and enhanced statistical power to detect amyloid accumulation compared to standard mutual-information based approaches. Second, to address the goal of improving workflow efficiency, this dissertation developed and evaluated deep learning (DL) segmentation models as alternatives to traditional, time-consuming segmentation methods. These DL models, especially our own developed Label Extraction of Neuroimaging (LEON) model, were found to provide quantitative and diagnostic results comparable to the FreeSurfer standard. Finally, to enhance clinical accessibility, this work focused on developing an MRI-less quantification workflow. A novel DL MRI-less segmentation model was developed for PET/computed tomography (CT) data, which achieved statistically equivalent quantification to the MRI-based reference standard when detecting amyloid positivity. In conclusion, this dissertation contributes a series of validated computational methods that offer improvements to the accuracy, efficiency, and potential clinical utility of PET quantification. This work presents a potential pathway for bridging advanced research methodologies and broader clinical application, aiming to better support AD diagnosis and disease monitoring.
ProQuest ID
Recommended Citation
Chiu, Sheng-Chieh, "Optimizing Quantification In Alzheimer'S Disease With Pet Imaging Through Advanced Imaging And Deep Learning Techniques" (2025). All ETDs from UAB. 7423.
https://digitalcommons.library.uab.edu/etd-collection/7423
