Advancing Multi-Center 129Xe Gas Exchange MRI Through Advanced Segmentation, Artifact Mitigation, and Establishing Reference Distributions

Abstract

Pulmonary diseases impose a major clinical burden, yet many standard clinical assess-ments remain limited in their ability to capture regional dysfunction and to support robust comparison across subjects, sites, and scanner platforms. Hyperpolarized 129Xe MRI/MRS offers a powerful way to probe lung function by providing spatially resolved ventilation imaging as well as compartment-resolved gas-exchange measurements that reflect transfer from alveolar airspaces to the interstitial membrane and to red blood cells. However, translating these capabil-ities into quantitative, multi-site–ready biomarkers requires analysis workflows that are auto-mated, standardized, and resilient to dominant technical confounders.The central hypothesis of this dissertation is that robust multi-site quantification can be achieved by (i) reducing reliance on auxiliary acquisitions and manual processing steps that in-troduce variability, (ii) establishing harmonized healthy reference distributions that enable con-sistent interpretation across vendors, and (iii) identifying and controlling confounders—particularly intensity inhomogeneity, incidental gas-phase excitation, and decay-correction as-sumptions—that can systematically bias derived metrics.Chapter 2 reviews pulmonary physiology and the core principles of 129Xe MRI/MRS that motivate ventilation and dissolved-phase biomarkers, including how compartment-resolved imaging supports interpretation of gas-exchange impairments. Chapter 3 addresses a key scalability bottleneck—thoracic cavity segmentation—by de-veloping a xenon-only deep learning framework that eliminates dependence on registered pro-ton scans. Because annotated training volumes are scarce in 129Xe MRI, the framework com-bines template-based augmentation with Pix2Pix GAN-based synthesis to expand anatomical and defect-pattern diversity, and demonstrates strong agreement between model-derived and expert-derived thoracic cavity volume and ventilation defect percentage.Chapter 4 establishes standardized healthy reference distributions and normative values for multi-site 129Xe gas-exchange MRI/MRS in a young healthy cohort spanning Siemens, Philips, and GE platforms, providing a foundation for consistent quantitative interpretation across centers. Chapter 5 addresses incidental gas-phase excitation during dissolved-phase imaging by implementing and directly comparing two leading correction strategies (dual-echo and spectros-copy-informed frameworks) under matched acquisition conditions on Siemens, and by deriving quantitative thresholds for tolerable contamination to support reliable gas-exchange metrics.Chapter 6 extends robustness and physiologic interpretability in three ways. First, it evaluates multi-step versus standard N4ITK bias-field correction for 129Xe ventilation MRI us-ing blinded expert reader analysis, supporting multi-step correction as a preferred approach for improving uniformity and defect depiction in multi-center workflows. Second, it outlines a conceptual framework for imaging collateral ventilation beyond a single breath-hold by inte-grating a spiral-based acquisition strategy and optimized breathing maneuvers, with planned comparisons to sampling-efficient trajectories. Third, it motivates spatially resolved estimation of membrane and RBC T2* using 129Xe chemical shift imaging to improve decay correction beyond whole-lung spectroscopy assumptions.Chapter 7 concludes by outlining future directions to improve robustness and clinical utility, emphasizing trachea-aware/user-guided segmentation, expanded normative reference cohorts, and more physiology-informed correction strategies for multi-site deployment.Together, these contributions advance 129Xe MRI/MRS from a powerful functional im-aging capability toward a standardized and scalable quantitative biomarker platform suitable for multi-site studies and broader clinical translation.