<p>Since the inception of hyperpolarized 129Xe MRI, the field has yearned for more efficient production of more highly polarized 129Xe. For nearly all polarizers built to date, both peak 129Xe polarization and production rate fall far below theoretical predictions. This thesis sought to develop a fundamental understanding of why the observed performance of large-scale 129Xe hyperpolarization lagged so badly behind theoretical predictions.</p><p> This is done by thoroughly characterizing a high-volume, continuous-flow polarizer using optical cells having three different internal volumes, and employing two different laser sources. For each of these 6 combinations, 129Xe polarization was carefully measured as a function of production rate across a range of laser absorption levels. The resultant peak polarizations were consistently a factor of 2-3 lower than predicted across a range of absorption levels, and scaling of production rates deviated badly from predictions based on spin exchange efficiency.</p><p> To bridge this gap, we propose that paramagnetic, activated Rb clusters form during spin exchange optical pumping (SEOP), and depolarize Rb and 129Xe, while unproductively scattering optical pumping light. When a model was built that incorporated the effects of clusters, its predictions matched observations for both polarization and production rate for all 6 systems studied. This permits us to place a limit on cluster number density of <2 × 109 cm-3.</p><p> The work culminates with deploying this framework to identify methods to improve polarization to above 50%, leaving the SEOP cell. Combined with additional methods of preserving polarization, the polarization of a 300-mL batch of 129Xe increased from an average of 9%, before this work began, to a recent value of 34%.</p><p> We anticipate that these developments will lay the groundwork for continued advancement and scaling up of SEOP-based hyperpolarization methods that may one day permit real-time, on-demand 129Xe MRI to become a reality.</p> / Dissertation