Precision measurements of brain oxygen utilization with positron emission tomography

Positron emission tomography (PET) with [15O] tracers provides the definitive standard for measuring blood flow, oxygen extraction, oxygen metabolism, and more recently, aerobic glycolysis.  PET with  is especially prominent for the neurosciences, broadly supporting fundamental and clinical studies of stroke and neurodegeneration.  Precision measurements of these pathophysiologies by PET are valuable for attaining medically relevant advances, including the validation of more accessible instrumental approaches, such as those based on magnetic resonance imaging.  However, quantitative PET with [15O] poses technical challenges concerning sensitivity and resolution of imaging an energetic and short-lived radiotracer, and the requirement of input functions for models of tracer kinetics.   In this work, we comprehensively address these challenges using Bayesian inference, implementing dynamic nested sampling.  Quantitative physiologic phenomenologies derive from established, integrable kinetic models, which we enhanced to include finite sampling and rectangular window moving averages.  We also extended kinetic models to include a realistic and generative phenomenology for tracer input functions.  The extended phenomenology accommodates both invasively sampled arterial input functions (AIF) and image-derived input functions (IDIF).  Nested sampling estimates data evidence and IDIFs demonstrated more favorable evidence than AIFs.  These findings suggest extensions to quantitative tracer kinetics that are principled, computationally tractable, and more scalable to larger human studies.