An evaluation of Bayesian regression for estimating cerebral oxygen utilization with oxygen-15 and dynamic PET

The use of Bayesian regression for improving the estimation of cerebral oxygen utilization rate (CMRO) was investigated. Based on plasma time-activity curves of (15)O oxygen and (15)O water that were obtained previously in a normal subject, realistic tissue time-activity curves of (15)O radioactivity in gray and white matter as obtained in dynamic PET (positron emission tomography) (15)O oxygen studies were computer-simulated for various noise conditions and blood flow variations. Bayesian regression and regular least-square regressions were applied to estimate CMRO from the kinetics. The accuracy and precision of the estimates were evaluated. The improvement in precision was as high as 50% for gray matter of 1 cm (3) size in a study of 0.5 million counts/slice over 3 min. Similar improvement was achieved for white matter in 1-million-count studies. The amount of improvement is larger for higher noise cases.