Maximizing model parameter precision by optimizing dose in 3D [15O]H2O PET scans

Dynamic PET imaging with [¹⁵O]H2O allows estimation of tumour perfusion in humans. For small regions of interest these estimates have limited precision due to statistical noise in the images. We have optimized the injected dose in order to improve the precision of these estimates. This optimization considers the effect of the scanner count-rate upon the local image variance and the subsequent effect on the parameter estimation. The work includes examination of variance images reconstructed via Alpert’s method and Monte Carlo simulations of the kinetic modelling process. Optimal doses are derived from [¹⁵O]H2O PET studies performed on a modern whole-body 3D PET/CT scanner (n=7). These scans (mostly abdominal) show that changes in local image variance can be predicted by the global count-rates to within 15 % over a wide range of injected doses and for the majority of the time-course of the dynamic scan. We find optimal doses for estimation of perfusion (F) in the range of 250-530 MBq, and 520-730 MBq for estimation of the volume of distribution (VT). Injection at 100-200 MBq greater than the calculated optimum rarely degrades the precision of the parameter estimates. Hence, a standardized dose of 500-700 MBq would be appropriate for these subjects if estimating F and VT simultaneously.