Development and validation of a variance model for dynamic PET: uses in fitting kinetic data and optimizing the injected activity

Abstract

The precision of biological parameter estimates derived from dynamic PET data can be limited by the number of acquired coincidence events (prompts and randoms). These numbers are affected by the injected activity (A0). The benefits of optimizing A0 were assessed using a new model of data variance which is formulated as a function of A0. Seven cancer patients underwent dynamic [15O]H2O PET scans (32 scans) using a Biograph PET-CT scanner (Siemens), with A0 varied (142–839 MBq). These data were combined with simulations to (1) determine the accuracy of the new variance model, (2) estimate the improvements in parameter estimate precision gained by optimizing A0, and (3) examine changes in precision for different size regions of interest (ROIs). The new variance model provided a good estimate of the relative variance in dynamic PET data across a wide range of A0s and time frames for FBP reconstruction. Patient data showed that relative changes in estimate precision with A0 were in reasonable agreement with the changes predicted by the model: Pearson's correlation coefficients were 0.73 and 0.62 for perfusion (F) and the volume of distribution (VT), respectively. The between-scan variability in the parameter estimates agreed with the estimated precision for small ROIs (<5 mL). An A0 of 500–700 MBq was near optimal for estimating F and VT from abdominal [15O]H2O scans on this scanner. This optimization improved the precision of parameter estimates for small ROIs (<5 mL), with an injection of 600 MBq reducing the standard error on F by a factor of 1.13 as compared to the injection of 250 MBq, but by the more modest factor of 1.03 as compared to A0 = 400 MBq.