Reliability of SUV estimates in FDG PET as a function of acquisition and processing protocols

Standardized uptake values (SUV) are commonly used in FDG PET to characterize suspicious high uptakes. To better understand the reliability and the limits of SUV, we studied the accuracy of SUV estimates using phantom data. Using the Data Spectrum thorax phantom in which spheres were inserted, we studied the effect of the sphere sizes, of out-of-the-field-of-view-activity, of the emission scan duration and of attenuation and partial volume effect (PVE) corrections upon SUV biases. Considering a specific acquisition and processing protocol as a reference, we determined the changes in SUV when modifying this protocol. For a true SUV of 8, estimated SUV were strongly dependent on lesion size, but also on the acquisition and processing protocol. Depending on the method used to derive the attenuation map, estimated SUV could change by more than 50% for lung spheres. Using PVE correction, SUV increased by a factor greater than 2 for spheres less than 2 cm in diameter. The very definition of SUV could also change the SUV by more than 50%. Even with CT-based attenuation correction and PVE correction, SUV estimates remained underestimated by more than 20% in lesions less than 2 cm in diameter. Biases in SUV estimates strongly depend on the acquisition and processing protocols. This suggests that comparing SUV between studies make sense only if the acquisition and processing protocols are strictly identical.