WE‐G‐214‐03: Investigating the Repeatability of Quantitative PET Imaging of Heterogeneous Tumors

Purpose: High precision within quantitative imaging studies is required to define biological targets for dose painting, and to evaluate responses to therapy. However, within PET imaging systems there are position‐dependent systematic fluctuations in contrast‐recovery. The objective of this study was to determine the impact of patient set‐up on PET‐based quantification of heterogeneous tumors. Methods: Ten dogs with sinonasal tumors received six repeated [18F]FDG‐PET scans. Between each PET acquisition, known set‐up errors were produced by varying patient position in precisely controlled 1.0±0.5 mm steps. Image noise was investigated by retrospectively varying acquisition time of a stationary scan. Resulting changes in SUVmax and PET‐based target volumes were evaluated. Correlation between corresponding voxels of co‐registered images determined repeatability of spatial distributions. A theoretical upper limit of repeatability was estimated from simulated PET images using empirical models of position‐dependent contrast‐recovery. Results: From the simulated PET images, a theoretical upper limit of 0.90 was estimated for the repeatability of spatial distributions when set‐up errors are present. When only image noise was varied between [18F]FDG‐PET scans, repeatability of spatial distributions was measured at 0.95. However, set‐up errors lead to significant decreases (p < 0.01) in voxel correlations. Introducing set‐up errors between [18F]FDG‐PET acquisitions reduced the repeatability of spatial distributions to 0.80, and caused changes of 5% in SUVmax and 10% in volumes of PET‐based targets. Conclusions: Set‐up errors as small as a few millimeters (< 3mm) reduced the reproducibility of quantitative PET imaging of heterogeneous tumors. PET‐based quantitative values generally varied within 10%, but target volumes within some tumors changed by 30%. Errors during PET acquisition will lead to uncertainty in quantifying changes in tumor function, and to limited accuracy and precision of PET‐based biological target volumes. Therefore, tumor delineation must be independent of SUV thresholds and uncertainty margins are required for PET‐based response quantification.