A Virtual Clinical Trial of FDG-PET Imaging of Breast Cancer: Effect of Variability on Response Assessment

Abstract

INTRODUCTION: There is growing interest in using positron emission tomography (PET) standardized uptake values (SUVs) to assess tumor response to therapy. However, many error sources compromise the ability to detect SUV changes. We explore relationships between these errors and overall SUV variability. METHODS: We used simulations in a virtual clinical trial framework to study impacts of error sources from scanning and analysis effects on assessment of SUV changes. We varied tumor diameter, scan duration, pretherapy SUV, magnitude of change in SUV, image reconstruction filter, and SUV metric. Poisson noise was added to the raw data before image reconstruction. Variance from global sources of error, e.g., scanner calibration, was incorporated. Two thousand independent noisy sinograms per scenario were generated and reconstructed. We used SUVs to create receiver operating characteristic (ROC) curves to quantify ability to assess response. Integrating area under the ROC curve summarized ability to detect SUV changes. RESULTS: Scan duration and image reconstruction method had relatively little impact on ability to measure response. SUVMAX is nearly as effective as SUVMEAN, especially with increased image smoothing and despite size-matched region of interest placement. For an effective variability of 15%, we found the Positron Emission Tomography Response Criteria in Solid Tumors criteria for measuring response (±30%) similar to the European Organization for Research and Treatment of Cancer criteria (±25%). CONCLUSIONS: For typical PET variance levels, tumor response must be 30% to 40% to be reliably determined using SUVs. PET scan duration and image reconstruction method had relatively little effect.