TY - JOUR T1 - Algorithms to harmonize SUVmax in clinical 18F-FDG PET scans to a standard injection-to-acquisition time interval (uptake time) JF - Journal of Nuclear Medicine JO - J Nucl Med SP - 261 LP - 261 VL - 60 IS - supplement 1 AU - Brenda Kurland AU - Huang Lin AU - Haoyi Fu AU - Robert Krafty Y1 - 2019/05/01 UR - http://jnm.snmjournals.org/content/60/supplement_1/261.abstract N2 - 261Objectives: FDG PET, including standardized uptake value (SUV) reporting, is widely used in clinical practice for solid tumor cancer staging and restaging. However, use of the SUV as a pharmacodynamic biomarker or clinical trial surrogate endpoint is limited by measurement error, introduced by inherent biological variability, image acquisition procedures, and post-acquisition data analysis. Valuable research has sought to standardize acquisition protocols and to harmonize data acquired using different protocols. This work concerns straightforward correction to a common deviation from scanning protocols: variation in injection-to-acquisition time. We describe 3 correction algorithms, and (following abstract submission) will compare their performance in independent dual-acquisition datasets that mimic a common desired correction for multicenter trials: SUVmax at 60 min when the scan was performed substantially later. Methods: The first algorithm extends both the data (60-min dynamic scans before and during neoadjuvant chemotherapy for locally advanced breast cancer) and the method (linear regression models) from the 2003 Beaulieu correction (1). The number of scans for model development is increased to 150 from 20, and the linear regression model is extended to restricted maximum likelihood (REML) estimates of fixed and random effects, with SUVmax at 60 min estimated with an associated confidence interval. The second algorithm, introduced in 2014 by van den Hoff (2), emphasizes the role of blood FDG clearance in time-dependence of SUVmax. Since we focus on measurements available in a standard clinical scan, we use a simplified version of the algorithm that does not require blood measurement. The third approach is empirical and nonparametric, using kernel regression (bandwidth selected by least-squares cross validation) to summarize plausible tumor time-activity curves within the range of SUV values and uptake times of interest to produce uptake-time-corrected estimates of SUVmax and of precision of the estimate. Results: Preliminary results are examined for an expected scenario for multicenter clinical trials and other applications of FDG PET as a pharmacodynamics biomarker: acquisition begins at 90 min post-injection (because a site desires better tumor-to-blood signal, or due to equipment- or patient-based delay) but the SUVmax value for 60 min is desired. For common SUVmax values observed at 90 min the estimated 60 min SUVmax varies substantively between different methods (updated Beaulieu, simplified van den Hoff, and empirical, respectively): for 3.5 at 90 min the 60 min estimates are (3.7, 2.6, 3.1); for 6 estimates are (5.2, 4.5, 5.0); for 10 estimates are (7.3, 7.6, 8.3). Encouragingly, these empirical estimates (3.1, 5.0, 8.3) using 1000 plausible curves generated by independent sampling from kinetic parameter estimates (see (3) and (4)) are similar to empirical estimates (3.1, 5.1, 7.9) from fitted curves from the 150 dynamic scans used for the updated Beaulieu correction. Further development (completed by SNMMI 2019) will complete precision estimates (confidence bands), formalize comparisons among algorithms, and evaluate performance in independent datasets from studies with dual acquisition (static scans at 60 and 90 min). Conclusions: Unexpectedly, preliminary results suggest that empirical estimates from simulated dynamic FDG-PET scans may produce more accurate uptake time adjustments than model-based adjustments fitted to 150 actual dynamic scans. If confirmed in the full analysis, an empirical adjustment is consistent with the straightforward nature of the uptake time correction problem and demonstrates the robustness of the 2-compartment model for tumor FDG uptake, with k4=0 through 120 min and a population arterial input function (3, 4). ER -