Measurements of tumor glucose use by (18)F-FDG PET need to be standardized within and across institutions. Various parameters are used for measuring changes in tumor glucose metabolic activity with (18)F-FDG PET in response to cancer treatments. However, it is unknown which of these provide the lowest variability between observers. Knowledge of the interobserver variability of quantitative parameters is important in sarcomas as these tumors are frequently large and demonstrate heterogeneous (18)F-FDG uptake.
Methods: A total of 33 patients (16 men, 17 women; mean age, 47 +/- 18 y) with high-grade sarcomas underwent (18)F-FDG PET/CT scans before and after neoadjuvant chemotherapy. Two independent investigators measured the following parameters on the pretreatment and posttreatment scans: maximum standardized uptake value (SUVmax), peak SUV (SUVpeak), mean SUV (SUVmean), SUVmean in an automatically defined volume (SUVauto), and tumor-to-background ratio (TBR). The variability of the different parameters was compared by concordance correlation coefficient (CCC), variability effect coefficient, and Bland-Altman plots.
Results: Baseline SUVmax, SUVpeak, SUVmean, SUVauto, and TBR averaged 10.36, 7.78, 4.13, and 6.22 g/mL and 14.67, respectively. They decreased to 5.36, 3.80, 1.79, and 3.25 g/mL and 6.62, respectively, after treatment. SUVmax, SUVpeak, and SUVauto measurements and their changes were reproducible (CCC > or = 0.98). However, SUVauto poorly differentiated between responding and nonresponding tumors. The high intratumoral heterogeneity of (18)F-FDG resulted in frequent failure of the thresholding algorithm, which necessitated manual corrections that in turn resulted in a higher interobserver variability of SUVmean (CCCs for follow-up and change were 0.96 and 0.91, respectively; P < 0.005). TBRs also showed a significantly higher variability than did SUVpeak (CCCs for follow-up and change were 0.94 and 0.86, respectively; P < 0.005).
Conclusion: SUVmax and SUVpeak provided the most robust measurements of tumor glucose metabolism in sarcomas. Delineation of the whole-tumor volume by semiautomatic thresholding did not decrease the variability of SUV measurements. TBRs were significantly more observer-dependent than were absolute SUVs. These findings should be considered for standardization of clinical (18)F-FDG PET/CT trials.