Quantitation of cancer treatment response by 18F-FDG PET/CT: Multicenter assessment of measurement variability

Abstract

The aim of this study was to assess the interobserver variability of quantitative 18F-FDG PET/CT parameters used in assessments of treatment response across multiple sites and readers. Methods: Paired pre- and posttreatment 18F-FDG PET/CT images of 30 oncologic patients were distributed to 22 readers across 15 U.S. and international sites. One reader was aware of the full medical history (readreference) of the patients, whereas the 21 other readers were unaware. The readers selected the single hottest tumor from each study, and made SUV measurements from this target lesion and the liver. Descriptive statistics, percentage changes in the measurements, and their agreements were obtained. Results: The intraclass correlation coefficient for the percentage change in SUVmax (%D SUVmax) of the hottest tumor was 0.894 (95% confidence interval [CI], 0.813-0.941), and the individual equivalence coefficient was 1.931 (95% CI, 0.568-6.449) when all reads were included (n = 638). When only the measurements that selected the same target tumor as the readreference (n = 486) were included, the intraclass correlation coefficient for the %DSUVmax was 0.944 (95% CI, 0.841-0.989), and the individual equivalence coefficient was 20.688 (95% CI, 21.810 to 20.092). The absolute change in SUVmean of liver corrected for lean body mass showed upper and lower limits of agreement (average bias ± 2 SDs) of 0.13 and 20.13 g/mL. Conclusion: The quantitative tumor SUV changes measured across multiple sites and readers show a high correlation. Selection of the same tumor target among readers further increased the degree of correlation.