: PET volume determination reproducibility

International audienceThe objective of this study was to establish the repeatability and reproducibility limits of several volume-related PET image-derived indices-namely tumor volume (TV), mean standardized uptake value, total glycolytic volume (TGV), and total proliferative volume (TPV)-relative to those of maximum standardized uptake value (SUV(max)), commonly used in clinical practice. METHODS: Fixed and adaptive thresholding, fuzzy C-means, and fuzzy locally adaptive Bayesian methodology were considered for TV delineation. Double-baseline (18)F-FDG (17 lesions, 14 esophageal cancer patients) and 3'-deoxy-3'-(18)F-fluorothymidine ((18)F-FLT) (12 lesions, 9 breast cancer patients) PET scans, acquired at a mean interval of 4 d and before any treatment, were used for reproducibility evaluation. The repeatability of each method was evaluated for the same datasets and compared with manual delineation. RESULTS: A negligible variability of less than 5% was measured for all segmentation approaches in comparison to manual delineation (5%-35%). SUV(max) reproducibility levels were similar to others previously reported, with a mean percentage difference of 1.8% +/- 16.7% and -0.9% +/- 14.9% for the (18)F-FDG and (18)F-FLT lesions, respectively. The best TV, TGV, and TPV reproducibility limits ranged from -21% to 31% and -30% to 37% for (18)F-FDG and (18)F-FLT images, respectively, whereas the worst reproducibility limits ranged from -90% to 73% and -68% to 52%, respectively. CONCLUSION: The reproducibility of estimating TV, mean standardized uptake value, and derived TGV and TPV was found to vary among segmentation algorithms. Some differences between (18)F-FDG and (18)F-FLT scans were observed, mainly because of differences in overall image quality. The smaller reproducibility limits for volume-derived image indices were similar to those for SUV(max), suggesting that the use of appropriate delineation tools should allow the determination of tumor functional volumes in PET images in a repeatable and reproducible fashion