Automatic aorta delineation and blood SUV determination for computation of standard uptake ratios (SUR)

Ziel/Aim:

Accurate calibration is a prerequisite for SUV quantification in PET and usually performed via phantom measurements. However, such a calibration is only valid for the used phantom since attenuation and scatter correction limitations affect quantification accuracy (QA) in-vivo. In recent studies in-vivo QA was assessed using bladder image vs. urine samples comparisons and showed a systematic underestimate of SUVs when relying on phantom-based calibration. A notable disadvantage of this approach is that it requires additional effort, making it unattractive for clinical use. The goal of this work was evaluation of an alternative method utilizing image-derived arterial blood SUVs (BSUV) averaged over a sufficiently large number of subjects.

Methodik/Methods:

We analyzed 681 patient scans from 3 sites which underwent routine 18F-FDG PET/CT or PET/MR. BSUVs were determined in the descending aorta using a roughly cylindrical ROI (delineated in the attenuation image and spatially transferred to PET). Minimum ROI volume was 5 ml and a safety margin used to avoid partial volume effects. Mean BSUVs, standard deviation (SD), and standard error of the mean (SEM) were computed for subgroups corresponding to 3 scanner calibrations of each site (9 subgroups, n=54-100). Intra- and inter-site variability was computed.

Ergebnisse/Results:

Relative SD (SEM) of BSUV in the subgroups ranged 14.3%–20.7% (2.8%–4.8%). BSUV-differences between intra-site groups were 1.8%–6.4% and mostly (5/6) insignificant. Inter-site BSUV-differences were significant and much larger (12.6%–25.1%, P<0.001).

Schlussfolgerungen/Conclusions:

Due to low inter-site BSUV-variability, group-averages can be computed with high statistical accuracy (<5%) in groups of ~70 patients and used for comparison of relative in-vivo QA. For multi-center trials this method can reveal inter-site differences at the 10% level and might even be used to calibrate image data between sites.