Measuring bias in quantitative PET biomarkers in-vivo.

Objectives: Quantitative imaging biomarkers are widely used in PET for both research and clinical applications, yet bias in the underlying image data is not well-characterized, particularly for studies below the head. In the absence of a readily available reference standard for in-vivo quantification, the bias of PET images has to be inferred using physical phantoms, even though arrangements of this sort provide only a poor approximation of the imaging environment in patient studies. Phantom-based evaluations may give an unrealistically optimistic impression of PET quantitative accuracy because they do not reflect the complexity of the scatter and attenuation distributions in real patients. In this study we used data acquired in patient volunteers to assess PET quantitative bias in-vivo. Image-derived activity concentrations in the descending aorta were compared with simultaneously acquired blood samples counted on a calibrated gamma counter.

Methods: Ten patients with prostate cancer were studied using the prostate-specific membrane antigen PET imaging agent [¹⁸F]DCFPyL (ClinicalTrials.gov NCT02981368). For each patient, 3 whole-body PET/CT image series were acquired using a Biograph mCT: immediately after radiotracer injection and approximately 1 and 4 hours later. Venous blood samples were obtained from an indwelling catheter in the vein of an arm at 8 time points over an 8 hour period and whole-blood was counted on a NaI gamma counter (Wizard 2480). Both the PET scanner and gamma counter were calibrated with respect to a NIST traceable ⁶⁸Ge standard, which was itself cross-calibrated for ¹⁸F. PET images were reconstructed using 3D OSEM + TOF, 2 iterations, 21 sub-sets, 5 mm Gaussian and standard corrections for attenuation, scatter, randoms and deadtime. A 10 mm diameter cylindrical volume-of-interest was automatically placed in the descending thoracic aorta to estimate the PET-derived activity concentration in blood (C_PET). No corrections were applied for partial volume or for time differences between the aorta and the peripheral sample point. A tri-exponential function was fit to the gamma counter blood data and used to estimate the activity concentration (C_gamma) at the time of each PET acquisition. The difference (d) was calculated as d = C_PET - C_gamma and the relative difference (D) was calculated as D = 100(C_PET - C_gamma) / 0.5(C_PET + C_gamma).

Results: No data were lost or compromised and 30 pairs of quantitative measurements were available for analysis. C_PET and C_gamma were linearly related with R² = 0.985. The mean difference between the PET and gamma counter data was 290 Bq/mL with the PET measurements tending to be greater. These data (d) were normally distributed (Shapiro-Wilk test, p=0.97) and the 95 % lower and upper limits of agreement were -495 and 1076 Bq/mL respectively. In relative units (D) the mean bias corresponded to +4.8 %. These relative difference data (D) were not normally distributed (Shapiro-Wilk test, p = 0.03). However for a subset of the data above 5,000 Bq/mL (corresponding to an SUV > 1 in a 74 kg patient following 370 MBq) the relative difference data (D) were normally distributed (Shapiro-Wilk, p=0.38). In this range, the 95 % limits of agreement were -2.6 % and +13.0 %. Over the 9 month period covering the dates of patient data acquisition, scanner quality control using an ¹⁸F-filled phantom had a mean SUV of 0.993 ± 0.005. Conclusions: Human image data acquired on a whole-body PET/CT system with a typical clinical protocol differed by an average of around 5 % compared to blood samples counted on a calibrated gamma counter. This relatively low bias is encouraging, particularly as it was measured in the complex imaging environment encountered in the chest, and may be partly attributable to residual uncorrected scatter. These data provide a unique opportunity for the assessment of PET bias in-vivo, supporting the quantitative application of [¹⁸F]DCFPyL, with positive implications for other radiotracers. Research support: Progenics Pharmaceuticals Inc; R01-CA134675