Participants and equipment

Seven members of the MRTDosimetry consortium participated in the comparison exercise (Azienda USL-IRCCS di Reggio Emilia, The Christie NHS Foundation Trust, Lund University, Oxford University Hospitals NHS Foundation Trust, Royal Surrey NHS Foundation Trust, “THEAGENIO” Anticancer Hospital, and University Hospital Würzburg). The participants were required to have access to a SPECT/CT system with high-energy collimators as well as methods for attenuation and scatter correction. In total, eight SPECT/CT imaging systems (5 × General Electric (GE) and 3 × Siemens) were included in the study. The individual setup details are given in Table 1, where four different combinations of reconstruction and correction methods were performed with system 2 (setups S2a-S2d). The description of the acquisition and reconstruction settings are provided in later sections.

Table 1 SPECT/CT systems and setups used in the comparison exercisePhantom design

SPECT/CT activity calibration factors are often determined in large-volume phantoms to reduce the influence of the partial volume effect. However, the production of such a large cylinder uniformly filled with a sufficiently high activity concentration of solidified 133Ba is challenging. Therefore, the comparison exercise was performed based on a set of four smaller cylinders of different sizes, similar to those proposed by Zimmerman et al. [10]. Although it was not the primary aim of this study, the resulting multi-centre dataset was used to provide a comparison of the relative impact of spatial resolution and partial volume effects between 131I and 133Ba within the cross-comparison.

Computer-aided designs (CADs) for cylinders with four active volumes (Table 2) were produced (Fig. 1a). A wall thickness of 3 mm was used, with no difference in attenuation expected due to the nearly equivalent attenuation of water and resin in the expected energy range (0.078% difference in attenuation for 344 keV [15]). To enable an evaluation of the partial volume effect, different height diameter ratios at a constant height were used. Two versions of cylinder caps were designed, one for containing a resin mixed with 133Ba and one for injection of a solution with 131I. A thread was glued to the bottom of the cylinders to enable mounting the cylinders to a custom-made baseplate using double-threaded plastic rods. Cylinders and caps were produced using a stereolithography (SLA) 3D printing system (Formlabs Form 2) using the Formlabs Tough (v5) photopolymer resin formulation (density when cured = (1.15–1.20) g·cm−3), resulting in a durable and partially transparent model (a useful feature for judging the level when filling). Threads (M3) were added to the injection caps before being fixed into position using epoxy resin and tested to ensure the resulting models were watertight. The designs can be downloaded from the MRTDosimetry data repository [11].

Table 2 Geometries and activities for the cylindrical solid 133Ba sources produced within this project at Laboratoire National Henri Becquerel (CEA) and the Czech Metrology Institute (CMI). Reference date (UTC): 15 December 2018 (12:00). All uncertainties are expanded uncertainties (k = 2)

To optimise the placement of the sources in a standard Jaszczak phantom (cylinder with a fillable volume of 21.6 cm diameter and 18.6 cm height), the collimator-dependent spill-out of counts was estimated by the convolution of a Gaussian function with 20 mm full width at half maximum. This value was chosen to provide a representative worst-case scenario of the reconstructed spatial resolution for 131I based on previous experience with the calibration of SPECT/CT systems with radioiodine in a clinical setting, but was not specifically measured in this study. A laser-cut baseplate for attachment of the sources was produced according to this optimal positioning. For a SPECT/CT measurement, the baseplate and the support rods were mounted in the Jaszczak cylinder, four cylinders (either one of two sets of 133Ba solid sources or liquid 131I sources) were attached to the mounting baseplate, and the Jaszczak cylinder was filled with water.

Solid 133Ba source production

Two sets of 133Ba cylinders (four inserts each) were produced, one at the Laboratoire National Henri Becquerel (CEA) and one at the Czech Metrology Institute (CMI).

Commissariat à l'énergie atomique et aux énergies alternatives (CEA) sources

At CEA, a two-component epoxy resin (Stycast 1264, [16]) was spiked with 133Ba. The amine component was mixed with a limited amount of radioactive aqueous phase, and after mixing with the epoxy component for 30 min, the spiked resin was cured at room temperature (~ 48 h total solidification time). The resulting spiked resin has a density of (1.140 ± 0.011) g·cm−3 (k = 1). The uniformity of the 200 mL batch was assessed by measuring eighteen 2.5 mL subsamples by 4π gamma counting, aliquoted throughout the whole filling process of the four geometries. The dispersion was below 0.1%. The activity of the resin measured by gamma spectrometry for each geometry was in agreement with the spiking value derived from weighing. The relative combined standard uncertainty on the activity in each cylinder is 1.3%, which is below the 2.0% target limit. Following ISO 9978 standard, leakage and contamination tests were conducted: wipe test on the spiked resin, wipe test on the closed vessels, and an immersion test. For the wipe tests, the detection limit reached was below 1 Bq and no activity was detected. For the immersion test, a closed geometry filled with 1 MBq of 133Ba spiked resin was immersed in water at room temperature for up to 4 days. The water was measured by gamma spectrometry, and no activity was measured (detection limit of 0.21 Bq). Figure 1b shows the unfinished sources during production.

Czech metrology institute (CMI) sources

CMI developed a set of reference 133Ba sources with the radionuclide fixed in two-component silicone rubber Lukopren (Lučební závody Kolín, Czech Republic). Drops of 133Ba water solution were added into the liquid rubber and stirred well. Pouring the second component of the rubber resulted in solidification of the solution within a few tens of minutes, fixing the radionuclide in the matrix. Distribution of the radionuclide was uniform within ± 1% in the whole volume of the solidified source. As with the CEA sources, an immersion test according to ISO9978, in Sect. 5.1.4, was applied to measure the radioactivity leakage from the manufactured encapsulated sources. Results of measurement with liquid scintillation analyser Tri-Carb 2910TR (PerkinElmer, USA) were below the detection limit of 1 Bq for all four sources. The relative combined standard uncertainty of the activity of the manufactured sources reached 1.1% and consisted of the uncertainty of the silicone rubber density (1.0%), activity of 133Ba stock solution (0.4%), and weighting (0.06%).

Solid 133Ba phantom preparation

The two sets of sources were sequentially distributed to the participating centres during the comparison exercise. In addition, each participant received a mounting baseplate, support rods, a set of screws, and a set of empty 3D-printed cylinders to be filled with the 131I solution.

Liquid 131I phantom preparation

Two phantoms containing 131I were prepared at each site: a uniformly filled Jaszczak cylindrical phantom to assess the setup-specific image calibration factor (ICF) and a phantom containing the set of four 3D-printed fillable cylinders to assess the partial volume effect.

The 131I activity for the ICF assessment with the Jaszczak phantom was measured in the local radionuclide calibrator before and after injection into the phantom using a traceable 131I calibration factor (dial setting specific to the radionuclide, measurement container, measurement geometry inside the calibrator, and filling level). To ensure a uniform activity distribution of the volatile 131I in the phantom, a carrier solution of sodium hydroxide (0.1 mol dm−3) with inactive iodine (10 μg g−1) was used. The activities of the ICF measurement at the time of SPECT scanning had a median value of 39.9 MBq (range 34.2–86.9 MBq).

For the four fillable cylinders, a stock solution was prepared at each site by weighing a container before and after adding 160 mL of sodium hydroxide (0.1 mol dm−3) with inactive iodine (10 μg g−1) as carrier solution as well as ~ 30 MBq of liquid 131I (from traceable activity measurement as described for the ICF measurement). The activity concentration was calculated as the ratio of dispensed activity to volume. The cylinders were filled by weighing each empty cylinder separately, injecting the stock solution, and re-weighing the filled cylinder. The activity inside each cylinder was then calculated as the product of activity concentration and active volume. The total 131I activity concentrations at the time of SPECT scanning had a median of 0.18 MBq·mL−1 (range 0.16–0.21 MBq mL−1) corresponding to total activities of 0.30, 1.2, 4.9, and 19 MBq in the cylinders with 7.5 (C1), 15 (C2), 30 (C3), and 60 (C4) mm diameter, respectively. As with the 133Ba cylinders, the 131I cylinders were attached to the mounting baseplate with the rods, and the phantom was filled with water.

Data acquisition and reconstruction

The measurements included the ICF determination with the Jaszczak phantom and separate measurements of the three sets of four small cylinders (CEA 133Ba, CMI 133Ba, and liquid 131I) mounted in the water filled Jaszczak phantom. All measurements within the scope of the exercise were performed according to a dedicated SOP, containing information on the required equipment, instructions on energy peak alignment (mandatory), uniformity quality control (optional), phantom filling and positioning, SPECT/CT acquisition parameters, reconstruction and correction methods, delineation of volumes of interest (VOI), and file transfer.

SPECT/CT imaging of the phantoms was performed according to the acquisition parameters given in Table 3. The acquisitions were performed with the phantom (placed with the largest cylinder phantom insert closest to the patient bed) oriented axially using a high-energy collimator and with a standard low-dose CT protocol for attenuation correction. An example of the fully assembled phantom and its positioning on a SPECT/CT system can be seen in Fig. 1c. For 133Ba acquisitions, an energy peak alignment was performed using the smallest 133Ba source. A 60-min acquisition was performed for the 131I ICF measurement, while 30-min acquisitions were performed for all cylinder measurements (CEA and CMI 133Ba, and 131I). The images were reconstructed using an ordered subset expectation maximisation (OSEM) iterative reconstruction with 30 iterations and 2 subsets without post-filtering [17]. Convergence had previously been verified at a representative site. The use of triple energy window (TEW) scatter correction and resolution recovery methods was recommended. However, the comparison exercise also included participants with small differences in the reconstruction software and corrections applied, as in the case of participant 3, which used in-house reconstruction software and the ESSE scatter correction. Details on the reconstruction software and scatter correction method used by each participant are shown in Table 1. Examples of SPECT/CT reconstructions of solid CEA 133Ba sources and 131I-filled cylinders are given in Figs. 1d and e, respectively.

Table 3 SPECT/CT acquisition parametersData analysisAnalysis of the solid 133Ba and the liquid 131I measurements

First, the ICF for 131I was calculated based on the SPECT images of the Jaszczak cylinder as:

$$ICF=\frac_}_\cdot _}$$

(1)

Here, \(_\) is the number of counts in an enlarged CT-based VOI placed around the uniformly filled Jaszczak cylindrical phantom to account for partial volume effect, \(_\) is the acquisition time duration, and \(_\) is the activity in the phantom as measured in the radionuclide calibrator and decay corrected to the time of acquisition.

For each of the cylinder inserts (131I and 133Ba), a pseudo-ICF value, by which ICF values that are influenced by partial volume effects (reduced ICF due to spill-out) will be described hereafter, was calculated as:

$$Pseudo}ICF=\frac_}_\cdot _}$$

(2)

Here, \(_\) is the number of counts in the cylinder VOI, \(_\) is the acquisition time duration, and \(_\) is the activity in the phantom decay corrected to the time of acquisition. The choice of VOI drawing technique was left to the participating sites to reflect the heterogeneity in clinical workflows. However, care was taken in the evaluation to only compare VOIs whose volumes matched the cylinder volumes. While one site chose to draw the VOIs based on the nominal cylinder dimensions (S1), all other sites used a thresholding method to match the physical active volume of the given cylinder (S2-S8).

The combined standard uncertainty in the ICF and pseudo-ICF values was calculated as the square root of the sum of the squared standard uncertainty components, including counts, time, and activity. As the study design required on-site reconstruction, projection data were not available, and therefore, the uncertainty in the counts within the considered volume was assumed to follow Poisson statistics and calculated as the square root of the number of counts [18]. A one-second uncertainty in scan duration was assumed for all the scans acquired. The uncertainty on the radionuclide calibrator measurements included the uncertainty on the calibration setting, reproducibility, linearity, uncertainty due to background correction, uncertainty associated with decay correction, and statistical uncertainty. The sources of uncertainty related to the measurement on the radionuclide calibrators were considered to follow a normal distribution. The weighing uncertainty was also considered for each cylinder, and a rectangular distribution was assumed. The contribution to the standard uncertainties was combined in quadrature to estimate the combined standard uncertainty of the activity dispensed to the cylinders.

The differences between the pseudo-ICFs obtained with the CEA and CMI 133Ba sources were assessed for statistical significance using a nonparametric two-sided Wilcoxon signed-rank test under the null hypothesis of zero median for the differences between the paired CEA and CMI pseudo-ICF values.

Cross-calibration between 133Ba and 131I

The relationship between the pseudo-ICFs for liquid 131I and solid 133Ba was studied based on a setup-specific cross-calibration line, which relates the 131I-based counts measured on a specific system for the cylinder geometry of different sizes to the 133Ba-based counterparts. Due to setup- as well as radionuclide-specific differences, the linear relationship differs from a cross-calibration based on only the ratio of 133Ba and 131I emission intensities. To test the performance of an experimental site-specific cross-calibration, the correlation between 133Ba and 131I pseudo-ICFs for each setup was determined by fitting a non-weighted linear model. The relative percentage change between 131I and 133Ba, corrected with both setup-specific and emission probability-based cross-calibrations, were calculated to assess the performance of both cross-calibration methods. In addition, to assess the potential use of 133Ba sources to calculate a volume-dependent partial volume correction, pseudo-recovery curves (pseudo-ICF against volume) were determined for 131I, uncorrected 133Ba, and 133Ba corrected with both cross-calibration methods by fitting a non-weighted non-linear model.