Introduction

The use of 68Gallium (68Ga)-labelled peptides for PET imaging has increased in the past years with the market authorisation for 68Ga/68Ge-generators. The main applications include imaging of neuroendocrine tumours using somatostatin analogues and prostate cancer imaging using the prostate-specific membrane antigen [1, 2]. Though the interpretation of 68Ga-PET/CT is mainly based on visual assessment, quantitative measures should be used to evaluate or predict therapy response.

Previous experience with 18Fluorine (18F) expressed the need for standardisation of acquisition and reconstruction protocols in order to retrieve comparable quantitative imaging data. The EANM Research Ltd. (EARL) provides an accreditation programme to ensure PET/CT system harmonisation in multicentre 18F-FDG PET/CT studies [3]. This approach is based on standardizing the recovery coefficient (RC) for six phantom spheres with different sizes, thereby minimising inter- and intra-institute variability. For other isotopes, quantification should be evaluated separately as isotope characteristics can result in different image quality and quantification accuracy. For example, Makris et al. studied 89Zirconium (89Zr) PET and showed the need for a specific harmonisation step including post-reconstruction smoothing to enable comparable quantitative measures among PET/CT systems [4]. In contrast, a recent 18F performance study showed that post-reconstruction filtering is not required for state-of-the-art PET/CT systems in relation to this isotope [5]. However, for 68Ga, such studies are not yet available.

In general, PET quantification accuracy depends on reconstructions, noise, and spatial resolution [6]. For 68Ga, the lower positron yield (89%), long positron range due to high initial positron energy (max 1.90 MeV, mean 0.84 MeV), short physical half-life (68 min) and small prompt gamma branching (3.2%, 1.077 MeV) may result in an inferior image quality compared to 18F [7]. Therefore, the aim of this study was to assess 68Ga-PET/CT quantification accuracy and reproducibility in a multicentre setting based on EARL standards.

Materials and methods

Clinical protocol evaluation

A survey among eight Dutch hospitals was performed to evaluate factors that affect quantification and to assess variability in clinical 68Ga-PET/CT acquisition protocols. Questions focussed on administered activity, PET/CT system, and acquisition- and reconstruction settings.

18F and 68Ga PET/CT phantom acquisitions

Eight European hospitals with 13 PET/CT systems performed phantom acquisitions, of which 11 systems were EARL accredited, but all had recoveries within the published EARL specifications. Six Biograph mCT systems (Siemens Healthineers, Erlangen, Germany), three Discovery systems (GE Healthcare, Milwaukee, WI, USA) and four Philips systems (Philips Healthcare, Eindhoven, The Netherlands) were included.

18F and 68Ga acquisitions were performed at the end of 2017 and beginning of 2018 with two phantoms which were prepared using a standardised procedure by experienced staff from each centre. First, the NEMA PET cylindrical phantom was filled with 6–13 kBq/ml of 18F and 68Ga. Second, the NEMA NU-2 Image Quality (IQ) phantom was imaged using a 1:10 ratio with 2.0 and 20.0 kBq/ml of 18F and 68Ga in background compartment and spheres (37, 28, 21, 17, 13, and 10 mm diameter), respectively. Acquisitions of both phantoms were performed with minimal two bed positions and at least 5 min per bed position. Images were reconstructed according to local settings, including corrections for decay, randoms, dead time, CT-based attenuation, and scatter.

Data analysis

Image noise was characterized for 68Ga only using the coefficient of variation (CoV) along a 30 × 30 × 160 mm bar in the centre of the cylindrical phantom. Image quality was based on the RC of all six spheres, analysed by the EARL semi-automatic tool [5, 8]. The RCmax, RCpeak and RCmean were determined as a function of sphere size based on the maximum voxel value (RCmax), the 1.0 cm3 volume with the maximised average value (RCpeak) and the mean value of 50% isocontour of the maximum voxel value (RCmean) with contrast correction, respectively. A spherical volume-of-interest (VOI) of ~ 300 ml in the centre of the cylindrical phantom and ten VOIs in the background of the IQ phantom were used for local PET and dose calibrator cross-calibration. IQ phantom background volume was 9400 ml, unless specified otherwise by the institute.

Results

Eight Dutch hospitals provided their clinical acquisition- and reconstruction protocols (Table 1), which showed to be different.

Table 1 Acquisition and reconstruction settings of clinical 68Ga PET/CT imaging for prostate cancer and neuroendocrine tumours. One hospital per row is presented
Full size table

An overview of all PET/CT systems and reconstruction settings is provided in Table 2. For local cross-calibration, most systems performed within 10% deviation of the dose calibrator (Fig. 1). The median [IQR] ratio was 0.93 [0.91–0.98] and 0.99 [0.97–1.01] for 68Ga and 18F, respectively. Two systems showed identical calibration accuracy for both isotopes (system 2 and 11), all other show a consistent underestimation for 68Ga. The 68Ga CoV in the centre of the cylindrical phantom was below 10% (Fig. 2).

Table 2 PET/CT reconstruction settings for phantom measurements
Full size table
Fig. 1
figure 1

Accuracy of the measured activity by the PET/CT system and local dose calibrator, based on the average between the cylindrical and IQ phantom. Numbers correspond to Table 2

Full size image
Fig. 2
figure 2

Noise across the cylindrical phantom filled with 68Ga, visualized as coefficient of variation (CoV)

Full size image

The 18F RC-curves of all PET/CT systems satisfied the current EARL specifications (Fig. 3a–c). However, for 68Ga the RC-curves were located around the lower limit of the EARL specifications (Figure 3d-f). In addition, 68Ga showed a reduced mean recovery and larger variation between PET/CT systems compared to the 18F. The variation for all spheres of the RCmean, RCmax and RCpeak for 18F was 6%, 6% and 8%, respectively. For 68Ga, the mean range was 11%, 11% and 15% (largest variation was 19%). Furthermore, the mean RCmax and RCmean were both 11% lower compared to the mean EARL specifications for 18F. The mean 68Ga/18F calibration difference within one scanner was 7% (range 1–13%).

Fig. 3
figure 3

RC for 18F with the current EARL standards and RC of 68Ga. Solid lines: maximum and minimum values according to EARL limits as applicable before 2019

Full size image

After correction for the local difference between 68Ga/18F cross-calibration (Fig. 1), the 68Ga RC curve was within EARL limits for all but two scanners (Figure 4). The mean 68Ga RCmax and RCmean were accordingly 5% lower compared to mean EARL standards.

Fig. 4
figure 4

68Ga RC-curves corrected for the 18F/68Ga calibration mismatch according to local cross-calibration. Solid lines: maximum and minimum values according to EARL limits as applicable before 2019

Full size image

Discussion

In this study, quantitative 68Ga PET/CT performance was evaluated in a multicentre setting. In a survey across Dutch hospitals, differences in clinical acquisition and reconstruction protocols were observed, underlining the need for clinical harmonisation. Although 11 out of the 13 PET/CT systems were EARL accredited, all systems showed 18F recovery performance within EARL standards. For this reason, all systems were included for 68Ga evaluation.

The absence of local and central dose calibrator cross-calibration for 68Ga is a limitation in this study. This would increase local calibrator harmonisation and improves PET/CT comparability across sites. Most institutes use a long-lived (137Ceasium) source to assess constancy and accuracy of the dose calibrator on a daily basis, and perform actual cross-calibration with the PET/CT system at least once a year using 18F. Still, in all but three PET/CT systems the measured 18F and 68Ga activity concentrations were within 10% deviation from the local dose calibrator. High energy prompt gammas emitted by 68Ga are likely detected by the dose calibrator causing a disconcordance, yet in fewer extent by the PET system. Because of this, the dose calibrator overestimates 68Ga-activity, and a persistent underestimation for 68Ga compared to 18F is seen in Fig. 1. A recent study by Bailey et al. also showed an underestimation of ± 15% for 68Ga, which was primarily related to an inaccurate scaling factor for the dose calibrator of a specific vendor [9]. To avoid these issues, they calibrated the dose calibrator towards the PET, after verifying that the scanner has a good response for 18F. These results are also supported by the fact that on specific Siemens scanners (scanners 1 and 2), a traceable 68Germanium (68Ge) source was used to verify absolute PET response independent of a dose calibrator. When imaging the 68Ge-source, the PET/CT system did not show the same offset as was observed when imaging the 68Ga cross-calibration phantom (roughly a deviation of < 1% vs. 6% and 7%, respectively). For the sake of simplicity, we would suggest to correct the RC curve for the local 68Ga/18F discrepancy, as after correction for this 68Ga/18F difference (Fig. 4) all but two scanners were within EARL specifications. This correction has to be performed offline in multicentre quantitative studies. The 68Ga used for this study was produced either locally or by a pharmaceutical institution and was therefore not traceable to a central dose calibrator. We expect that the response between the dose calibrator and the PET-system could be uniform in future clinical 68Ga-PET/CT studies if a traceable (NIST) source is used to harmonise protocols between centres.

68Ga image noise was below 10% for all PET/CT systems which is in concordance with the EANM/EARL guidelines [3, 8]. The RC variation is larger for 68Ga compared to 18F (Fig. 3). However, 68Ga performance nearly reached EARL performance specifications after correction for the local 68Ga/18F ratio. Surprisingly, the RCpeak variation (8% and 15%) is larger in contrast to RCmax and RCmean (both 6% and 11%) for both 18F and 68Ga, respectively. The study of Kaalep et al. showed the opposite result in RCpeak variation [5]. The RCpeak is expected to be less prone to noise compared to RCmax; therefore, it was expected to be more comparable over all PET-systems. The difference could be explained by the fact that the standard deviation of RCmax and RCpeak are similar: 8.4% and 8.6% for 68Ga and 4.8% and 5.0% for 18F, respectively. Yet, the mean RCpeak value is lower; therefore, resulting in a higher CoV. Next to that, the larger 68Ga variation in the RC-curves compared to 18F is likely related to the higher positron energy of 68Ga and thereby revealing a lower signal-to-noise ratio. This effect is enhanced by post-reconstruction filtering. Finally, previous single-centre studies show 68Ga RC-curves similar [10] or somewhat better due to point spread function reconstruction [11] as observed in the current study. The EARL limits as applicable before 2019 (EARL1) are shown in Figs. 3 and 4, as all acquisitions were acquired before 2019 and therefore site-specific acquisition and reconstruction protocols are designed to meet the EARL1 limits. RCpeak specifications are not available for EARL1 and are therefore not shown in Figs. 3 and 4. EARL2 limits (applicable from 2019) for RCmax and RCmean increased with ~ 25% in comparison to EARL1. We expect that the gap between 18F and 68Ga recoveries will further increase with these new limits, as already for EARL1 not all scanners agreed to EARL1 limits after 68Ga/18F correction (Fig. 4).

Based on the results, we propose to correct 68Ga recovery towards the 18F recovery to correct for the current dose calibrator deviation. We suggest, therefore, to apply the EARL acquisition and reconstruction protocol and to correct for 68Ga/18F cross-calibration mismatch. One can assume that 68Ga recovery is steady if 18F specifications of a PET-system are stable during regular yearly assessment. Unless the acquisition and reconstruction protocol is changed or major maintenance is performed to the PET/CT-system, we recommend to perform additional 68Ga IQ acquisitions only when regular 18F evaluations are deviating. An EARL accreditation programme for 68Ga can thus be based on the 18F accreditation but extended with a cross-calibration verification between 68Ga measured by the dose calibrator and PET/CT system only, similarly as proposed by Kaalep et al. for 89Zr [12]. In addition, frequent 18F cross-calibration acquisitions using the cylindrical phantom are advised, especially after PET/CT system maintenance.

Conclusion

This evaluation of multicentre 68Ga PET/CT performance showed that 68Ga RCs perform at the lower limits of current 18F EARL standards. For practical reasons, we recommend to use the 18F EARL approved reconstruction settings and to correct for 68Ga/18F calibration mismatch based on local cross-calibration. Finally, we suggest to evaluate 68Ga PET/CT recovery performance once and repeat only when 18F specifications are changed.