The TARGET phantom Sub-Study aimed to evaluate inter-site variability in [99mTc]Tc-MAA imaging and specifically, to report on variability in two dosimetric quantities (LSF and T:N due to technical and procedural differences between sites. To fully leverage the benefits of [99mTc]Tc-MAA scout imaging, quantitative metrics derived from [99mTc]Tc-MAA SPECT should be comparable between scanners, sites and studies. This study demonstrates that the use of key image corrections, specifically AC and SC, significantly reduced inter-system variability, whilst standardization of other reconstruction parameters (iterations, subsets and post-filtering) did not improve consistency.

Of the two metrics considered in this study, the LSF investigation exhibited better consistency between sites. Results showed the LSF was overestimated by approximately 8.8% using the site-specific protocols and the LSF IQR between different protocols was 9.6–10.1. Greater variability was noted in the tumour to normal tissue investigation, where CRC was demonstrated to vary substantially when site-specific imaging protocols were used. As an example, for the largest sphere in the NEMA IQ phantom, the CRC IQR was 0.5–0.7 and the two most differing sites recorded a CRC 0.35 and 1.01 respectively, meaning that if the same patient were scanned in two participating centres, the outcome in apparent tumour absorbed dose could differ by more than a factor of two. This demonstrates the potential variability that can be expected when sites use different imaging systems and methods.

An additional aim of the study was to investigate the impact of a harmonization strategy, involving implementation of a standardized imaging protocol. Firstly, considering the standardised protocol for the LSF investigation, the LSF as measured on the standardised protocol was underestimated by 5.0%, less than that noted for the site-specific protocols, however the IQR was not reduced (8.4–9.0). This indicates that the standardised protocol did not have a positive impact on variability but did improve accuracy. For the tumour dosimetry investigation, imposing the standardized protocol was evidenced to improve average performance. For the majority of centres (70%), the average CRCs of the largest sphere demonstrated a positive bias compared to site-specific protocols, and the S:B error was reduced, indicating accuracy was improved. The IQR however was not improved, as an example, the largest sphere CRC IQR increased from 0.16 (0.53–0.69) for the site-specific protocols where no specific corrections were imposed, to 0.21 (0.61–0.82) for the standardized protocol. It is evident, that implementing a single, standardised protocol does not necessarily reduce variability, as it is still necessary to account for the different properties of the collimators and the reconstruction algorithms for the various cameras. A key finding was that eliminating sources of possible variation in image corrections substantially improved inter-system quantification variability. For the subset of sites that provided both site-specific and standardised datasets, and who largely included attenuation and scatter correction as part of their site-specific protocols, the initial large variability in recovery coefficients was reduced. By removing inconsistency in only two parameters (AC and SC) the IQR in CRC for the 37 mm sphere halved from 0.2 to 0.1.

Our results demonstrate that technical factors have a non-negligible impact on [99mTc]Tc-MAA image-based dosimetry and dose targets reported in multi-centre trials should be interpreted in this context. The benefits of defining specific dose targets when imaging practice remains markedly inconsistent is inherently limited. Efforts to define accurate dose thresholds must be matched by efforts to standardise imaging practice to maximise efficacy.

A simple step that may be taken to maximise consistency between centres, is to ask centres to perform key image corrections (i.e. scatter correction and attenuation correction). Since the site-specific subgroup of images that all applied AC and SC, were in fact more consistent than those obtained from the standardised protocol, this would suggest there is merit in investigating the implementation of image corrections and procedural guidelines but leaving the specific application to the discretion of individual centres with greater insight of their own imaging system.

Whilst the standardized protocol did require centres to perform these corrections, the protocol was prescriptive and did not leave much room for centres to optimise, leading to some inconsistent behaviours between different imaging systems. There were several factors that could introduce inconsistencies, for example, the standardized protocol stipulated that a 5 mm gaussian filter be used for post-filtering, however these filters may be implemented differently in the various reconstruction algorithms used by the respective vendors. Similarly, the energy window width in the standardized protocol was required to be 15% (140 keV ± 7.5%). Many centres had a default of 20% window width and thus, in order to adhere to the standardized protocol, the energy window settings were changed. For some scanners this would also require the scanner to be peaked. If the peaking procedure was skipped, this would result in poorer quality images. Finally, by tuning of the number of iterations, site-specific protocols typically balanced image noise and reconstruction time at the expense of a lower CRC. Since convergence rates vary between the various reconstruction algorithms for individual imaging systems, the standardized protocol utilized a relatively high number of iterations to assure full convergence and optimize the CRC. However, this resulted in high noise levels in several cases. A key takeaway therefore is that close agreement between sites via implementation of a standardised protocol for SPECT is only partly relevant, and potentially only pertinent for the acquisition parameters. Due to the different reconstruction algorithms and collimator specifics, a better harmonization approach could involve focusing on the CRC metric itself, and tuning reconstruction parameters accordingly leading to different settings for different cameras. Ideally this should be facilitated through a central entity to analyse the data, an approach which has been successfully demonstrated in the EARL initiative. A similar methodology, of implementing a standardised acquisition protocol and performing reconstruction centrally, has been successfully implemented for SPECT quantification of 177Lu [17] and 99mTc [18] in phantoms.

This study has several limitations. Firstly, site specific differences in phantom preparation and activity dose measurement likely contributed to variability in the results, however as a specific preparation protocol was made available and the selected phantoms are standard in the field, this variance is expected to be small as compared to the objective of this investigation (i.e., the variability in acquisition and reconstruction). The LSF was inferred from two phantoms that were both homogeneously filled with activity and water, representing the ‘liver’ and ‘lungs’. In reality lungs are much less dense, activity is non-uniformly distributed, and the liver and lungs often overlap in the planar view, which in practice results in substantial overestimation of the LSF. For this reason many centres consider SPECT/CT as an alternative to planar imaging (alternative methods using SPECT/CT have been investigated and proven to be superior [19,20,21]). Finally, the standardized protocol for this Phantom sub-study was designed considering a system from a specific vendor, individual cameras from the various other imaging vendors implement different reconstruction algorithms and thus, a standardized protocol based on one imaging system was not capable of encompassing all scanner processes.

The dataset collected in this work, encompasses a wide range of system types and [99mTc]Tc-MAA SPECT/CT imaging protocols and thus provides a representative insight into the large variability evident in the field. Future research should build on existing works [12] to establish an evidenced-based standardised practice of acquiring and reconstructing [99mTc]Tc-MAA SPECT/CT images for the purpose of pre-treatment dosimetry. The harmonization of imaging procedures is now endorsed by several professional societies and organizations [22,23,24]. Much focus has been given to reducing variability of PET image quantification in multi-centre settings i.e. The EARL initiative [25], and more recently the AAPM launched a scheme to enhance consistency in 90Y Bremsstrahlung imaging, again based on quality control procedures using phantoms for standardisation [26]. However, as yet there is no standardisation programme for pre-treatment [99mTc]Tc-MAA SPECT/CT dosimetry. Investment in an accreditation scheme similar to EARL for [99mTc]Tc-MAA SPECT/CT dosimetry would be a valuable future endeavour to help advance the use of quantitative [99mTc]Tc-MAA SPECT/CT imaging. In the interim, publications on dose–effect relationships and reported dose thresholds should comment on their centre-specific imaging factors (e.g., system type, protocol parameters, image quality as measured via CRC) so that other centres may put results into context before implementing clinically.

In conclusion, this study shows that quantification of [99mTc]Tc-MAA SPECT/CT is feasible in a multi-centre phantom study, and high quality clinically relevant data can be obtained. Over the range of cameras and site-specific planar protocols investigated, comparable performance was noted in the lung shunt investigation, which suggests suitability for quantitative analysis of LSF in a scenario analogous to that of pre-treatment dosimetry work-up. Site-specific SPECT protocols included in this study were not capable of consistently reconstructing [99mTc]Tc-MAA activity distributions and there were large differences in CRC between different protocols for the same size structure. By eliminating sources of difference in image corrections between protocols, variation in quantification was reduced. A subset of site-specific protocols that implemented key image corrections (AC and SC) had a reduced range compared to the full site-specific dataset. The standardised protocol did not improve consistency between sites in either the LSF investigation or tumour dosimetry investigation but did improve accuracy.