PET is a widely adopted molecular imaging modality, being sensitive, inherently quantitative and able to image a wide range of molecular processes. It has important established clinical and research applications. In oncology, particularly using 2-fluoro-2-deoxyglucose (FDG) it has proven application for diagnosis, staging, treatment response and detection of relapse which results in significant changes in disease staging, treatment modality and intent (8, 9). In addition to FDG there are now a wide range of other PET radiopharmaceuticals in clinical and research use. Quantitation in PET is of particular importance as there is increasing evidence that quantitative parameters predict outcome (10) and is essential for biodistribution and dosimetry calculations for both clinical and research applications (11, 12).
Accurate AC is fundamental to high-quality, highly sensitive and accurate quantitative PET imaging. Initially, AC for PET was performed with sealed line sources (13), however approximately 20 years ago Beyer and colleagues described development of the first combined PET CT scanner, using the CT component for generation of attenuation maps for PET reconstruction (1). Combined PET CT is now widely accepted as superior to PET alone (2) and is now standard of care for PET imaging. In addition to providing AC, the CT component of the PET CT study may be used for anatomic localisation or diagnosis. Current practice guidelines recommend selecting CT acquisition parameters depending on the intended use, with recommendations for reduced voltage and/or current when the CT is used for AC only and with higher exposures required when CT is also used for diagnosis (14). In most clinical and research purposes, in addition to AC, the CT component of a PET/CT study will also be used for anatomic localisation or diagnosis. However, in select circumstances the CT component may only be required for AC, particularly when multiple repeated AC CTs are required over a short period of time. One example of this is for gated studies where an AC CT is required for each phase of the gated study. Another situation where CT may only be performed for AC is early phase human biodistribution and dosimetry studies where PET/CT scans are performed repeatedly in the same subject over a short timeframe, typically hours (12, 15). In this setting, accurate AC maps are required for accurate quantitation, however in the short interval between scans no anatomic change would be expected and thus the CT component would serve no other purpose. In these settings, the repeated acquisition of CT may contribute more to the overall patient radiation exposure than the injected radiopharmaceutical, and thus it is essential that radiation exposure from the CT component be minimised according to the ALARA principle. This study was undertaken to establish the minimum CT exposures and optimise reconstruction parameters for accurate biodistribution and dosimetric assessment in preparation for a first in human study of a novel PET radiopharmaceutical for imaging cell death (16).
There is limited data regarding optimisation of the CT component of the PET CT acquisition. Recently, Bertolino and colleagues undertook a systematic review of CT protocols performed within a PET CT scan. Their rationale for undertaking this was the observation that unlike the PET acquisition, there is a lack of robust scientific literature regarding the optimisation of CT protocols used in PET CT. They concluded that dose is heavily dependent on the protocol intent (AC, anatomic localisation, or diagnosis). They did not conclude on specific parameters for CT acquisition within a PET CT rather suggesting periodic quality control considering technological advances (17). There is very little data regarding dose optimisation of CT performed for AC. Faye and colleagues used five different anthropomorphic phantoms (newborn to medium adult) to assess the impact of acquisition parameters on CT image noise and adequacy of PET AC. They reported that significant dose reductions could be achieved, reporting that in paediatric patients adequate AC could be obtained with very low dose and with only an increase in tube voltages required to prevent under correction in adults. There are several differences between this previous study and this study. Firstly, Faye and colleagues assessed the adequacy of AC for PET quantitation qualitatively by visual inspection of the images – no quantitative analysis was performed on the reconstructed PET images (although this was undertaken on the CT AC map). Secondly, the study was performed on a PET CT scanner without capability for IR of CT (18).
Brady and Shulkin undertook a phantom and retrospective patient study to assess ultralow dose CT protocols reconstructed using adaptive statistical iterative reconstruction (ASIR) on PET and CT image quality and quantitation. With this protocol they reported no change in SUV, background uniformity or spatial resolution of PET with up to 90% dose reduction and that there was an average deviation of only 2% for all cylindrical/spherical target lesions. In contrast to the current study, regions of interest were not considered outside of the target lesion (beyond background uniformity) and the scanner was from a different manufacturer with a different IR algorithm (19).
The paucity of published literature, and the absence of any specific data related to equipment at this institution or the specific application of quantitative imaging for first in human biodistribution, radiation dosimetry calculation and imaging, was the impetus for undertaking this study. With the intent of doing whole-body biodistribution studies, accurate quantitation at all sites (not just lesional sites) is essential and hence the region of interest analysis assessed both lesional and non-lesional regions. Voxel analysis of the reconstructed PET datasets were similarly undertaken to provide the broadest insight into subtle quantitative changes throughout the study.
In this study, determining what level of change to regard as significant is difficult and contentious and is dependent on many factors including technical, biologic and physical (20). Both proportional and fixed changes were considered, with advantages and disadvantages to both approaches. Using a proportional change in activity relative to the concentration in the ROI may be appropriate as SD is higher with higher activity concentrations. However, proportional changes are also problematic as at lower activity concentrations proportional changes deemed significant may be of such a small absolute magnitude that they are not relevant in terms of diagnostic and dosimetric quantitation. At high activity concentrations relatively small proportional changes below the level deemed significant may still have significant impact on clinical and research quantitation. Ultimately a fixed threshold change equivalent to 1 SD of SUV measured in the ROI within the liver was deemed significant, which equates to a change in SUV of 0.1 or ~4% of the mean SUV of the liver. It is acknowledged that this is a small change and in isolation would not be regarded as significant. However, to enable accurate comparison between studies whether performed for clinical indications (such as for assessment of treatment response following commencement of therapy) or for biodistribution and dosimetry calculations, Boellard described a wide range of factors which can affect PET quantification and stresses the importance of standardisation to minimise variability and improve accuracy of quantification. In particular, Boellard identified reconstruction parameters has a potential source of variability of up to 30% (20). In defining the PERCIST 1.0 criteria, Wahl and colleagues use the SD of uptake within the liver in the formula for calculations both before and following treatment, and in addition state that liver SUV should generally be within 0.3 from study to study and much of this variability will be accounted for by biologic factors. Hence, an SUV change of 0.1 is approximately 30% of the expected interstudy reproducibility of mean liver uptake (21). Given that this study was used to define parameters for multiple time point scanning for a first in human study it was considered particularly important that a level of significance be set that was consistent with established clinical and research standards such as defined in PERCIST 1.0.
Both ROI and individual voxel analysis demonstrates that at very low CT exposures there is a systematic underestimate of mean SUV in all ROIs, which is greater when IR is used compared to FBP. With increasing CT exposure mean SUV in ROIs converges to the mean SUV obtained from the RR. When CTDI reaches 1mGy the difference is insignificant between TR and RR irrespective of whether IR or FBP are used on both ROI and voxel analysis. Based on ROI analysis but not voxel analysis, CTDIs less than 1mGy still result in differences of mean ROI SUV less than QSD between TRs and RR when reconstructed with FBP. However, as this study was undertaken as a prelude to a quantitative first in human biodistribution study, it was considered that both ROI and voxel analysis should not differ significantly between the RR and TR. Based on this, CT parameters chosen for AC for PET for the first in human study was 25mAs, 100kVp, pitch 0.828, FBP which delivers a CTDI of 1mGy.Compared to the RR CTDI of 3.3mGy this represents an approximately 70% dose reduction.
In CT scans performed for diagnosis, IR has been reported to result in improved image quality while reducing CT dose, however unexpectedly, it was observed that at the lower CT exposures IR of the CT used for AC of PET resulted in greater underestimation of activity compared to when FBP CT reconstruction was used for AC of PET. This is contradictory to that observed by Brady and Shulkin who also observed no change in image noise (19). In this study image noise expressed as SUV COV was much more variable at low CT exposures; in some regions SUV COV was greater with IR and in other regions SUV COV was greater with FBP. SUV COV was similar at higher CT exposures regardless of reconstruction algorithm. Given this variability, it is unlikely that image noise alone is the explanation for the greater SUV underestimation with IR. Other possible explanations include differences in the IR algorithms and subsequent generation of segmented CT AC maps. The PET CT scanner used in this work has the option for differing levels of noise suppression (iDose levels) applied to IR and further work is warranted to investigate these. However, this study has demonstrated a 70% reduction in CTDI is possible without compromising quantitative accuracy. Further investigation of different iDose levels would be beneficial to see if further dose reductions are possible however the further gains would be relatively modest. Additionally while use of IR at very low dose CT exposures may potentially enable better CT image quality, given the very low CT exposure even if further image quality improvements can be made it is unlikely the improvement will enable more than anatomic localisation which can already be adequately achieved with FBP CT images. These differences between this study and Brady and Shulkin’s study highlights the need for scanner and reconstruction specific assessment and periodic quality control audit as suggested by Bertolini and colleagues (17).
In conclusion, this study demonstrates the impact of CT acquisition parameters and reconstruction algorithms on AC for PET reconstruction and identifies appropriate parameters and algorithms to minimise exposure when CT is performed only for AC of PET studies on the Philips Ingenuity TF scanner. More generally it demonstrates a method for assessment of the impact of CT acquisition parameters and reconstruction algorithms on quantitative accuracy of PET reconstructions that is broadly applicable to all PET CT scanners to enable scanner and reconstruction specific CT dose optimisation.