The Netherlands protocol for standardisation and quantification of FDG whole body PET studies in multi-centre trials

Positron emission tomography (PET) using 2-[¹⁸F]fluoro-2-deoxy-d-glucose (FDG) has become an important tool for diagnosis, staging, prognosis and treatment-response monitoring in oncology [1]. Vansteenkiste and Stroobants [2] reviewed the role of FDG-PET in respiratory oncology. In their paper, the authors addresses various aspects of FDG-PET in oncology and discuss amongst others that improved mediastinal lymph node staging of patients is obtained using PET compared with computed tomography(CT) alone. De Geus-Oei et al. [3] performed a systematic review on the predictive and prognostic value of FDG-PET in non-small cell lung cancer and concluded that the degree of FDG uptake is of prognostic value at initial presentation, after induction treatment before resection, and in case of recurrence. Weber et al. [4] studied the use of PET for monitoring cancer therapy and for predicting outcome. In his paper, the author addresses many issues involved in quantitative assessment of therapy-induced changes in FDG tumour uptake. To date, there are numerous papers on the promising role of FDG-PET in oncology [5‐9].

While the important potential of FDG-PET in oncology has been demonstrated in many investigations and papers, many of these studies discuss the need for standardisation, especially for multi-centre trials and for meta-analyses. Standardisation becomes especially important when quantification of FDG-PET studies using standardised uptake values (SUV) is performed [10‐12]. Although visual inspection of FDG-PET images remains very important for diagnosis and response assessment, it has been shown that semiquantitative analysis (SUV) allows an objective assessment for lesion characterisation [13], prognostic stratification [3] and monitoring treatment response [14]. The latter is generally measured by the relative change of SUV during treatment. At the same time, comparison of SUV results obtained from different centres is hampered by the disperse variability in methodology of acquisition, image reconstruction and data analysis procedures applied. Boellaard et al. [10] showed that many technical factors, such as image reconstruction settings and region of interest strategies, can have a major impact on the outcome of SUV. Jaskowiak et al. [15] showed the impact of OSEM reconstruction settings on SUV. Stahl et al. [16] and Krak et al. [17] evaluated the effects of various clinical and technical factors on test–retest variability and absolute outcome of SUV and the impact of using various SUV measures on observed treatment response. Recently, Westerterp et al. [12] evaluated quantification of FDG-PET studies in multicentre trials with a focus on methodological variability among centres. These and other studies showed the need for harmonisation of FDG-PET across centres.

The need for standardisation was already addressed in 1999 by a position paper of Young et al. [18]. In that paper, various methods for quantification of FDG-PET studies, such as visual inspection, use of SUV or full kinetic analysis, were discussed. In addition, factors affecting FDG uptake are described. Taking into account these factors, a set of recommendations was presented. More recently, in 2006, Shankar et al. [19] published the concensus recommendations for FDG-PET studies as indicator of therapeutic response in patients in National Cancer Institute Trials. The guidelines given in the latter paper focus on patient preparation, image acquisition, image reconstruction, quantitative and semiquantitative image analysis, quality assurance, reproducibility and other factors in FDG-PET studies before and after a therapeutic intervention. Larson and Schwartz [20] stated that these NCI guidelines provide an excellent starting point for standardisation of the use of FDG-PET in clinical trials. Other authors, such as Zijlstra et al. [21] and Juweid et al. [22], stressed on the need for harmonisation of interpretation of FDG-PET studies in lymphoma either using visual inspection and/or using relative changes in SUV. More guidelines are being developed as technology moves forward [23]. Recently, Delbeke et al. [24] therefore provided procedure guidelines for tumour imaging with FDG PET-CT.

Although many issues are addressed in the aforementioned papers, there is still a need for standardisation of FDG-PET studies in case of multi-centre studies focussing on quantification using SUV. Therefore, a PET imaging working group was formed by the Dutch Society of Haemato-Oncology (HOVON, Hemato-Oncologie Vereniging voor Volwassenen in Nederland) and by Nuclear Medicine Physicians, representatives of the Dutch Society of Nuclear Medicine, and Medical Physicists, representatives of the Dutch Society of Nuclear Medicine and Dutch Society of Clinical Physics, with the task of formulating a protocol for standardisation and quantification of FDG Whole Body PET studies in multicentre trials in The Netherlands. The protocol specifically aims at standardisation of FDG-PET studies in order to allow inter-institute interchangeability of SUV. The first version of the protocol was finalized in November 2007 and was approved by the Dutch Society of Nuclear Medicine. The purpose of the present paper is to provide a description of the protocol and the rationale for recommendations given. First, factors affecting SUV outcome will be summarized shortly as these have been addressed extensively elsewhere. Next, guidelines and recommendations will be described in the main part of the paper. Finally, limitations, pending issues and future work will be addressed.

Various factors and sources of error affect SUV outcome. These factors can be assigned to physiological effects, sources of error and physical effects.

Physiological factors affecting SUV are, for example, the plasma glucose level during FDG-PET scanning, FDG plasma clearance during the phase of FDG uptake, the scanning period and the effects of patient motion and breathing. As FDG uptake is time dependent, the time interval applied between FDG administration and execution of the PET study will affect SUV as well. For response monitoring, the time interval between the baseline FDG-PET study and start of treatment as well as the time intervals between subsequent FDG-PET studies and cycles of treatment are of importance. This is especially important when inflammatory processes may occur as may be the case during or after radiation therapy. These factors and their impact on SUV outcome are summarized in Table 1.

Table 2 presents some of the most common errors affecting SUV, as also described by Weber et al. [4] and Shankar et al. [19]. Correct calculation of SUV depends on an accurate (cross-)calibration between PET or PET-CT scanner and the dose calibrator used for measuring patient dose. Moreover, the net dose given to the patient needs to be known exactly. Thus, any remaining activity in the syringe after administration should be minimized to within 1% using appropriate procedures. Alternatively, remaining activity in the syringe may be measured after the administration. In the majority (95%) of the cases, remaining activity in the syringe is smaller than 3% of the specified dose, although occasionally (5% of the cases) remaining activity was more than 10% (unpublished data), which mainly occurred when specific activities are still high (i.e., shortly after production). The effects of dead volume in the syringe preparation system and administration system should thus be carefully taken into account. A subtraction method to derive the amount of activity in a syringe may be used. The latter is generally based on subtracting a small amount of FDG from a stock, positioned in a dose calibrator, and measuring the amount of activity of the stock prior and after withdrawal of the patient dose. If the stock contains a large amount of activity relative to that needed for the patient dose, care has to be taken that dead volume or statistical errors do not affect accuracy and reproducibility of net dose assessment. With proper procedures, however, this can be handled easily. A frequently overlooked problem may arise from an incorrect synchronisation of clocks in a PET system with those of the dose calibrator used (or computer attached to it for read-out and/or) or with clocks in various rooms/locations. Moreover, as the patient dose is specified at a certain dose calibration time, which is generally not exactly equal to the actual injection time, correct decay corrections have to be applied. This means that decay between calibration time and PET acquisition time needs to be applied to the dose specified at the dose calibration time. Obviously, paravenous injection may still allow visual interpretation of the PET study, but will make quantification based on SUV incorrect. A reproducible and correct administration of FDG is required.

Physical or technical factors affecting SUV outcome (Table 3) are described in various papers. Data acquisition settings such as time per bed position, the amount of overlap between subsequent bed positions, acquisition mode (2D or 3D) and FDG dose affect scan statistics and/or the noise equivalent count level of PET studies. Poorer scan statistics and lower image signal-to-noise ratio (SNR) will result in an upward bias of SUV, especially when using the maximum SUV as final outcome parameter [10].

Nowadays, attenuation corrections will be mostly derived from a dedicated CT scan (CT-AC) acquired on a PET-CT scanner. Patient motion (e.g., breathing) may result in a mismatch between PET and CT and thus result in incorrect attenuation correction. The latter may be minimized by breathing instructions (breath holding at mid-inspiration volume or shallow breathing) during CT scanning. Use of respiratory gating would be beneficial, but is not yet routinely available in most centres. Use of contrast agents may affect accuracy of conversion of CT-AC image Hounsfield units into 511 keV linear attenuation correction factors and thereby accuracy of SUV. Hopefully, sophisticated algorithms to perform this conversion will be introduced allowing for use of contrast agents, as will be discussed in more detail later. Finally, attenuation correction based on CT could be affected by truncation of the CT image. Note that CT truncation does not only affect the quantification of PET at the position of CT image truncation alone, but may affect SUV of tumours located elsewhere in the same axial planes as well.

Image reconstruction methods and settings determine both quantitative accuracy and final image resolution. With iterative reconstruction methods, a sufficient amount of iterations (or the product of the number of iterations and subsets) need to be applied to ensure sufficient convergence of the algorithm. Insufficient convergence will result in an object- (and/or surroundings) dependent SUV outcome [15]. In addition to these settings, generally some filtering is performed during or post-reconstruction. Typically, Gaussian filters of 5 to 10 mm full width at half maximum (FWHM) are applied resulting in a clinical image resolution of 7 to 12 mm FWHM. Image matrix size and zoom factors applied during reconstruction determine the final voxel size and may lead to additional image resolution loss when voxel size is larger than half the intended FWHM (Nyquist criteria). The final image resolution has a large effect on so-called partial volume effects (PVE), which results in an underestimation of SUV for small objects. PVE increases with decreasing image resolution. PVE becomes mainly important for lesions smaller than three times the FWHM. To date, no accurate and precise PVE correction are available for objects smaller than about two times the FWHM (i.e., generally <15 to 20 mm diameter). Consequently, image resolution must be matched as closely as possible among centres in multi-centre trials to avoid differences in SUV due to difference in image resolution. Other data acquisition setting and reconstruction parameters, such as the use of time of flight and relaxation parameters, affect image quality and SUV outcome. When images have been collected, additional variability in SUV results from different data analysis strategies, especially region of interest (ROI) definitions. Fixed size ROI providing ‘SUV peak’, maximum pixel values providing ‘SUVmax’ or the average ROI value within a 2D or 3D ROI providing ‘SUV mean’ are being used. Moreover, 2D and 3D ROI can be generated manually or using various semiautomatic procedures, the latter generally based on some percentage threshold value relative to maximum SUV. Examples of the effects of the number of OSEM iterations and different image resolutions using various ROI on SUV are shown in Fig. 1. Finally, various SUV normalisations are being utilized such as SUV normalised by body weight, body surface area or lean body mass.

The protocol for standardisation and quantification of FDG-PET whole body studies is based on principles for minimizing intersubject and inter-institute variability of SUV measures:

The recommendations in this paper aim at standardisation of FDG-PET studies in order to achieve interchangeability of SUV in multi-centre trials. Yet, a few issues are still open for debate and need to be addressed in the near future. Also, during multi-centre phantom measurements, some limitations were encountered and will be discussed below.

First of all, the recommended FDG dose takes into account patient weight, time per bed position, applied bed overlap and scanning mode (2D or 3D). Although patient weight is used directly, the variation in image quality does not depend linearly on patient weight, and the relation between weight, length and gender is ignored. An alternative parameter to vary dose could be measures such as lean body mass, body surface area or body mass index, but their relation with image quality is not resolved as well. Secondly, it is assumed that effects of minimal versus 50% bed overlap on image SNR equals a factor 2. Similarly, it is also assumed that 2D and 3D acquisition mode results in a factor 2 difference of SNR. Ideally, it is optimal to assess image SNR for all scanners using various anthropomorphic phantoms (with various sizes and levels of outside field of view activities) to fully determine differences in image quality amongst various scanners and use those data to specify recommended FDG dose. The recommendations given in this paper, however, offer an advantage over the usually (fixed) dosages applied in most centres, while they are still feasible to be used in a clinical setting. The image quality QC experiments are used to assess any remaining impact of image SNR on SUV quantification. So far, it has been observed that differences between scanners and centres are within 10% provided recommendations are followed strictly (Fig. 2).

Occasionally, the possibility and flexibility to set acquisition and reconstruction parameters are limited. In practice, however, it was observed that present default reconstruction settings, as implemented by the manufacturer, closely fulfilled the recommendations given in this paper. The limited possibilities to change or overrule the manufacturer-provided settings is a way of standardisation by itself and may therefore be considered as ‘beneficial’ for the moment. Yet, the impacts of future software and hardware upgrades on SUV quantification need to be carefully monitored. Any change in default acquisition and reconstruction algorithm and their settings will directly have an effect on observed SUV. Any software and hardware upgrade should therefore be clearly specified by the manufacturer, including a description of the effects of the upgrade on image quality (resolution, SNR). The authors therefore hope that PET and PET-CT scanner manufacturers are willing to clearly specify the meaning of acquisition and reconstruction parameters used in their software and graphical user interface. Moreover, sufficient flexibility in changing parameter settings would be beneficial to facilitate matching of image SNR, convergence of iterative reconstruction methods and image resolution in a multicentre studies.

Another observation (data not shown) is that calibration and image quality, especially uniformity of pixel values of reconstructed images, may differ between (same) scanners of the same manufacturer at different sites. The consequence of this observation is that all QC measures have to be performed at each site to verify that the scanner characteristics and specifications fulfill the criteria given in the ‘NEDPAS’ protocol.

Occasionally, we observed that cross-calibration between PET scanner and dose calibrator revealed a (relative) difference of more than the recommended 5%. The cause of this discrepancy is not fully understood, as the phantom used to calibrate the scanner to the dose calibrator is or can be used to verify correct calibration during the cross-calibration QC measurement. One hypothesis may be that acquisition and reconstruction settings and algorithms used during the calibration procedure are not equal to those during the calibration QC (or verification) measurements and thus not equal to that used during patient studies. Discrepancies may also be caused, e.g., by the use of ‘calculated’ attenuation correction rather than using a CT-AC or a transmission scan. Other causes might be the use of (slightly) different reconstruction or correction algorithms or settings. Moreover, the calibration must be performed at (a range of) count rates normally encountered in clinical studies. It is therefore of utmost importance that the scanner is calibrated using acquisition and image reconstruction algorithms and settings identical to those used during patient studies. In this way, it is assured that results obtained during calibration are applicable to those obtained during the calibration QC and the patient studies.

An issue to resolve is the use of contrast agents during PET-CT studies. Several studies have been performed to assess the effects of intravenous contrast agents on the accuracy and image quality of PET-CT scans [33‐39]. Impact of the use of IV contrast on SUV of lesions (tumours and nodes) ranged from 0% to 25%. Amplitude and frequency of observed increase in SUV depend on contrast agent used and specific imaging protocols [40]. Moreover, effects of using intravenous contrast may differ between baseline and posttreatment scans [37]. A simple strategy to avoid any artefact or quantification error is to acquire a dedicated low-dose CT-AC scan prior to PET scanning. After concluding the entire PET procedure, a diagnostic CT scan including use of intravenous contrasts can then be made. Using the latter strategy, it is also feasible to acquire the diagnostic CT at maximal inspiration to further enhance diagnostic accuracy, while the CT-AC can be used for localisation of FDG avid lesions.

In case of oral contrast agents, this scanning setup cannot be followed to avoid presence of contrast agents in the patient during CT-AC. To date, use of (diluted) oral contrast agents during PET-CT studies is a matter of debate. Various studies have shown that oral contrast agents affect quantification of FDG-PET studies possibly up to 103% in the case of high density barium contrast [41], but effects on SUV seem to be small when using diluted or other types of contrast agents [42‐45]. The amplitude of this effect thus depends strongly on type of contrast agent and a priori unknown local accumulations. Moreover, increase of SUV seemed to be mostly observed in the ascending colon [46]. An interesting alternative might be the use of negative oral contrast agents, which do not affect the accuracy of CT-based attenuation correction, and is therefore a promising possibility for PET-CT studies [47, 48]. Use of negative oral contrast will hopefully be of sufficient quality for diagnostic use, which needs to be assessed by the radiology community.

The potential effects of contrast agents on SUV may vary between scans and can create (extra) artificial SUV responses in longitudinal studies. Although effects of 0% to 25% seem small, it is just one of the many such (small) factors affecting SUV accuracy and variability. The accumulated resulting inaccuracy led us to argue that all (small) factors contributing to variability in SUV across centres should be controlled as much as possible, and this reasoning is, in fact, the main driver of many of the presented recommendations. Based on phantom experiments (Fig. 2), inter-institute variability due to technical issues could be minimized to within 7% (1 SD) provided upon strict standardisation of PET procedures. The clinical benefit of using contrast agents should therefore be carefully balanced against potential increased SUV variability and/or incorrect assessments of treatment response up to 10–20% in some cases [33, 37].