Standards for PET Image Acquisition and Quantitative Data Analysis

Patient Preparation Procedures

The procedures used for patient preparation affect ¹⁸F-FDG uptake in both tumors and surrounding healthy tissues. Patient preparation procedures therefore should be aimed at maximizing uptake in tumors and minimizing uptake in healthy tissues, thereby optimizing image quality and reducing SUV variability among subjects. The various studies described earlier all provided guidelines for patient preparation, and there seemed to be a general consensus on the optimal preparation procedure. In general, guidelines were given for a fasting period to achieve euglycemic conditions, hydration, use of sedatives and waiting conditions, bladder voiding or use of diuretics, and limits for blood glucose levels. In most cases, additional guidelines were provided for diabetes mellitus patients. A complete delineation of patient preparation guidelines can be found in the articles listed in Table 2.

Until recently, there was still some debate on the optimal time interval between ¹⁸F-FDG administration and the start of a PET study. Lowe et al. (25) and Shankar et al. (35) reported that ¹⁸F-FDG uptake was still rising up to 120 min after injection, although uptake curves seemed to become flatter at 60–90 min after injection. At present, an interval of 60 min with a tolerance of 5–10 min seems to be considered acceptable in most guidelines. In other publications (17,32,34), a minimum uptake period of 30–40 min was recommended. The shift toward a longer uptake period may reflect the trend toward using PET in a quantitative rather than a qualitative manner.

Additional recommendations are still needed for the time interval between (the end of) therapy cycles and the execution of a PET study. For chemotherapy, a minimum interval of 14 d is usually applied, but more detailed recommendations are given by Juweid et al. (7). For radiotherapy, intervals between the end of treatment and the start of a PET study of even 3 mo may be required. The optimal interval may therefore be study specific, and further investigations are required (11). The appropriate timing of PET studies is one of the topics addressed further in other contributions in this supplement issue of The Journal of Nuclear Medicine.

It is also necessary to measure weight and, depending on the SUV normalization procedure, the height of the patient at the time of each PET study. Moreover, the net administered dose specified at the dose calibration time or injection time, which can vary, must be known with certainty. Because all of these values are entered into the SUV equation, they should be reported on a scan report form or entered into the PET system during acquisition, so that the data are stored within the DICOM (Digital Imaging and Communications in Medicine) file headers of the PET scan (43).

¹⁸F-FDG Administration Procedures

The net amount of the administered dose is directly used in the SUV calculation. Consequently, the exact ¹⁸F-FDG dose given to a patient must be known; that is, the dose must be corrected for residual activity in the syringe or administration system. Moreover, decay corrections must be applied to compensate for the radioactive decay of ¹⁸F between the dose calibration time or injection time and the beginning of acquisition. Therefore, clocks in the PET or PET/CT system must be synchronized with those in the dose calibrators used to measure or determine the dose of ¹⁸F-FDG injected. A detailed discussion of these issues can be found elsewhere (9,43).

PET Study Acquisition, Image Quality, and SNR

PET acquisition parameters, such as acquisition mode, scan duration per bed position, and amount of bed overlap in subsequent bed positions, in combination with patient weight and ¹⁸F-FDG dose, affect PET image quality. It has been shown that poorer image quality (increased noise levels) may result in an upward bias of SUV measurements (26,51). To optimize image quality, recommendations are generally given for uptake period, scan duration, and ¹⁸F-FDG dose. The dose can be selected from a range of generally used doses (17,32,35); as a function of patient weight (34); or as a function of the combination of patient weight and scanner type, scanning procedure, and scan duration (43). The last approach attempts to minimize variability in image quality between different types of scanners.

For PET/CT, it is recommended that a patient be positioned with the arms above the head to improve CT-AC (31,32,43). Breath holding at midinspiration or shallow breathing is recommended to minimize artifacts attributable to mismatches in registration (and blurring) between CT-AC and PET emission scans.

Both Delbeke et al. (32) and Boellaard et al. (43) discussed the use of contrast agents during CT-AC. Delbeke et al. (32) and Coleman et al. (31) indicated that intravenous contrast agents may cause attenuation correction artifacts in PET images but that these effects are usually modest with modern PET/CT scanners. The standardization protocol described by Boellaard et al. (43), however, advises that no contrast agent be used until it has been established that attenuation correction artifacts are completely absent when oral or intravenous contrast agents are used during CT-AC. This more conservative approach has been followed because these recommendations focused mainly on the quantification of PET studies. Whether contrast agents can or cannot be used during CT-AC is, however, a topic for further discussion and evaluation (31,43).

Image Reconstruction and Image Resolution

In most articles that make recommendations, guidelines for optimal reconstruction settings are limited because image reconstruction algorithms and specifications are highly specific to manufacturer and scanner type, making generalizations difficult.

Schelbert et al. (34) indicated the need for reconstruction of PET studies with correction for decay during subsequent bed positions and both with and without attenuation correction. Obviously, attenuation correction is needed for quantification. Young et al. (17) recommended the use of attenuation correction but otherwise provided no image reconstruction parameters. However, at that time, data on the effects of image reconstruction algorithms and settings on PET quantification were sparse, and filtered backprojection was still the most commonly used reconstruction method.

Later, Shankar et al. (35) recommended the use of the same scanner and reconstruction algorithms and settings for multiple scans of a given patient. This approach is justified when ¹⁸F-FDG PET studies are used for tumor response measurements based on relative changes in SUVs. It has been shown (45,49) that relative changes in SUVs are almost independent of applied methodology, provided that it is consistently used for all longitudinal PET studies of a given patient and that the scanner is continuously properly calibrated against the dose calibrator.

Delbeke et al. (32) indicated that PET image reconstruction should include corrections for detector efficiency (normalization), system dead time, random coincidences, scatter, attenuation, and sampling nonuniformity.

Similar recommendations were provided by Boellaard et al. (43) to ensure that ¹⁸F-FDG PET images are fully quantitative. Apart from the application of all of the usual corrections for quantification, however, differences in PET image resolution are probably major factors contributing to variability in SUVs among centers (45). Therefore, in multicenter studies, when absolute SUV results are used, it is crucial for image resolution to be matched as much as possible across centers and scanners. This is especially true because methods for overcoming quantitative inaccuracies attributable to (differences in) resolution, so-called partial-volume effects (29), are not widely available yet and, more importantly, are still not sufficiently accurate and precise for structures smaller than 3 times the “clinical” scanner resolution (the clinical resolution is usually ∼7 mm full width at half maximum [FWHM]). To some extent, partial-volume effects may be minimized by use of the SUV_max within a lesion. However, differences in image resolution still contribute to interinstitutional differences in the SUV_max (Fig. 1) (26,43,45). Furthermore, PET image resolution is not equal to and is usually worse than the scanner resolution measured according to NEMA NU 2 specifications, a fact that is often overlooked or not understood. Boellaard et al. (43) therefore provided reconstruction settings for various scanners but explained that reconstruction should be performed so that activity concentration recovery coefficients as a function of sphere size meet the multicenter QC specifications given in the same study. To this end, dedicated QC phantom experiments were proposed. With this approach, resolution and (iterative reconstruction) convergence matching across various scanners and institutes can be achieved, regardless of the fact that the scanners may have different reconstruction algorithms and software.

Data Analysis Procedures and SUV Normalization

After data acquisition and image reconstruction, the extraction of quantitative measures starts with segmentation of the tumor, that is, definition of an ROI. To this end, various ROI approaches have been used; these approaches include manually defined tumor boundaries, semiautomated 2D and 3D region-growing techniques that involve the application of a fixed or relative (to maximum uptake) threshold, fixed-size regions (SUV_peak), and maximum uptake over the entire tumor (SUV_max). Variations in SUVs are also caused by imprecise performance of these “simple” ROI methods, especially for small lesions (smaller than 3 times the FWHM), as can be deduced from the estimated lesion sizes in Figure 1. Therefore, there is an urgent need to develop objective and highly automated volumetric delineation methods that accurately yield true lesion size, such as those described by Geets et al. (52) and Hatt et al. (53). These and other sophisticated tumor segmentation methods are presently being developed but are still being evaluated. However, it is important to remember that different ROI methods will result in different quantitative outcomes (26).

Young et al. (17) suggested that the whole tumor uptake be reported and that the same ROI volumes be used on subsequent scans. Mean uptake and maximum uptake, expressed in MBq/L, should be recorded. Shankar et al. (35) did not recommend a specific ROI methodology but indicated that consistent ROI methodology should be used for a given patient during a longitudinal study. Both articles specifically addressed the use of ¹⁸F-FDG PET for assessing the tumor response to therapy, for which consistency has been shown to be of utmost importance. Boellaard et al. (43) recommended the use of various 3D ROIs (i.e., volume of interest [VOI]) based on relative (to maximum uptake) thresholds, either with or without background corrections (adaptive thresholds). Because a larger VOI may be more precise, it was suggested that the largest VOI be used to provide stable and accurate VOIs corresponding to the tumor location and extent on all subsequent scans. However, this strategy is only applicable to the assessment of a response to therapy. Nevertheless, it was specified that in all cases, the maximum uptake should be reported and the same VOI method should be used for a given patient and across institutes participating in a multicenter trial.

Finally, SUV normalization variables and the serum glucose level affect SUVs. The SUV is usually normalized to body weight, but lean body mass and body surface area are being used as well. Young et al. (17), Weber (9), Shankar et al. (35), Lammertsma et al. (18), and Boellaard et al. (43) all indicated that SUV normalized to body surface area might be more appropriate during longitudinal studies in case of weight loss during therapy, as has been demonstrated in other studies (22,23). The most appropriate method for SUV normalization is still a matter of debate, but the method should be standardized for multicenter trials.

A higher blood glucose level results in a lower SUV. The possible effects of blood glucose levels were investigated by Eary and Mankoff (54) and discussed by Young et al. (17), Weber (9), Delbeke et al. (32), Shankar et al. (35), Lammertsma et al. (18), and Boellaard et al. (43). They stated that the blood glucose level should be checked before the administration of ¹⁸F-FDG. In general, 2 strategies for minimizing the effects of blood glucose levels were recommended. First, with a strict patient preparation protocol (including, e.g., at least 4 h of fasting), a blood glucose level within the reference range (70–130 mg/dL, corresponding to approximately 4–7 mmol/L) usually can be achieved. When needed, the blood glucose level may be reduced by the administration of insulin (17,32), but rescheduling is generally recommended for patients in hyperglycemic states (32,35,43). The blood glucose level threshold for rescheduling ranges from 130 to 200 mg/dL, corresponding to 7.2–11.1 mmol/L. Delbeke et al. (32) and Shankar et al. (35) did not specify an exact threshold but indicated that, in general, a threshold of 150–200 mg/dL (8.3–11.1 mmol/L) is applied by most institutes.

The second strategy for overcoming variability in SUVs because of variations in blood glucose levels involves incorporation of the blood glucose level as a correction factor in the SUV calculation, as suggested by, for example, Lammertsma et al. (18) and Boellaard et al. (43). However, the use of blood glucose level correction is still a matter of debate, and a possible improvement in SUV accuracy should not be counterbalanced by a disproportionate increase in SUV variability. At the very least, a standardized and validated method for measuring the blood glucose level is needed (i.e., bedside devices should not be used (43)), and the method should be calibrated among centers.

QC Measures and Qualification of Personnel

Technical prerequisites for performing quantitative PET studies are that a scanner is operating according to specifications and that it is calibrated correctly with a phantom that has known activity concentrations, as described elsewhere (9,32,35,43). In addition to QC measures, Delbeke et al. (32) provided additional guidelines for minimum PET scanner performance based on NEMA NU 2 specifications and provided guidelines for the qualification of personnel.

Correct functioning of a PET or PET/CT scanner is usually verified on a daily basis with a set of manufacturer-supplied daily QC routines. In addition, a PET scanner should be cross calibrated against a dose calibrator; that is, the activity measured by the PET camera should be directly compared with the injected activity measured by the dose calibrator being used clinically (12,43,51). Boellaard et al. (43) indicated that this relative calibration should be accurate to within ±5% on the basis of multicenter calibration experiences (43–45).

These QC measures for PET are usually sufficient when quantitative ¹⁸F-FDG PET studies are performed to assess the response to therapy on the basis of relative changes in SUVs. However, absolute SUV results can be used for differentiating between benign and malignant lesions, determination of prognosis, and response monitoring by evaluation of residual uptake after therapy or a combination of relative changes and residual uptake during or after therapy; therefore, absolute SUV results are being used more frequently. In all of these cases, especially multicenter studies, additional QC measures are required. Takahashi et al. (55) showed that a variation in the SUV of up to 47% was observed when 5 different scanners were compared and that this variation was reduced to within 22.6% when standardized protocols were used. Boellaard et al. (43) observed that by (very) strict standardization of PET acquisition, image reconstruction, and data analysis procedures, variability in the SUV as a function of sphere size could be reduced to approximately ±10%. Therefore, it was suggested that an assessment of activity concentration (or SUV) recovery coefficients as a function of sphere size be included in multicenter QC procedures. Specification of activity concentration or SUV recovery coefficients for each sphere (within the NEMA NU 2 2001 image quality phantom) is an attempt to achieve resolution and reconstruction convergence matching across scanners and institutes.

Is There a Need for Different Levels of Standardization?

The required level of standardization may depend on the intended use of ¹⁸F-FDG PET. When oncologic ¹⁸F-FDG PET studies are interpreted visually for staging and diagnostic purposes, the PET procedure should focus on optimizing image quality for tumor detection. In such a study or clinical practice, specifications for patient preparation, ¹⁸F-FDG dosage, and scan duration are required, but it is likely that standardization of reconstruction settings across various institutes will be less stringent.

More strict guidelines in a multicenter setting are needed for quantitative PET studies. A distinction can be made between the use of relative changes in quantitative outcomes for measuring the response to therapy and the use of absolute quantitative measures for diagnosis, determination of prognosis, or prediction of responses (e.g., residual ¹⁸F-FDG uptake during or after therapy). With relative changes in SUVs, that is, the ratio of the SUV in response studies to the SUV obtained during a baseline study, the impact of most factors affecting individual SUVs in response studies is minimized (i.e., canceled out). Indeed, it has been shown that the consistent application of a certain methodology, in addition to all other actions undertaken to optimize image quality, is likely to be sufficient (45,49). However, constant scanner performance and calibration over time must be verified (32,35,44). Moreover, if changes in metabolic volume occur, differences in ROI methodologies and image resolution will still affect the observed responses (45), so that standardization of ¹⁸F-FDG PET protocols is still mandatory.

When absolute quantitative measures are used in a multicenter trial, all participating centers must follow strict guidelines for patient preparation, PET acquisition, ¹⁸F-FDG dosage, image reconstruction (i.e., matching of resolution and iterative reconstruction convergence), data analysis procedures, and QC measures (43).

Finally, Mankoff et al. (56) and Lammertsma et al. (18) emphasized the need for the verification of responses based on SUVs against responses obtained from a full kinetic analysis. For this purpose, dynamic quantitative ¹⁸F-FDG PET studies of a small series of patients should be done for comparison with established associations of SUV and kinetic analyses. Several centers, including my own, have large datasets for such comparisons, increasing statistical power. The execution and data analysis of these studies are much more complicated than those needed for SUV quantification. Therefore, these studies are likely to be performed at a limited number of sites. However, at least one single-center validation study should be included during phase I or II studies to correctly interpret SUV responses in larger clinical trials (17). To date, there are no guidelines for performing such quantitative dynamic ¹⁸F-FDG PET studies, although some considerations have been described (17,18). It is obvious that such guidelines should incorporate the same recommendations for patient preparation, ¹⁸F-FDG dosage, administration procedures, and image reconstruction settings as those used for studies based on SUVs alone (Table 2). Guidelines for data analysis have not yet been officially provided, although Young et al. (17) suggested Patlak analysis as the preferred kinetic method for analyzing dynamic ¹⁸F-FDG PET studies.

In summary, different levels of standardization could be based on the intended use of ¹⁸F-FDG PET, ranging from visual inspection to full kinetic analysis. Depending on the capabilities of a certain site, different levels of accreditation could be assigned, thereby allowing or not allowing a site to participate in certain clinical multicenter trials. However, PET centers should be encouraged to meet the highest feasible quality standards. It is important to remember that many small factors affect SUV results and that it is therefore of utmost importance to standardize all of these factors; that is, many small factors can add up to considerable variations in SUVs—up to 50% or more (9,26,45,49,55)—and guidelines therefore should be followed strictly.

Issues in Maintaining and Updating Future Standards

Once a standardized protocol has been established and implemented, regular verification of correct implementation of ¹⁸F-FDG PET guidelines is needed. To this end, multicenter QC measurements should be obtained regularly and centrally reviewed and archived.

Because different software platforms or programs provide different SUV outcomes, even when identical datasets are used (57), standardization of ROI methodology and data analysis software is urgently needed. At present, some initiatives, such as that of the Quantitative Imaging Biomarkers Alliance (QIBA), are being undertaken to address these issues. Alternatively, in the absence of standardized ROI software and implementation, quantification of ¹⁸F-FDG PET studies in a multicenter setting may need to be reviewed and analyzed centrally.

A potential threat for maintaining standard procedures across various institutes and PET/CT scanners over time is the limited flexibility in defining acquisition and reconstruction settings for most modern PET/CT scanners. This limited flexibility is some form of standardization and may, at present, be considered beneficial. Nevertheless, the impact of future software and hardware upgrades on SUV quantification needs to be monitored carefully. Any change in acquisition and reconstruction settings will have a direct effect on derived SUVs. Any software or hardware upgrade therefore should be clearly specified by the manufacturer, including a description of the effects of the upgrade on image quality, that is, on resolution recovery and SNR (55). Moreover, manufacturers should allow for the continuous use of previously implemented acquisition and reconstruction methods and settings to ensure the consistent use of methodology. This matter is of utmost importance when clinical decisions are being made on the basis of observed changes in SUVs. In other words, an artificial change in an SUV because of a change in methodology could incorrectly be interpreted as tumor response or progression; obviously, such a situation should be avoided. An alternative strategy is for each manufacturer to assist in the implementation and maintenance of acquisition and image reconstruction protocols to meet the specifications of a multicenter standardized protocol.

Furthermore, future PET/CT scanners are likely to have higher sensitivity, improved image quality, and higher spatial resolution. Higher sensitivity and improved image quality resulting from more advanced image reconstruction methods may be used to reduce the ¹⁸F-FDG dose, shorten the scan duration, or both. Changes in spatial resolution, however, have a greater impact on SUV quantification (26,27,45,49). Therefore, as mentioned earlier, it is of utmost importance that any PET study can be reconstructed in such a way that it meets the SUV recovery coefficient requirements of current ¹⁸F-FDG PET standardized protocols. A simple solution could be the generation of a second (standardized) PET image by another reconstruction of the PET study with earlier software versions or settings or the simple downsmoothing of the higher (spatial)-resolution PET image with an appropriate filter to meet the (lower) resolution requirements of ongoing multicenter studies. An additional benefit of this strategy would be the collection of paired (high- and low-resolution) quantitative PET datasets. The use of such paired datasets would allow a proper evaluation of the clinical benefits of new technology and might facilitate the design of new ¹⁸F-FDG PET standards, that is, the “translation” of existing protocols, SUV thresholds, or response criteria to newer ones.

Finally, apart from improvements in PET and PET/CT technology, other multimodality imaging devices, such as PET/MRI scanners, will be introduced (31,58–62). Both their clinical use and optimal data acquisition procedures will need to be assessed. Future recommendations will therefore need to focus on PET combined with CT or MRI. Other new developments may be the increasing use of tracers other than ¹⁸F-FDG or of multitracer studies (63–65) to further enhance clinical accuracy or in other applications, such as radiation oncology (66,67). These issues are beyond the scope of the present review, but multitracer and multimodality studies are likely to play an increasingly important role in oncology in the future (62,65).