This study confirms that estimation of renal perfusion based on 82Rb PET/CT using AA as the IF in a 1-tissue compartment model is feasible, as previously indicated by Tahari et al. [21]. Additionally, our results support the use of the AA-VOI in a single FOV as an alternative IF to the LVBP; the low intra-assay coefficients of variation are acceptable with excellent inter-observer reliability, thus allowing renal clearance of 82Rb to be determined using a single FOV assessment of the kidneys in their entirety. To our knowledge, this is the first study to assess method precision and determine intra-assay variation and inter-observer variability for clearance estimates with 82Rb PET/CT.
82Rb as renal perfusion tracer
There are many advantages to using 82Rb PET/CT for measurement of renal perfusion: it is non-invasive and does not require blood sampling or urine collection, making the procedure less burdensome for patients; it allows for single kidney perfusion analysis and is readily available from 82Sr/82Rb generators which are already in situ at sites routinely using 82Rb for assessment of myocardial blood flow, thus making it cost effective. In comparison, the “ideal tracer”—15O-water—can be utilised only in centres with on-site cyclotron access [31]. The combination of a short 82Rb half-life of 75 s and short acquisition time allows for repeated scans of the same subject within a short timeframe, presenting unique opportunities to examine acute effects of differing drugs on renal perfusion. For example, 82Rb PET/CT may be especially suitable for use in cross-over studies exploring interventional effects.
No absolute contraindications exist to the use of 82Rb; thus, patients suffering from all stages of AKI and CKD can undergo the examination without risk of deterioration of renal function.
Since renal 82Rb accumulation exceeds myocardial 82Rb accumulation, half the tracer dose of cardiac studies is sufficient to perform good quality renal imaging, resulting in a low effective radiation dose (~1 mSv) for a single scan of the kidneys in their entirety, including the AA for use as IF. Additionally, for modern digital scanners with high sensitivities, even lower tracer doses may be sufficient to perform the examination.
Input functions and necessary data correction
Pharmacokinetic modelling requires an IF, where sampling of peripheral arterial blood to produce an arterial TAC is the gold standard method for obtaining an accurate estimation. However, the short half-life of 82Rb necessitates an alternative to the arterial sampling derived input curve. This can be achieved using image-derived input curves based on, e.g., PET/CT scanning, where LVBP- and AA-TACs are examples of IDIFs. Accurate quantitative IDIF estimation is dependent on many parameters, relating to both the individual PET/CT scanner and reconstruction parameters used for imaging, the geometry, size and placement of analysis VOIs with respect to structural organs-of-interest boundaries, the ratio of neighbouring activity concentrations, as well as requiring calibration of the 82Rb-tracer injector system and imaging scanner with associated dose calibrators, the 3 primary sources of quantitation error being scanner count efficiency, PVE and spill-over. As a minimum, an understanding of what corrections are, and are not, automatically included in an individual scanner-systems software, is necessary to verify correct method implementation and data analysis locally. If one can ensure the relative contributions from PVE and spill-over are negligible in all VOIs used to obtain organ-specific TAC data, the assumption that any global scanner-specific error will cancel out in the kinetic modelling should be adequate and allow for evaluation of activity concentrations without need for cross-calibration of all systems. However, if it is not possible to ensure negligible PVE and spill-over effects in one or more of the VOIs, then, the relative activity concentrations defining the TACs will not be correct with respect to each other and will result in an erred kinetic analysis of renal flow. One method to reduce PVE is to define VOIs as 1-cm3 volume spheres centred on the highest activity voxel in the organ of interest and measure peak-activity concentrations. However, placement of smaller VOIs is variable and observer dependent, especially in highly inhomogeneous (biological) activity distributions. As such, it can be advantageous to use mean values to define the activity measurements. Additionally, if the maximum voxel count lies in proximity to an organ boundary, PVE and spill-over will not be reduced and will still have to be accounted for in the data analysis.
We obtained uncorrected IDIFs from TACs based on VOIs placed in both LVBP and AA in the dynamic PET images. Use of IDIFs based on large-size vascular structures, combined with the high resolution of modern PET scanners, reduces PVE in activity measurement [32, 33]. Additional investigation of measured activity accuracy as a function of distance from structural boundaries (specific to our scanner and reconstruction method), using a phantom containing known 82Rb activity concentrations in geometrical structures simulating the volumes, shapes and sizes of the LVBP and AA (Additional file 1), showed that to ensure negligible PVE and spill-over effects when defining a VOI, its placement needs to be a minimum of 15 mm from organ boundaries, i.e., a minimum of 3–5 voxel distances, dependent on the choice of imaging matrix. In the smaller AA structure, even though there is very little background to give unwanted spill-in, this criterion was not met, indicating that PVE is present and requires correction. On the other hand, the large LVBP volume indicates that PVE is reduced. However, due to significant uptake of 82Rb activity in the left ventricular wall and the use of hot-contouring producing VOIs with, at most, 1–2 voxel distances from the myocardium, the LVBP also required correction for PVE and spill-over. Both corrections were performed based on the method of Katoh et al. [28] using Eqs. 1 and 2, with the necessary correction factors experimentally determined from phantom measurements. Additionally, a global calibration of our scanner’s count efficiency in a large (> 100cm3) homogeneous volume was made, to provide kidney VOI data correction. Here, PVE and spill-over are negligible, but counting efficiency was not automatically corrected by the Siemens scanner software. Based on these arguments, three differing values for “organ-specific” β values were required to ensure the correct relative relationships between the corrected organ-TAC data; one cannot assume that a single, global scanner and reconstruction-dependent correction factor will cancel out in subsequent kinetic analysis, unless the employed VOI definition protocol ensures independence from PVE and spill-over in all organs.
Comparison of our IFs with those of the first (and to date only published) human renal 82Rb PET/CT study by Tahari et al. [21] show both similarities and differences. In both studies, the uncorrected activity in the AA is observed to be lower than LVBP activity. As there is no known metabolism of 82Rb in its passage through the aorta, it is assumed that the activity concentrations in the left ventricle and the aortic lumen are equal, and as such, observed measurement differences will be caused by any scanner and image reconstruction quantification inaccuracies, as discussed above. Tahari et al. [21] assessed the effect to arise from PVE and performed the correction using a simple scaling of their measured AA activity to match the observed maximum LVBP activity. It is unclear whether LVBP activity in [21] was corrected for PVE and spill-over. Our more systematic approach, in which calibrated phantom measurements determined the recovery coefficient β necessary to correct VOI specific activity measurements for a given organ geometry, gave β values of 0.71, 0.612 and 0.643 respectively for LV, AA and kidney TAC corrections, where the numerical value for β does not differentiate between the relative contributions from PVE, spill-over or count efficiency, but provides a “global” factor accounting for all contributions. Application of these organ-specific β values increased the measured peak values for both LVBP and AA TACs (Figs. 6 and 7) resulting in the corrected-AA IFs being scaled to match the corrected LVBP IFs (Fig. 8), supporting the assumption that activity concentrations in the left ventricle and aortic lumen are equal. This agrees with the IFs shown in Tahari’s study. The main difference is that our K1 values obtained from AA IFs differ from their AA K1-derived values, due to our correction of kidney-TACs for system counting efficiency.
Using the β-corrected TAC data, we find for both AA and LVBP, the intra-assay coefficients of variation are acceptably low, indicating that 82Rb PET/CT is a precise method for evaluation of K1, hence allowing for determination of changes in K1. Additionally, the inter-observer variability assessment supports the use of AA as IF as a robust image-derived method for determining renal perfusion, with excellent reliability demonstrated for both kidneys using AA, compared to good to excellent reliability using LVBP.
Renal clearance—measurement of K1 and interpretation of clearance values
High renal 82Rb uptake and accumulation were confirmed. To minimise errors in uptake estimation caused by regional differences, it is important to measure uptake in the entire kidney. In our study, 13 out of 18 completing subjects (72%) showed both LVBP and the entire kidneys in FOV-A, such that 5 analyses were performed on truncated kidneys. In the article by Tahari et al. [21], only 3 out of 8 subjects (38%) had the LVBP and kidneys in the same FOV (corresponding to FOV-A), and 5 out of 8 subjects had the LVBP and kidneys in separate acquisitions. As a global quality control, we found no significant difference between K1 values derived from AA activity curves in FOV-A and those in FOV-B, supporting the assumption that in the studied population with healthy, lesion-free kidneys, quantitation obtained from truncated images of the kidney tissue is representative of values which would be obtained from imaging the kidneys in their entirety. This may not, however, be universally true. Since blood flow differs between the renal cortex and outer and inner medulla, the extent of differential renal tissue included in the analysis VOI may affect values of measured blood flow.
The poor quality of the low-dose CT, used for AC-correction only, did not allow discrimination of the cortex and medulla in our renal VOIs, and even the relatively good quality images and high renal uptake observed in 82Rb PET/CT imaging could not ensure a reliable discrimination between diverse flow regions. This differential flow measurement is potentially possible using CT contrast enhancement or even PET/MR from which to define the kidney VOIs, but was not available for the present studies.
The conversion of K1 values (ml/min/cm3) to total clearance values for both kidneys (ml/min) can be approximated by multiplying the K1 values with the total volumes of the renal VOIs assuming 1 cm3 to be equivalent to 1 g of tissue. These clearance values are summarised in Table 5.
The calculation of total clearance is dependent on the measurement of the K1 values, which are observed to be quite robust and are only to a limited extent dependent on the size of the VOIs used in their determination. However, the delineated volumes of the renal VOIs provide only a rough approximation to the actual anatomic volumes examined. This is due to the less-than-optimal resolution of the PET-scanner, which for this study used a matrix with voxel size 6.3 × 6.3 × 3 mm corresponding to a voxel volume 0.12 cm3 and can lead to errors in determination of the volume analysed: it is likely that our renal VOI volumes are somewhat overestimated. This is based on the observation that the average total volume used for analysis was 296 ± 30 ml, corresponding reasonably to the total anatomical VOI for both kidneys being 300 ml. This corresponds to the reported value for “total parenchymal three-dimensional volume” of 302 ml (range 215–499 ml) as determined by CT-volumetry in a recent study by Gardan et al. [34]. As this volume corresponds to the entire kidney volume, including low flow regions contributing little to the total clearance, then the actual “renal volume” responsible for high flow rates must be smaller. In the same CT-volumetry study, the “cortical renal volume” was measured to be 189 ml (range 126–308). Use of this volume would render total clearance values of 527 ml/min and a BSA normalised total clearance of 489 ml/min, but would probably provide a slightly underestimated value for the total clearance. This illustrates that accurate determination of total clearance values based on PET/CT methodology requires accurate knowledge of the renal volume of relevance. Thus, to obtain a reliable estimation of total clearance using 82Rb, future studies will need to have a better, and preferably standardised, definition of the regions responsible for the flow values in question, which will then allow for reliable comparison to existing reference methods as, at least for ERPF determination, these are independent of this variable.
or this study, our measured total clearance values are observed to be low when compared to previously published mean values for RBF: ~1100–1500 ml/min [35, 36]. However, they are quite similar, if somewhat at the high end, to previously published mean values for ERPF with values 345–700 ml/min/1.73 m2 [35, 37, 38]. This suggests 82Rb PET/CT may actually be estimating ERPF and not RBF as is the current understanding.
Whether we measure estimated RBF or ERPF with 82Rb depends on the distribution of the tracer between plasma and erythrocytes in whole blood. Early studies on potassium permeability of the erythrocyte membrane showed a very slow exchange of radioactive potassium and rubidium between plasma and erythrocytes amounting to 1.8–2.1% per hour and even less over 8 min of study [25, 26]. Hence, most 82Rb is present in the plasma during renal uptake studies, implying that the measured renal uptake values most likely represent estimated RPF after correction for extraction, if EF is close to unity.
Assuming our data represents RPF, estimated RBF can be calculated by correcting with the haematocrit value which is easily measured. For canines, EF is estimated to be 0.89 (0.80–0.95) [19], but to our knowledge, remains to be determined in humans due to difficulty in calibrating and measuring blood activity for 82Rb. However, if we assume the extraction values to be similar for humans, after correction for the haematocrit value (in the present population 42 ± 0.02), we find an average total estimated RBF value normalised to BSA of 1494 ± 221 ml/min/1.73 m2, which lies at the upper end of the expected general range for RBF in healthy subjects. Since our study population consisted of a highly homogeneous group of young, healthy subjects, our estimation of a high ERPF and consequently a high RBF is to be expected; our results being consistent with 2 previous studies of similar population groups, published in 1959 and 1960 respectively [35, 36], where mean RBF values in the range 1100–1500 ml/min, were calculated based on measurement of PAH-clearance and using conversion for extraction fraction and haematocrit. Specifically, our mean and range of estimated RBF values are fully consistent with the published range of individual RBF values (1150–2350 ml/min) in the study by Brodwall et al. [36].
Study strengths and limitations
The major strengths of this study are a combination of the randomised cross-over design, the standardisation of pre-scan conditions (fluid intake, exercise level, duration of fasting period), and the consecutive acquisition of the four 82Rb PET/CT scans over a short 45-min period, enabling optimal evaluation of intra-assay coefficients of variation and hereby precision. Additionally, the use of measured recovery coefficients provides reliable numerical correction of the IFs which are specific to our PET/CT scanner, imaging reconstruction method and choice of VOI definitions.
The homogeneous study population consisted of healthy adults, providing estimated RBF measurements uninfluenced by age and medical therapy. However, this is also a potential bias as it is not certain that results from this study can be directly applied to a population of elderly subjects, nor to subjects suffering from hypertension or renal disease; additional feasibility studies may be needed for these populations. Additionally, before 82Rb PET/CT can be implemented for clinical estimated RBF determination, further evaluation is required of day-to-day variation as well as the quantitative accuracy of the method.
A technical limitation is the use of the automatic injection system used to provide the bolus administration of activity; in practice, a short infusion of 82Rb is administered, which depending on the age of the 82Sr/82Rb generator can have a duration between 20 and 40 s, and as such does not represent a true bolus injection which should ideally be administered within 10 s. For this reason, low activity may be present in the kidneys even before the activity in the blood pool has peaked, as seen in Fig. 8.
Another limitation, when estimating RBF from the measured K1 values, is the assumption that the extraction fraction and blood volume fraction, as derived from animal experiments, are valid for calculation of RBF in humans from measurement of ERPF. Due to a general lack of published literature for 82Rb renal flow measurement, with early research based nearly exclusively on animal studies (1990s and earlier) and the first (and as far as we are aware, only) human investigation performed and published by Tahari et al. in 2014, no solid data is available for humans, such that animal-based values are the best, and only, data available to us.
Also, a significant limitation is that it does not provide comparison to a reference method; the accuracy of 82Rb PET/CT for RBF estimation cannot be evaluated. Based on the available literature during study design, the assumption was that the method provides a measurement of RBF for which an appropriate reference method would be in comparison with 15O-water studies [31]. However, in light of the results presented here, if we are in fact measuring plasma flow and not, as originally assumed RBF, then additional reference methods become available, such as PAH/OIH-clearance methods [35, 37, 38]. In fact, comparison of 82Rb measured K1 flow values, with a reference method for ERPF evaluation, could help answer the question regarding which quantity is actually being measured in 82Rb PET/CT studies. However, as discussed above, the accurate calculation of total clearance/flow rates is highly dependent on the volume analysed, and for future studies, a standardised segmentation of the regions analysed will be needed to provide accurate analysis and comparison with these reference methods.
