Introduction
Prostate cancer (PCa) is one of the most frequently occurring cancers in the male population [
1] and is now the second most common cause of death from malignancy in this group [
2]. Positron emission tomography (PET) is an integral part of PCa management, with prostate-specific membrane antigen (PSMA)-targeted tracers used for primary staging and also in the evaluation of biochemical relapse, assisting in therapy planning and disease management [
3,
4].
PSMA is a cellular transmembrane surface protein that is overexpressed in PCa. Unfortunately, PSMA is not prostate-specific [
5] and also binds to cells in other tissues as well as to the neovasculature in other malignancies [
6]. Several PSMA radiotracers have been produced [
7]. Labelled predominantly with
68Galium (
68Ga) and
18Fluoride (
18F), these radiotracers have been made commercially available and disseminated in clinical trials and daily clinical practice. The use of
68Ga radiolabelled compounds is widely implemented. However, production of
68Ga-radiotracers requires an on-site
68Ga generator, and
68Ga radiotracers have a relatively shorter half-life and longer positron range than
18F labelled tracers, resulting in images with lower spatial resolution.
Conventional static PSMA PET imaging is usually performed 60 min after tracer administration, although modified protocols with scanning at different time points have been suggested to improve image quality [
8,
9]. However, most previous studies have been limited by the single timepoint evaluation of conventional PET, and only few comprehensive studies of PSMA pharmacokinetics have been attempted. These have mostly applied a dynamic protocol on conventional field of view cameras constrained to two bed positions over the pelvis [
10‐
15], with only one truly “whole-body” dynamic acquisition study performed on a total-body PET/CT scanner [
16]. Using dynamic PET with [
68Ga]Ga-PSMA-11 has shown to increase identification rates of both primary PCa and local recurrence [
12,
17], suggesting that dynamic whole-body (D-WB) PSMA PET imaging could outperform static PSMA PET.
For the analysis of PSMA kinetics, a two-tissue compartment model can be applied to describe the tracer exchanges between plasma and tissue. In this case, the first compartment (unbound compartment) represents the free unbound tracer in interstitial fluid, while the second compartment (bound compartment) represents tracer bound to the PSMA receptors, with the transport rate k
3 correlating with tracer binding and internalization, and k
4 with the dissociation of the tracer from the receptor and externalization. For PCa lesions, the k4 ≈ 0, as the binding is predominantly irreversible [
11].
Previously, dynamic PET acquisition was limited to a single field of view and technically difficult to perform due to the necessity of acquiring arterial blood samples to accurately determine the input function. Given the disseminated nature of malignancies such as PCa and the volume of patients referred for pre-therapy PET, dynamic PET has not been implemented in routine oncological work-up. However, this may be about to change.
Recently introduced methodology [
18,
19] allows for D-WB PET acquisitions in conventional PET scanners, by applying the linear Patlak model [
20,
21] to a multi-pass continuous WB dynamic PET acquisition. This imaging protocol provides not only the conventional standardized uptake value (SUV) images, but also multiparametric images based on Patlak kinetic modelling [
22]. These images are the
Ki images (representing the effective tracer binding by the PSMA receptors) and DV images (representing the distribution volume of non-trapped tracer in the reversible compartments and the fractional blood volume). These advances have prompted renewed interest and research in the use of dynamic PET for oncological imaging as reviewed elsewhere [
23,
24]. While the multiparametric D-WB protocol has been successfully applied for 2-[
18F]fluoro-2-deoxy-D-glucose ([
18F]FDG) [
25‐
27], it has until now only to a lesser degree been explored for other tracers.
At our department, we recently transitioned from a 68Ga labelled tracer ([68Ga]Ga-PSMA-11) to an 18F labelled tracer ([18F]PSMA-1007) for routine clinical PSMA PET allowing for a comparison of the two radiotracers. With this study, we therefore aimed to evaluate the tissue pharmacokinetics of these two tracers, the image quality and clinical impact of multiparametric D-WB PSMA PET imaging, and the quantitative accuracy of the resulting parametric values.
Materials and methods
Patient population
This study was a retrospective analysis of data. Participants were recruited from all patients referred for PSMA PET/CT as part of their clinical evaluation if they were deemed fit to lie still for 70 min during scanning. The study was approved by the local ethics committee in the Central Denmark Region (1-10-72-188-19).
D-WB PSMA data were obtained from 20 male patients with known prostate cancer. Ten patients were scanned with [68Ga]Ga-PSMA-11 and ten patients with [18F]PSMA-1007.
Data acquisition and image reconstruction
The study participants were scanned on a Siemens Biograph Vision 600 PET/CT scanner (Siemens Healthineers, Knoxville, USA) with a 26.2-cm axial field of view. A fully automated multiparametric PET/CT acquisition protocol (Multiparametric PET Suite AI, Siemens Healthineers, Knoxville, USA) was used.
[68Ga]Ga-PSMA-11 cohort (N = 10): These subjects were scanned with a 76-min multiparametric PET acquisition protocol, started at the time of an injection of [68Ga]Ga-PSMA-11 (2 MBq/kg). The PET protocol consisted of 1) a 6-min dynamic scan with the bed fixed at the chest region, and 2) a 70-min dynamic WB PET scan consisting of seven continuous 10-min WB passes.
[18F]PSMA-1007 cohort (N = 10): A 70-min multiparametric PET acquisition protocol was started at the time of an injection of [18F]PSMA-1007 (2 MBq/kg). The PET protocol consisted of 1) a 6-min dynamic scan with the bed fixed at the chest region, and 2) a 64-min dynamic WB PET scan consisting of 16 continuous bed motion passes: 7 × 2-min WB passes followed by 9 × 5-min WB passes.
The dynamic image acquisition protocols were therefore not entirely identical. In practice, this meant that 40–70 min direct parametric reconstructions for [
68Ga]Ga-PSMA-11 were performed on 3 10-min images, while for [
18F]PSMA-1007 they were calculated from 6 5-min images. However, since all multiparametric images were based on 30-min D-WB PET data, we expect this variation of the frame length to have a minimal impact on image quality and noise of the multiparametric images (
Ki and DV) [
18].
For both tracers, multiparametric images (Ki and DV) were reconstructed using the data from 40-to-70-min post-injection and the image-derived input function (IDIF). This reconstruction protocol was performed using the direct Patlak reconstruction in the Multiparametric PET Suite AI software from Siemens Healthineers. A standard-of-care static SUV image was reconstructed using data from 60-to-70-min post injection. The PET reconstruction parameters for D-WB: For the 10-min SUV image, we used TrueX + time-of-flight, four iterations, five subsets, 440 × 440 matrix, 2-mm Gaussian filter, and relative scatter correction (reconstruction time 2.5 min). Parametric images of Ki and DV were generated using the direct Patlak reconstruction method with non-negativity constraints using list-mode data from multiple passes (40–70 min), TrueX + time-of-flight, eight iterations, five subsets, 30 nested loops, 440 × 440 matrix, 2-mm Gaussian filter, and relative scatter correction (reconstruction time 13.5 min). For image-based kinetic analyses, we also made a 0–6-min dynamic series of the chest region (12 × 5 s, 6 × 10 s, 8 × 30 s; reconstruction time 5 min), and a 6–70-min dynamic WB series (16 passes, reconstruction time 23 min), using the same reconstruction parameters as the static SUV image. This results in complete 0–70-min dynamic PET data coverage of the chest region.
After image acquisition, the automated multiparametric scan protocol automatically identified the aorta on the low-dose WB CT scan a technology from Siemens Healthineers known as automated learning and parsing of human anatomy (ALPHA) [
28] and placed a cylindric volume of interest (VOI) (1.6 mm
3) on the descending aorta to extract the IDIF from the full dynamic PET series of the chest region. Such an IDIF is robust and can be used to replace an arterial blood input function for precise quantitative Patlak modelling [
29].
Image analysis and VOI delineation
Multiparametric images were visually evaluated by two nuclear medicine physicians using Hermes Gold Client v.2.5.0 (Hermes Medical Solutions AB, Stockholm, Sweden). VOI delineation of the multiparametric images was performed using PMOD® 4.0 (PMOD Technologies Ltd, Zürich, Switzerland). Semiquantitative values of SUVmax and SUVmean were obtained from the conventional PET reconstructions, whereas Ki and DV values were extracted from the multiparametric images.
For each patient, VOIs were analysed from areas of tissue without evidence of pathology. Specifically, we performed delineation of an area of the liver, spleen, parotid gland, lacrimal gland, healthy bone, muscle, benign ganglia (with active PSMA signal) in the pelvis and thorax, and bladder. Areas with pathologically increased uptake of PSMA were identified and delineated using an isocontouring method of 55% of SUV
max in the VOI [
30]. Thus, we outlined the primary tumour in the prostate, as well as lymph node and bone lesions. In patients with an uncountable number of active lesions, for example in disseminated skeletal disease, up to ten individual foci were chosen for delineation. Background regions were delineated in the vicinity of these target lesions, corresponding to an elongated ROI drawn in adjoining tissue in at least three consecutive slices. The individual methodology used to delineate these areas can be found in Additional file
1: Table S1, and an example of a lesion and background delineation can be found in Additional file
1: Fig. S1.
We used target-to-background ratio (TBR) as an objective metric for quantitative assessment of ‘lesion detectability’. Detectable lesions require a TBR > 1, and a higher TBR indicates better lesion detectability.
Comparison of multiparametric and image-based Ki and DV values
The estimates of kinetic parameters obtained through indirect image-based analysis can differ from those obtained by direct reconstruction of parametric images, with the latter exhibiting more favourable bias and noise characteristics, as demonstrated in reference [
31]. The noise and bias in
Ki images are influenced by factors such as the specific implementation of the optimization algorithm, the mathematical formulation of the Patlak model, and the utilization of non-negativity constraints [
32,
33]. We therefore compared the kinetic parametric estimates using the two methods. The image-based parameters were calculated by linear Patlak analysis in PMOD® 4.0, using the general kinetic modelling tool (PKIN) with the lumped constant set to 1 and discarding fits with negative values. The direct reconstructed values were obtained using the Multiparametric PET Suite AI from Siemens Healthineers.
Kinetic analysis
70-min dynamic scan data from the fixed bed at the chest region was analysed using a two-tissue compartment model (2CM) and the 70-min IDIF using PMODs PKIN module. More specifically we analysed VOIs in the liver, spleen, healthy bone (thoracal vertebra), muscle (paravertebral) and any of the previously delineated PSMA avid lesions that were included in this limited scan field-of-view. Parameter estimates for a reversible (
k4 > 0) and irreversible (
k4 = 0) 2CM were obtained and compared with the parameters from the irreversible Patlak model [
20] and the reversible Logan Model [
34]. Akaike information criterion (AIC) [
35] was used to select the CM that best fitted each tissue and for each tracer.
Statistical analysis
Statistical analyses were performed using GraphPad Prism 9.2.0. Statistical tests were used for group comparisons (paired/unpaired) and to assess whether data were normally distributed. Welch’s T test was performed for normal distributed data (liver, spleen, bone, and benign ganglia), while the Mann–Whitney test was performed for non-normal distributed data (prostate lesions, lymph node lesions, bone lesions, parotid gland, lacrimal gland, and muscle).
Pearson’s correlation analysis was performed for the relation between Ki and SUV values. P values of < 0.05 were considered significant. Continuous group data are presented as mean ± SD or median (range) as appropriate. Time-series are presented as mean ± SEM.
Discussion
PSMA multiparametric PET images for both radiotracers were of good visual quality as reflected by the excellent TBR and overall image appearance. In general, lesion TACs were roughly similar between the two radiotracers, whereas organ TACs differed noticeably due to the different modes of excretion. Finally, the parametric values derived from the image-based kinetic analyses compared well with the parametric images using direct reconstruction allowing for simple acquisition of whole body PSMA kinetics.
Multiparametric
Ki imaging assumes an irreversible kinetic model in organs and lesions. Tissues with reversible uptake will have underestimated
Ki depending on the degree of reversibility that may differ between tissues. We found that the multiparametric
Ki values for lesions strongly correlated with
Ki values obtained from full 2CM analysis for both tracers, whereas the multiparametric
Ki values for normal organs, such as liver, spleen, and bone, were quantitative only for [
18F]PSMA-1007. Overall, these results indicate that [
18F]PSMA-1007 is better suited for quantitative multiparametric
Ki imaging than [
68Ga]Ga-PSMA-11 as more organs exhibit irreversible kinetics. The observed difference can perhaps be attributed to the different binding potentials of the two tracers [
36,
37]. Muscle tissue kinetics were best analysed by reversible 2CM for both tracers, probably due to the low muscular PSMA activity [
5]. For [
18F]PSMA-1007,
Ki values were quantitative both in healthy bone and bone lesions, which could allow for quantification of disease progression. This may be of clinical interest since the threshold for pathology on static SUV images has been hard to establish due to the varying “normal” PSMA uptake in bones. Thus, we recommend using [
18F]PSMA-1007 for quantitative multiparametric
Ki imaging in order to obtain images with unbiased quantifications of background organs and tissues, whereas both tracers can be used for lesion detection.
The main advantage of PSMA D-WB multiparametric images over static SUV images is that the former allow for differentiation between free unbound tracer (background) and tracer bound to the PSMA receptors.
Ki images are therefore characterized by improved target-to-background ratio, which in theory could improve lesion identification. However, even though tumour-to-background ratios were in general more favourable on our
Ki images, we identified no additional pathologic lesions in this patient cohort. This result is similar to that of previous larger studies [
11,
14] and attests to the known robustness of conventional SUV imaging.
[
18F]PSMA-1007 is often preferred by clinical departments due to its supposedly superior properties evaluating pathology in the pelvis, and the absence of reliance on an onsite gallium-68 generator. However, in our hands, both PSMA tracers identified primary prostatic disease with ease, regardless of whether images were static SUV or parametric
Ki images. However, the improvement in regional TBR contrast associated with [
18F]PSMA-1007 is likely to be more clinically relevant in relapse evaluation studies, of which we only had three in our cohort. Furthermore, it is recommended to administer furosemide shortly before or after administration of [
68Ga]Ga-PSMA-11 [
38], thus diminishing the high residual activity in the bladder [
39]. Our protocol lacks such intervention, as it would reduce patient compliance with the prolonged scan time. Finally, this disadvantage associated with [
68Ga]Ga-PSMA-11 scanning can also be circumvented using D-WB imaging which provides better TBR (in this case PCa to bladder) of the
early dynamic images [
12,
15,
40]. Whether this potentially translates to improved detection of recurrent disease by D-WB [
68Ga]Ga-PSMA-11 remains to be clarified in a larger study.
Both radiotracers readily identified lymph node metastases, although with higher
Ki values in the [
68Ga]Ga-PSMA-11 images. Likewise, skeletal lesion [
68Ga]Ga-PSMA-11 activity was greater in static SUV images and calculated
Ki values were higher in the parametric images. Coupled with the generally higher background signal in the bone observed in [
18F]PSMA-1007 PET images, these findings seem to suggest that [
68Ga]Ga-PSMA-11 PET should outperform [
18F]PSMA-1007 PET in both lymph node and bone lesion detection. However, previous head-to-head studies based on static images have shown more bone lesions detected using static [
18F]PSMA-1007 PET than [
68Ga]Ga-PSMA-11. This is now a known disadvantage of [
18F]PSMA-1007, as up to half of these additional bone ‘lesions’ have turned out to be benign [
41,
42], and consequently, no difference in radiotracer sensitivity to detect malignant skeletal lesions have been reported [
43,
44]. Although histological verification was not available in all lesions in the current study, it is evident to us that some of the delineated bone lesions with [
18F]PSMA-1007, particularly those in the ribs and with low SUV values, likely also represent unspecific benign lesions [
41,
42]. The presence of such unspecific bone lesions can contribute to the difference in distribution of SUV signal observed in Fig.
5, as the [
68Ga]Ga-PSMA-11 cohort included a larger amount of likely bone metastases. The bone lesions were also visible on parametric PSMA PET using both radiotracers, which is unsurprising even though these lesions are probably visualized due to a non-PSMA-related uptake mechanism [
45].
Consequently, parametric images cannot be used to differentiate these unspecific bone lesions from malignant disease, regardless of radiotracer used or scan protocol employed.
Whereas sensitivity to detect lesions was not improved by the parametric imaging, specificity appears to be slightly better. In our cohort of 20 patients, we observed two cases of ‘false-positive’ findings in soft-tissue lesions on the SUV images that showed no tracer uptake on the parametric reconstructions, as previously reported for [
18F]FDG [
25]. However, it is relevant to note that the presence of isolated soft-tissue or lymph node metastasis in the upper extremities is highly unlikely in prostate cancer.
Some limitations to the study must be acknowledged. First, the patient cohort is rather small, and for ethical reasons, patients were not subjected to repeat studies using the two different radiotracers. A direct comparison of findings in each study was therefore not possible. Studying the same patients with both tracers would have been optimal to minimize inter-individual variation in tumour biology, dissemination patterns and length of disease. Second, only a small fraction of patients in our cohort were scanned for disease relapse evaluation, which in theory should be the most promising referral indication, since putative lesions are located in the pelvic area with high background on [
68Ga]Ga-PSMA-11 PET. However, suspected disease relapse only represents a fraction of PSMA PET referrals at our department. Third, we lack histological confirmation of our findings. However high correlation between imaging and histopathologic findings has been previously demonstrated for PSMA tracers [
46,
47]. Finally, although all multiparametric images were based on 30-min D-WB PET data, the dynamic image acquisition protocols were not entirely identical. In a more elegant study setup, we would have preferred identical D-WB PET acquisition protocols.
In conclusion, it is possible to perform D-WB PSMA PET scans that generate lesion and tissue Ki values in a clinical setting by using a multiparametric acquisition protocol on standard FOV PET scanners. Both [68Ga]Ga-PSMA-11 and [18F]PSMA-1007 can be used for lesion detection, with parametric PSMA Ki images showing superior lesion TBR. However, in our small cohort, Ki images did not uncover any additional lesions. For quantitative whole-body Ki imaging, [18F]PSMA-1007 is the preferred choice due to its predominantly irreversible kinetics in organs and lesions, leading to unbiased quantitative values.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.