Background
Yttrium-90 (half-life, 64 h) is a radionuclide used in targeted radionuclide therapy, particularly for radioimmunotherapy (RAIT), peptide receptor radionuclide therapy (PRRT), and selective internal radiotherapy (SIRT). Promising clinical results have been obtained for the treatment of B-cell lymphomas with anti-CD20 and anti-CD22 monoclonal antibodies [
1,
2], while the treatment of neuroendocrine tumors with somatostatin analogues has proven its benefit in the last decade [
3,
4]. One RAIT product labeled with
90Y has been already approved by the regulatory authorities (Zevalin®, Bayer Corporation, Pittsburgh, USA). Encasing
90Y in a resin or a glass sphere has also provided a promising approach for the treatment of hepatocellular carcinoma (HCC) and unresectable liver metastases with SIRT [
5‐
7]. Two microsphere products are approved and available for clinical practice: TheraSphere® (glass microspheres; MDS Nordion, Ontario, Canada) and SIR-Spheres® (resin microspheres; Sirtex Medical, North Sydney, Australia).
The emission spectrum of
90Y is almost exclusively β
− (mean energy 933.6 keV), making the irradiation of macroscopic tumoral lesions possible. However, despite the increasing therapeutic applications of
90Y, imaging the distribution of this isotope in the patient's body remains a significant challenge when assessing the quality of tumor targeting after injection, as well as when performing systematic dosimetry studies. While pre-therapeutic imaging with surrogate tracers remains the best and only means of adapting the therapeutic activity to be injected (such as
111In or
68 Ga for RAIT or
99mTc for SIRT), it could pose significant organizational issues depending on local constraints (especially for
111In pre-therapeutic studies) as well as additional costs. Moreover, the biodistribution of a radiopharmaceutical labeled with a different radionuclide may imperfectly predict that of
90Y-labeled radiopharmaceuticals [
8]. Therefore,
in vivo imaging during the treatment phase can predict toxicity and efficacy, and can help optimize the injected activity in subsequent courses, given that radiotherapy treatments usually require repeat injections.
The current clinical practice for
in vivo imaging of such treatment involves a SPECT acquisition using bremsstrahlung and remains the method of choice to assess the distribution of
90Y-labeled radiopharmaceuticals [
9]. Although reconstruction algorithms suitable for bremsstrahlung imaging of
90Y are under development [
9,
10], the resulting images have the disadvantage of having a low spatial resolution and poor quantification performance, especially for small lesions when using parallel-hole collimators. However, a recent study showed the benefits of a using pinhole collimator for bremsstrahlung imaging of
90Y with very promising results [
11].
The prediction and discovery of the 0
+-0
+ transition of
90Zr [
12], which results in a β
+/β
− pair creation with a very low branching ratio of 31.86 × 10
−6
[
13], provide the opportunity to detect
90Y distribution using PET images [
14]. The first clinical applications of PET imaging with
90Y have been tested recently in the context of radioembolization [
15,
16], PRRT [
17], and RAIT for lymphoma [
18]. These studies focused primarily on the detection of lesions incorporating a very large amount of tracer (typically several MBq mL
−1, equivalent to several tens of Bq mL
−1 when considering the pair creation), thus facilitating the visualization and accurate measurement of radioactive concentration. Similarly, the value of using time of flight (TOF) information for this type of acquisition in this context of high radioactive concentration was recently investigated [
19]. In a RAIT protocol, the expected concentration in the lesions generally ranges from 0.01% or less to 1% of the total injected activity. This leads to a typical concentration of several tens of kBq mL
−1
[
20].The aim of this work was three-fold. The first objective was to determine the minimum detectable activity (MDA) with PET imaging using
90Y for a LSO-based acquisition system in relation to lesion size. The second objective was to study the impact of TOF reconstruction on detectability and quantitative accuracy according to lesion size and finally to correlate these results with analysis on patient data.
Discussion
This work addresses one aspect of the general question of PET imaging with very low activity, such as that encountered typically in PET monitoring of therapeutic ion irradiation [
32]. Here, we showed that although
90Y imaging is possible, it is useful only for a size and concentration above a given level of radioactive concentration and when optimized parameters are implemented.
Previous work in the area of
90Y imaging by PET has established this technique as being potentially useful for the precise assessment of the spatial distribution of microspheres and to determine the quantitative input activity map for subsequent dosimetric calculations [
15,
17]. The advantage of using TOF vs. non-TOF reconstruction has also been recently studied in a comprehensive manner [
19]. However, all these studies were conducted in the context of therapeutic radioembolization with microspheres [
15,
19] or PRRT [
17]. This type of treatment results in a radioactive concentration of several MBq mL
−1 or even tens of MBq mL
−1
[
17‐
19]. To our knowledge, no studies have reported the calculation of the minimum detectable radioactive concentration of
90Y by PET imaging, or its relationship to lesion size and its impact on the quantification of the signal. The current work was conducted in the context of a systemic injection when the radioactive concentrations are much smaller than SIRT, the ratio of lesion to background is, in theory, high and the signal from the natural radioactivity of
176Lu is of the same order of magnitude as that obtained for low concentrations of
90Y.
The experimental data allowed us to determine some practical aspects regarding the MDA or the influence of reconstruction parameters on detectability and quantification accuracy. Mainly, for lesions with an activity concentration exceeding 2 to 3 MBq mL
−1, the best detectability was provided by reconstruction using TOF or non-TOF information with one iteration whatever the lesion size. The TOF information and one iteration were required for an activity concentration below 2 to 3 MBq mL
−1 regardless of the lesion size, in terms of best detectability. These results are consistent with those reported previously in a context where the signal to background ratio was one order of magnitude lower [
19] than in our work. Table
2 summarizes the estimated MDA for all lesion sizes, showing that the best detectability was effectively reached with one iteration (TOF or non-TOF). Regarding the accuracy of quantitative information, the best compromise was reached using TOF information and three iterations regardless of the radioactive concentration and lesion size.
Table 2
Minimum detectable activity
28 | <500 | <500 |
22 | <500 | <500 |
17 | ≈600 | ≈1,000 |
13 | ≈900 | ≈1,400 |
10 | ≈1,000 | ≈3,000 |
These experimental findings were compared to those derived in clinical situation and allowed a comprehensive knowledge of the signal collected in patients.
We must emphasize that the signal from the
176Lu became a significant contribution for random coincidences for radioactive concentration below 1 MBq mL
−1 in a cold background (or with a high ratio between the signal in the background and the spheres). After reconstruction, the residual signal resulted in multiple isolated foci of moderate intensity in the reconstructed volume that may increase the number of false positive results when small lesions or low concentrations are involved. This reconstructed signal is further amplified by the necessary attenuation correction and appears more intense in the reconstructed volume, especially when three iterations are used. In this last case, the radioactive concentration of a false positive could be in the range of few MBq mL
−1 due to statistical fluctuations. Generally, when the concentration measured in a lesion extracted from a volume reconstructed with one iteration and TOF is greater than 1 MBq mL
−1, the lesion could be considered as true positive even if the concentration measured in a volume reconstructed with TOF and three iterations is in the range of few MBq mL
−1. Conversely, a concentration measured as high as few MBq mL
−1 for a reconstructed volume with TOF and three iterations but with an intensity of less than few hundred of kBq mL
−1, when the volume is reconstructed with one iteration and TOF, could be considered as false positive. This effect was clearly observed in the phantom measurements. This is a potential method to discriminate true positive from false positive in the context of clinical
90Y-PET imaging, but it still needs to be confirmed on larger clinical datasets by comparing it with
99mTc-MAA and contrast-enhanced CT. Finally, unlike the hypothesis provided by Campbell et al. [
33] and van Elmbt et al. [
19], we found a mean tumor-to-liver uptake ratio of roughly 10 for patients with HCC. This led to higher mean activity concentrations of 10 MBq mL
−1 in the tumor and approximately 900 kBq mL
−1 (range 200 to 1,300 kBq mL
−1) in the liver, higher than those reported in the latter studies.
One of the main limitations of our study was the estimated CNR based on the assumption of uncorrelated noise and was generally assessed for data reconstructed with linear algorithms such as FBP. The use of 3D OP-OSEM and a PSF model within the chosen reconstruction method led to a correlation between voxels, which may have affected the noise estimation. However, based on the chosen parameters, we assumed that this effect would be limited to the extent of 2 to 3 voxels only [
34], despite the post-smoothing Gaussian function (2 mm FWHM) and could be neglected in this work. Moreover, the definition of the limit of detectability of lesions (i.e., CNR > 8) is a complex function depending on the size and shape of the lesion. The value chosen in this work was adapted from a study with human observers within a framework of simulated noisy micrographs [
29]. The limit used in this study does not necessarily reflect that which is specific to PET imaging in a low count rate with an iterative reconstruction. Assumptions related to the Rose criterion [
27] were probably not completely fulfilled, but the quantitative results that we derived were consistent with the visual analysis. Moreover, an additional analysis (not shown) with a different limit of CNR > 5 did not substantially modify the main conclusions regarding detectability.
The CNR may also be significantly improved using a Gaussian post-filtering with a higher FWHM value to reduce noise in the reconstructed images [
26] but at the expense of a more pronounce correlation between voxels. In this work, we have set the post-filtering strength to that used for routine examinations of
18 F-FDG considering, firstly, the improvement in the reconstructed spatial resolution provided by the PSF modelling, and secondly to reduce the bias in quantitation for the reconstructed volume.
Regarding the estimation of scattered radiation, it should be noted that it relies on a prior quick analytical reconstruction uncorrected for scattering events [
35]. As these preliminary reconstructions are based on very noisy sinograms, we expect as well the scattering simulation may fail to calculate a reliable estimate. Similarly, van Elmbt et al. [
19] suggested an additional component of true coincidences affecting the ends of the profile tails of the rebinned sinograms. They assumed that this signal may come from pair production in the LSO crystals by the X-ray bremsstrahlung above 1.022 MeV. This component may have an impact on the scaling of the scattered sinogram to the emission sinogram. This uniform background may also affect the random correction process, but no specific correction was applied to account for.
Finally, we used a single acquisition for the extraction of figures of merit used in this work. The use of multiple independent acquisitions for each of the measuring points would have certainly strengthened the statistical robustness of each measurement, especially when the radioactive concentration in the spheres was less than 1 MBq mL−1. We reasonably assume that the use of experimental replicates does not change the basic conclusions of this work.
Conclusions
This study assessed the MDA for a 30-min 90Y-PET imaging session and the accuracy of reconstructed activity concentration according to lesion size and its concentration. The benefit of using TOF information is discussed and shown for concentrations below 2 MBq mL−1 or small lesion size. An activity concentration below 1 MBq mL−1 may be detected but with a variable quantitative accuracy depending on the lesion size and reconstruction parameters. 90Y-PET imaging is probably not feasible for most RAIT procedures due to the very low uptake by the lesions during this type of treatment.
However, the utility of 90Y-PET imaging after SIRT in hepatic tumors was demonstrated and could be an opportunity to retrospectively check the distribution of microspheres in the liver and the tumor, and to accurately compute the absorbed dose.
Finally, future developments will be dedicated to a better understanding of the physics and imaging properties of 90Y image and data, a further specialization of the acquisition protocols, and derive adequate correction schemes.
Competing interests
The authors declare that they have no competing interests.
Authors’ contributions
TC performed the data acquisition, analysis, and interpretation. TC is the main author of the manuscript. TE and EG contributed significantly to the patient study and reviewed and approved the final content of the manuscript. CBM, CA, CR, and LF contributed to the intellectual content (supervision) and critical review of the manuscript. JB and FS interpreted the results, gave a critical review of the work, and contributed to the enhancement of the manuscript. FKB participated to the study design and coordination, and helped draft the manuscript. All authors read and approved the final manuscript.