Background
Radioembolization with
90Y, or selective internal radiation therapy (SIRT), is a catheter-based therapy with the potential of delivering a high-radiation dose directly to liver tumors or metastases meanwhile minimizing the effects on healthy liver parenchyma [
1‐
6]. The radiopharmaceutical consists of non-biodegradable
90Y-imprinted resin or glass-based microspheres with a diameter of approximately 20–40 μm. The tumoricidal effect is produced by beta particles emitted from
90Y incorporated on the surface of the resin matrix or integrally bound within the microsphere matrix, in the case of glass beads. The selectivity of SIRT is based on anatomic and physiological aspects of liver tumors and metastases that can be exploited for the selective delivery of the microspheres [
7‐
9]. Microspheres are injected into the hepatic artery and delivered directly into the smaller blood vessels that feed the tumor, therefore being trapped within the tumor microvasculature.
It is worth noticing that in recent years, radioactive
166Ho poly(
l-lactic acid) microspheres (166Ho-PLLA-MS) have been developed as a viable alternative to liver radioembolization with
90Y microspheres. In addition to high-energy beta-radiation,
166Ho emits gamma radiation that allows gamma camera quantitative imaging and dosimetry [
10,
11]. Furthermore, holmium is highly paramagnetic and can be visualized using traditional magnetic resonance imaging (MRI), thereby allowing for MRI-based absorbed dose calculations [
11‐
13].
90Y is a pure β-emitting radionuclide with maximum and average energies of 2.28 MeV and 933.7 keV, respectively. The corresponding maximum and average path lengths of the emitted β particles in soft tissue (1 g/cm
3) are 11 and 2.5 mm. Although
90Y has been traditionally considered as a pure β
− emitter, the decay of this radionuclide has a minor branch to the 0
+ first excited state of
90Zr at 1.76 MeV that is followed by a β
+/β
− emission. This internal pair production has been largely studied in the past because it is generated by a rare electric monopole transition (E0) between the states 0
+/0
+ of
90Zr. A thorough explanation of the emission of β
+ particles via internal pair production in the 0
+–0
+ transition of
90Zr is provided elsewhere [
14].
In the last years, the small positronic emission has been exploited for
90Y–PET imaging studies after liver radioembolization, with the aim to provide quantitative post-treatment imaging and to prospectively improve the therapy [
15‐
27]. Furthermore, the possibility of detecting β
+ emissions from
90Y by PET scanners is paving the way for an accurate patient-specific dosimetry. Because microspheres can be assumed to be a permanent implant, the use of appropriate dose calculation algorithms allows quantitative
90Y–PET images to be converted into absorbed dose maps. The accuracy of dose calculation is of paramount importance in liver radioembolization as side effects and treatment outcome are related to absorbed doses. Over the last few years, there has been an intensifying debate revolving around the most accurate dosimetric approach for the assessment of the absorbed dose in molecular radiotherapy. Many authors obtained absorbed dose maps from
90Y–PET data performing three-dimensional convolution of the kernel function with the cumulated activity. A recent analysis by Pasciak et al. [
26] proposed the use of local deposition technique, while previous studies used the analytical MIRD (Medical Internal Radiation Dose) approach.
The aim of this study is to investigate the impact of different computational approaches on absorbed dose calculation, starting from the quantitative accuracy of 90Y–PET. The workflow followed in the present work was structured into three logically sequential steps, which are detailed below.
(1) Firstly, the quantitative accuracy of reconstructed
90Y–PET data was evaluated using a cylindrical phantom and an IEC (International Electrotechnical Commission) body phantom for the assessment of recovery coefficients. Then, the PET camera’s ability to produce
90Y quantitative data was validated using an anthropomorphic phantom with a liver cavity and a background compartment. A hollow cylindrical insert was fixed into the liver cavity of the anthropomorphic phantom with the intent to simulate a hepatic lesion. Acquisitions were performed using a General Electric (GE) Discovery ST PET/CT scanner with a 6:1 tumor to background activity concentration ratio. The anthropomorphic phantom was acquired at days 1, 4, 5, 6, and 12 post-phantom preparation down to an activity concentration of 0.31 ± 0.02 MBq mL
−1 for the lesion insert (initial activity concentration 5.5 ± 0.3 MBq mL
−1) and 50 ± 3 kBq mL
−1 for the liver background (initial activity concentration 0.89 ± 0.04 MBq mL
−1). In the present phantom study, activity measurements were performed using a well-type radionuclide calibrator available on site and traceable to the Italian National Institute of Ionizing Radiation Metrology for the geometry being measured (accuracy within ± 5% at
k = 2 level, as recommended by AAPM report 181 [
28]). PET images were reconstructed using the standard GE algorithm (full 3D-OSEM VUEPoint HD two iterations, 15 subsets) and corrected for attenuation, scatter, random, and dead time using manufacturer software. Random and scatter corrections are incorporated into the iterative algorithm, with scatter correction performed using a fully 3D approach that considers both the axial and trans-axial scatter components.
(2) The second part of the study was devoted to absorbed dose calculations. Absorbed doses to the cylindrical insert were calculated from
90Y–PET images obtained in the previous stage, benchmarking different computational algorithms including (I) full Monte Carlo (MC) using Raydose [
29]; (II) Kernel convolution using Philips Stratos [
30]; (III) local deposition [
26]; (IV) MC N-Particle code [
31] (MCNP4c); and (V) MIRD analytical approach [
32].
(3) Ultimately, absorbed doses calculated using MC codes were compared with those obtained performing experimental measurements with high-sensitivity LiF:Mg,Cu,P thermoluminescent dosimeters (TLDs) inside the same cylindrical insert filled with a homogenous
90YCl
3 solution. LiF:Mg,Cu,P chips were fully characterized in terms of absorbed dose to water at the Italian National Institute of Ionizing Radiation Metrology (ENEA-INMRI) using the available
60Co reference gamma beam [
33]. In the present study, the average absorbed dose to water obtained from TLD measurements was considered the gold standard.
The observation reported here provides key insight into dosimetry calculations related with 90Y–PET data. To the best of the authors’ knowledge, there are no published literature studies aimed to compare measured absorbed dose with dose values obtained from 90Y–PET quantitative data.
The results reported in the present paper are organized into three separate sections, reflecting the abovementioned workflow.
Discussion
This study offers key insight into how different approaches to dosimetry calculations can accurately assess the absorbed dose after liver radioembolization using
90Y–PET data. At a time of increasing evidence for an absorbed radiation dose-effect relationships in radioembolization with
90Y-laden microspheres [
41,
42], it is the authors’ belief that there is an urgent need for accurate dosimetry in patients undergoing SIRT therapy. In fact, it is generally accepted that the absorbed dose both to tumor areas and to healthy liver is likely to have a significant impact on treatment’s clinical effectiveness. Furthermore, since microspheres administered during therapy remain trapped within the microvasculature, imaging requirements reduce to a single-time acquisition, thus simplifying the whole dosimetry process. As a consequence, there is great potential for SIRT dosimetry to become routine in the clinical practice.
Patient-specific dosimetry in SIRT suffers from a number of image-degrading effects. Besides PVE, that is likely to play a major role in small lesions, low true coincidence statistic due to the combined interplay of low beta plus branching ratio and low achievable acquisition time are likely to impact the image quality. Furthermore, the high scatter component intrinsic of
90Y PET imaging may negatively affect image quantification. Ultimately, patient respiratory motion is the primary cause of image blurring possibly leading to systematic dose underestimations [
43].
In this study, we concerned ourselves with quantitative accuracy of 90Y PET and the impact that different dose calculation approaches have on image-based dosimetry after liver radioembolization. Yet another question of interest for us was to compare absorbed doses obtained from MC calculations with those obtained through absorbed dose measurements performed with TLD chips into a liquid radioactive environment. To the best of our knowledge, this line of research is the first to investigate the degree of agreement between calculated and measured dose using 90Y–PET quantitative data.
Our data confirm and expand previous observations. Since most scanner do not support
90Y as a viable radionuclide option, scanner calibration should be performed using surrogate radionuclides and applying proper decay correction factors (e.g.,
86Y [
44,
45] or
22Na [
46]). Quantitative analysis showed that partial volume effects dominate spheres of volume below 11.5 mL (diameter < 28 mm), approximately. Recovery of activity concentration measured in the largest sphere on day 0 of imaging underestimated the true activity concentration of − 15%, approximately. Our results are in keeping with previous studies [
18,
47]. However, the recent QUEST study [
48] suggests that ToF-PET scanners are likely to improve contrast of hot spheres and increase RCs.
Another important consideration is that the low branching ratio for
90Y positron emission entails long scan times in order to reduce image noise. This is especially true for non-ToF–PET scanner, where the acquisition time plays a key role. Quantitative analysis in anthropomorphic geometry showed that despite the tumor insert being uniformly filled with a homogenous
90YCl
3 solution,
90Y–PET images presented significant spatial non-uniformity. Notwithstanding this, for activity concentrations exceeding 1.5 MBq/mL, the recovered activity concentration underestimated the known value by 4% approximately, demonstrating that accurate quantitation of
90Y is possible if long scan time is performed and if partial volume effects are accurately compensated for. Quantitative accuracy decreases for decreasing activity concentration levels. Interestingly, our results are in close agreement with those obtained by Mille et al. [
49] which recovered 95.4% of the known activity in a large cylindrical phantom despite obtaining noisy images with a voxel variability of 21%.
In the present study, the degree of uniformity was quantified in terms of homogeneity index. Quantitative data demonstrated a noise-like behavior of the homogeneity index, with the latter decreasing in proportion to the square root of the number of emitted positrons. This finding suggests that for the considered geometry and settings, image non-uniformities are mainly imputable to the low-counting statistics. It is interesting to note that activity underestimation both in the tumor insert and in the liver compartment is associated with increasing homogeneity index.
Of additional concern has been the impact of different dose calculation algorithms in anthropomorphic geometry. As a matter of fact, to date only a few studies attempted to compare different dosimetry methods in liver radioembolization. Mikell and co-workers [
50] compared four voxel-based dosimetry dose algorithms based on
90Y bremsstrahlung SPECT images (namely, MC, soft-tissue kernel with density correction (SKD), soft-tissue kernel (SK), and LD method). Interestingly, the authors found that for tumor, non-tumoral liver mean absorbed doses calculated with SKD, SK, and LD are equivalent to MC (within 5%). Deviations in the mean absorbed dose values increase when dosimetry is performed in the lungs, with right lung dosimetry being strongly influenced by the liver–lung interface [
50]. In another work, Grassi et al. [
51] compared STRATOS with a homemade software package (VoxelMed), both in phantom and in patients undergoing radiopeptide therapy with
177Lu. Both software performed well, with an agreement within 5% in phantom. Larger deviations were observed in patients.
In the present study, the large voxel variability of absorbed dose values is related to the abovementioned intrinsically noisy nature of 90Y PET images. However, on average, absorbed doses to the tumor insert have been found to be consistent to within 9%, except for dose calculations performed on day 1, where the standard deviation over the average absorbed dose values is 20%.
The most striking observation to emerge from the data comparison is the difference between dose values obtained with MC codes (namely, MCNP and Raydose) and other algorithms. The foremost cause of the slightly lower dose values obtained with Raydose is most probably due to the resampling of the CT grid from 512 × 512 to 128 × 128, executed for efficiency purposes. The larger voxel size, together with the inhomogeneity in the underlying PET images, produced a smoothing of the activity data. This is evident in the fine distribution of the dose (Fig.
6a–e): DVHs obtained with Raydose do not present any high-dose tail as any hot spots in a single voxel are shared with adjoining voxels. This is likely to produce lower doses especially at the liquid–PMMA interface. On the other hand, the lower doses obtained using MCNP could be interpreted as being a result of: (i) the input activity concentration, supposed to be uniformly distributed over the entire cylindrical object and (ii) the absence of spill-in activity from regions around the cylindrical insert in MCNP simulations. This is likely to result in a lower dose to the cylindrical insert.
Furthermore, when interpreting the results of the current study, it is also important to note that the presence of the PMMA wall of the cylindrical insert may had an impact on dose calculations, possibly leading to biased estimates between methods that directly account for the acrylic edges (Raydose, MCNP) and those that do not (kernel convolution, LD, and MIRD). The impact of the PMMA wall on dose calculations is somehow expected and inevitable, given the non-negligible presence of acrylic material (4.98 g) if compared to the total mass of the insert when filled with
90YCl
3 (24.1 g). From a computational perspective, there is no substantial material difference between tumor and background liver. As a consequence, this effect is not expected in a clinical scenario, unless dose calculations are performed at the liver–lung interface, where difference in material density may play a key role [
50].
Of note, average absorbed doses determined with the LD method are in excellent agreement with those obtained using the MIRD and the kernel convolution dose calculation approach. This result provides a strong argument for encouraging LD algorithm for the evaluation of absorbed doses in the clinical practice, where resource–intensive software packages are not always available. Similar conclusions have been drawn in recent research on the matter [
26]. However, the large dose variability at the voxel level—most likely due to a scenario with low true coincidences and high random fraction—raises questions about the possibility of using dDVH for radiobiological modeling if acquisitions are performed with a non-TOF scanner.
During the last 20 years, there has been increasing interest in the measurement of absorbed doses from internal beta emitters used in molecular radiotherapy. In particular, TLDs have been extensively used in the past to measure the absorbed dose both in phantoms [
52] and in animals [
53‐
55]. However, to the best of the authors’ knowledge, this is the first work that compares absorbed doses obtained from
90Y–PET quantitative imaging procedures with those obtained from experimental measurements using TL dosimeters. Direct Monte Carlo radiation transport is presently considered to be the most accurate of all currently available dose estimation algorithms [
56]. As a consequence, its validation using dosimeters calibrated against absorbed dose primary standards has become even more pressing. Our research has confirmed the accuracy of MC calculations in
90Y–PET dosimetry.
\( {\overline{D}}_w^{meas} \) and
\( {D}_w^{MCNP} \) are in the ratio 1.088. This difference needs to be further investigated, especially in the light of the well-known uncertainty of MCNP cross sections at extremely low energies [
57]. On the other hand, measured doses were in excellent agreement with absorbed doses calculated using a full MC approach, being
\( {\overline{D}}_w^{\mathrm{meas}} \) and
\( {D}_w^{\mathrm{Raydose}} \) in the ratio 1.03, well within the stated uncertainties.
The current research acknowledges a few limitations that should be noted to aid interpretation of the result. It is worth noting that the ability of a PET scanner to perform accurate
90Y quantification relies on the knowledge of the internal pair production branching ratio. Therefore, precise knowledge of the branch ratio of the 0
+–0
+ transition of
90Zr is important for an accurate quantification of
90Y accumulated inside the target region and detected via PET acquisition. Most recent literature findings report an internal pair production branch ratio as large as (3.186 ± 0.047) × 10
−5 measured by Selwyn [
58] and colleagues using a HPGe detector. This branch ratio value was used in the present study for the absolute calibration of the PET system. It is desirable that in the near future, the current uncertainty on the internal pair production branch ratio (about 1.5%) will be reduced by more accurate experimental measurements [
59].
Furthermore, we focussed on absolute quantification and dosimetry on a single cylindrical insert simulating a liver lesion. The extent to which both our methods and results can be generalized to other geometries and different PET scanners certainly requires further investigation. Finally, it is worth noticing that PET calibration measurements should be repeated over a reasonable long period of time to assess the system stability and to evaluate the final uncertainty in the calibration factor. Unfortunately, it was not possible in the present experiment to investigate impact of repeated calibration measurements due to the limited resources available.
Contrary to external beam radiotherapy, in which individual patient dosimetry is mandatory and there are legal requirements for accuracy (within 5% to a reference point), there is a lack of standardization in molecular radiotherapy (MRT), and it suffers from isolated efforts to harmonize quantification and dosimetry approaches. The current main source of uncertainty in internal dosimetry is in taking the step from dose measurements on simple reference geometries to quantitative imaging measurements of the complex and varying geometries of the activity localized in real patients. This passage is essential to comply with EC Directive 2013/59/EURATOM, Article 56, which states that individual dose planning for radiotherapy patients (including MRT) must be enforced in legislation by EU member states by 6 February 2018. It is desirable that in the near future, the final standard uncertainty in absolute quantification in complex phantoms be well below well 5% to ensure the uncertainty in clinical absorbed dose estimations comply with the requirement of 5% to a reference point. Comprehensive guidance has yet to be presented in this field, and there is no doubt that an internationally endorsed protocol on 90Y PET/CT quantitative imaging would lead to further advances in this area.