Simultaneous 166Ho/99mTc dual-isotope SPECT with Monte Carlo-based downscatter correction for automatic liver dosimetry in radioembolization

Background Intrahepatic dosimetry is paramount to optimize radioembolization treatment accuracy using radioactive holmium-166 microspheres (166Ho). This requires a practical protocol that combines quantitative imaging of microsphere distribution with automated and robust delineation of the volumes of interest. To this end, we propose a dual isotope single photon emission computed tomography (SPECT) protocol based on 166Ho therapeutic microspheres and technetium-99 m (99mTc) stannous phytate, which accumulates in healthy liver tissue. This protocol may allow accurate and automatic estimation of tumor-absorbed dose and healthy liver-absorbed dose. The current study focuses on a Monte Carlo-based reconstruction framework that inherently corrects for scatter crosstalk between the 166Ho and 99mTc imaging. To demonstrate the feasibility of the method, it is evaluated with realistic phantom experiments and patient data. Methods The Utrecht Monte Carlo System (UMCS) was extended to include detailed modeling of crosstalk interactions between 99mTc and 166Ho. First, 99mTc images were reconstructed including energy window-based corrections for 166Ho downscatter. Next, 99mTc downscatter in the 81-keV 166Ho window was Monte Carlo simulated to allow quantitative reconstruction of the 166Ho images. The accuracy of the 99mTc-downscatter modeling was evaluated by comparing measurements with simulations. In addition, the ratio between 99mTc and 166Ho yielding the best 166Ho dose estimates was established and the quantitative accuracy was reported. Results Given the same level of activity, 99mTc contributes twice as many counts to the 81-keV window than 166Ho, and four times as many counts to the 140-keV window, applying a 166Ho/99mTc ratio of 5:1 yielded a high accuracy in both 166Ho and 99mTc reconstruction. Phantom experiments revealed that the accuracy of quantitative 166Ho activity recovery was reduced by 10% due to the presence of 99mTc. Twenty iterations (8 subsets) of the SPECT/CT reconstructions were considered feasible for clinical practice. Applicability of the proposed protocol was shown in a proof-of-concept case. Conclusion A novel 166Ho/99mTc dual-isotope protocol for automatic dosimetry compensates accurately for downscatter and allows for the addition of 99mTc without compromising 166Ho SPECT image quality.


Background
Radioembolization has rapidly developed over the past decade. Conventionally, the amount of injected activity is based either on the patients body surface area or on the target liver volume for the commercially available resin and glass yttrium-90-loaded microspheres, respectively (SirSpheres® from Sirtex Medical and Therasphere® from BTG International). These methods are applied under the assumption that microsphere distribution is homogenous in the treated volume. However, due to patient characteristics and especially the heterogeneity of the microsphere distribution, these methods are too simplistic to allow for reliable dosimetry. In recent years, more and more centers have adopted the partition model, defining a tumor and non-tumor compartment, and allowing more personalized activity calculation by comparison with minimal required tumor dose and maximum allowable healthy liver dose from literature. Although much more accurate, the downside of the partition model is the delineation of the compartments, which is usually done manually. This can be cumbersome and hampers clinical widespread adoption. An automatic protocol could solve this. For radioembolization treatments with holmium-166-loaded ( 166 Ho) microspheres (Quiremspheres®, Quirem Medical), we propose a dual-isotope SPECT/CT protocol using pretreatment 166 Ho scout dose as treatment simulation and technetium-99m stannous phytate (a radiocolloid) for healthy liver tissue delineation [1]. 99m Tc-stannous phytate only accumulates in Kupffer cells by phagocytosis of the stannous phytate particle. As Kupffer cells are absent in tumorous tissue, this radiopharmaceutical has been used for many decades for the detection of liver disease and liver malignancies. The main advantage of simultaneous SPECT acquisition of both the treatment simulation with 166 Ho microspheres and healthy liver tissue segmentation with 99m Tc colloid is the absence of miss registration, due to patient-related factors. This manuscript will focus on the technical challenges concerning image acquisition and reconstruction with this dual-isotope protocol, mainly related to crosstalk of the two isotopes. Accurate quantitative reconstruction of 166 Ho SPECT has been demonstrated in previous work by Elschot et al. [2], but the presence of 99m Tc during the acquisition causes a significant contamination in the 166 Ho energy window. Vice versa, the 99m Tc photopeak window is contaminated due to downscatter from high energy 166 Ho emissions. This crosstalk interaction is illustrated in Fig. 1, depicting an energy spectrum of both 166 Ho and 166 Ho + 99m Tc. The Utrecht Monte Carlo System (UMCS) was extended to be able to correct for these crosstalk interactions.

Implementation of the photon modeling
In previous work, Elschot et al. demonstrated an iterative OSEM reconstruction method (Utrecht Monte Carlo System, UMCS) for quantitative 166 Ho SPECT, which includes the Monte Carlo-based modeling of photon contributions from the full 166 Ho energy spectrum, including bremsstrahlung [2]. In short, besides Compton and photo-electric effects in the patient, fast simulation of all collimator and crystal effects was accomplished by incorporating a look-up-table of point spread functions (PSF), which were generated with MCNP, a general purpose Monte Carlo radiation transport code [3]. For this work, this method was extended to include the effect of 99m Tc downscatter in the 81-keV photopeak window of 166 Ho in a similar fashion. The PSF as detected in the 81-keV (15% width) energy window depends on the energy of the photons before the detection and the distance to the collimator. To represent this, multiple PSFs were simulated for the Siemens Symbia Medium Energy collimator at source-detector distances of 1 cm, 5 cm, 12 cm, 24 cm, and 40 cm; PSFs for intermediate distances are interpolated at runtime. In order to restrict computation time, PSFs for 8 specific energies are used to cover the full energy spectrum. Each photon is associated with a PSF, determined by its energy as illustrated in Table 1. These PSFs represent photons that have an initial energy outside the 81-keV window and require a collimator or detector interaction (e.g., partial energy deposition, lead x-ray emission) to generate a detection. In addition to the PSFs for these indirect detections, a separate PSF was generated to represent photons that are "directly" detected, i.e., photons that have an energy in the 81-keV window (e.g., after scattering in the patient). These events are weighted  by the detection probability, determined by the energy of the photon relative to the energy window and the energy resolution of the gamma camera.

Image reconstruction and validation
Image reconstruction of a dual-isotope acquisition was performed in three consecutive steps, in which UMCS was used as the forward projection simulator for each step. Firstly, 99m Tc reconstruction: crosstalk of 166 Ho in the 140-keV 99m Tc photopeak window was corrected for during iterative reconstruction by addition of an upper scatter window (centered at 170 keV). A k-factor of 0.93 (or 0.96 including the window width ratio) was applied to account for the slightly decreased 166 Ho scatter contribution around 140 keV (which can be appreciated from the small slope in the 166 Ho spectrum in Fig. 1 between 140 keV and 170 keV). The k-factor was determined as the mean count ratio between the projection images of these energy windows in 65 patient SPECT acquisitions.
Thirdly, 166 Ho reconstruction: crosstalk from 99m Tc was corrected for by adding the simulated 99m Tc-downscatter projections as a scatter window during iterative reconstruction of 166 Ho.

Evaluation
The crosstalk simulation and performance of the image reconstructions were assessed by conducting and comparing phantom studies. A 99m Tc line source centered between two 40 × 40 × 10 cm 3 slabs of polymethyl methacrylate (PMMA) scatter material and a 6.3-L cylindrical phantom, filled with 50 MBq 99m Tc, were scanned on a Siemens Symbia T16 SPECT/CT, recording the clinically used energy windows of both 99m Tc (140 keV, 15% width) and 166 Ho (81 keV, 15% width), along with two scatter windows centered at 118 keV and 170 keV (widths 12%). The recorded projections of both phantoms were compared with simulated projections of a digital phantom (of equal shape and activity, based on the attenuation CT image of the setup) to assess quantitative accuracy of the 99m Tcdownscatter simulations (i.e., the extent to which 99m Tc contaminates the 81-keV 166 Ho photopeak window). Projections of the 99m Tc line source were used to validate the shape of the PSFs corresponding to 99m Tc downscatter in the 81-keV energy window.

Determination of the 166 Ho-99m Tc activity ratio
To define a practical balance between the amount of administered 166 Ho and 99m Tc, an anthropomorphic torso phantom was measured (Fig. 2a). Within the 1200-ml water filled liver compartment of the phantom, a 130-ml insert was placed containing a 53-MBq 166 Ho solution. Two scans were performed. In the first scan, the phantom contained 166 Ho only, for the second scan, 35 MBq 99m Tc was added to the liver compartment. The data was reconstructed with our protocol and analyzed for quantitative accuracy for both the 166 Ho only scan and the dual-isotope scan. A clinically acceptable 166 Ho/ 99m Tc activity ratio was defined, based on the results of the latter phantom study and on a visual interpretation and consensus reading by two nuclear medicine physicians and a medical physicist.

Quantitative assessment
The quantitative reconstruction accuracy of a 166 Ho activity distribution is dependent on the size of the distribution and is further influenced by the presence of 99m Tc. The National Electrical Manufacturers Association (NEMA) image quality (IQ) phantom ( Fig. 2b) was used to determine the sphere size-based activity recovery for 3 different 99m Tc background concentrations. The 6 spheres in the NEMA IQ phantom (diameters 10, 13, 17, 22, 28, and 37 mm, volumes, 0.52, 1.15, 2.6, 5.6, 11.5, and 26.5 ml) were filled with a 0.8-MBq/ml 166 Ho solution and 3 acquisitions were performed with varying 99m Tc activity concentrations in the background compartment (0, 6, and 11 kBq/ml).
The phantom was partially filled with agar-agar to reduce the total volume of the background compartment (5.5 L instead of the conventional 9.7 L), in order to obtain a ratio between the total activity and the activity concentration that resembles a clinical situation more closely. Images of the 166 Ho activity distribution were reconstructed using the dual-isotope reconstruction protocol described above. Volumes of interest (VOIs) were placed over the spheres, where the size of each VOI matched the size of the corresponding sphere in the phantom. The positioning of the VOIs was performed automatically (using a fit routine to optimize the mean activity within the VOIs). The mean 166 Ho activity concentration within each VOI was divided by the known injected activity concentration to obtain the activity recovery coefficients. Because in clinical practice, the actual shape or volume of an activity deposition is not necessarily known beforehand, a VOI may be drawn wider around the hot spot to determine the total activity within the spot. To emulate this, additional to the VOIs that matched the sphere sizes, the diameters of the VOIs were increased by 20 mm. The increase of 20 mm was found to be sufficiently large to limit spill-out due to the partial volume effect, while not being so large as to overlap with neighboring spheres. Furthermore, activity recovery coefficients were computed for all spheres and for each iteration in the reconstruction (100 iterations using 8 subsets of 15 projections, 120 angles in total) to investigate the influence of the number of iterations on the quantitative accuracy of the reconstruction protocol.

Proof of concept in clinical setting
If a patient is a candidate for radioembolization, first, a visceral angiography is performed to assess the arterial blood supply of the liver and tumors. During the same angiography, positioning of the microcatheter for a radioembolization treatment is determined by the interventional radiologist. To simulate the actual radioembolization treatment, a scout dose of 250 MBq 166 Ho microspheres is administered in the predetermined microcatheter positions. Subsequently, a SPECT/CT is acquired to assess treatment safety (i.e., excluding extrahepatic depositions of activity) and assess the intrahepatic distribution of the particles for treatment dosimetry. As part of a prospective clinical study (HEPAR PLUS), informed consent for the acquisition of the proposed dual-isotope SPECT/CT in a patient was obtained [4]. After a regular 166 Ho scout dose procedure (with 250 MBq), a 166 Ho-only SPECT/CT was acquired. Subsequently, 10 min after intravenous injection of 50 MBq 99m Tc-stannous phytate, a dual-isotope SPECT/CT was acquired. The imaging protocol of the dual-isotope SPECT/CT was based on the results of our phantom study.

Results
Projection images of the 99m Tc line source in the 81 keV and 140-keV energy windows were simulated and compared with the measured projection images (Fig. 3, left column). Comparing the number of counts in the 81-keV window (C 81 ) with the number of counts in the 140-keV window (C 140 ), the simulation underestimated C 81 /C 140 by 8% (0.59 simulated versus 0.64 measured). Quantitative accuracy of the 99m Tc forward projections in the 140 keV and the 81-keV window was also assessed using the cylindrical phantom. A homogeneous activity distribution was imposed inside the digital phantom, matching the total activity as measured in a dose calibrator (Fig. 3, right column). The simulated projections overestimated the counts in the 140-keV energy window by 3.5%, and the counts in the 81-keV window by 10.7%. Since no cross-calibration was performed between the dosecalibrator and the scanner, the simulated projections were intrinsically quantitative (i.e., unscaled).
The projection images of the anthropomorphic phantom showed that, for an equal amount of activity, 99m Tc contributes approximately twice as many counts to the 81-keV window as 166 Ho, and 4 times as many counts to the 140-keV window. These numbers strongly depend on the distribution of the isotopes and the geometry of the patient. Based on the results of the anthropomorphic phantom study and the consensus reading, a clinical activity ratio of 5:1 (250 MBq 166 Ho:50 MBq 99m Tc) was chosen. Thus, counts contributing to the 81-keV window are mostly due to 166 Ho (5:2), while at the same time the intensity in the 140-keV window is approximately balanced between 166 Ho and 99m Tc (5:4). The 166 Ho only reconstruction of the anthropomorphic phantom overestimated the activity by 7% compared with the calibrated injected activity. Addition of 99m Tc in the liver compartment further reduced the accuracy, overestimating the 166 Ho activity by 14%.
The reconstructed 166 Ho images of the NEMA IQ phantom were analyzed by defining volumes of interest (VOIs) over the spheres that matched the actual sphere sizes. Figure 4 shows the activity recovery for the six spheres in which the expected activity was based on the dose-calibrator measurements (used as ground truth). In this phantom study, activity recovery was highest for the 166 Ho only images, i.e., addition of 99m Tc decreased the activity recovery. When the diameters of the VOIs were increased by 20 mm, however, adding 99m Tc increased the apparent 166 Ho activity.  Figure 5, top row, shows the rate of convergence for 3 different sphere sizes and the dependence on 99m Tc background activity concentration. Based on these results, 20 iterations were found to provide an acceptable degree of convergence for all but the smallest spheres. The bottom row in Fig. 5 shows how the increase in VOI diameter influences the recovered activity (after 20 iterations). Increasing the diameter of VOIs beyond 40 mm caused VOIs to start to overlap with neighboring spheres.
Visual interpretation of the 166 Ho-only SPECT/CT and the 166 Ho reconstruction of the dual-isotope SPECT/CT in the patient setting showed no differences, as shown in Fig. 6.   166 Ho-reconstruction SPECT/CT from dual-isotope acquisition (middle), and corresponding 99m Tc-stannous phytate reconstruction SPECT/CT from a dual-isotope acquisition (right) in a patient. Visual assessment of the images by two nuclear medicine physicians shows no differences and the 99m Tc-stannous phytate SPECT/CT shows a negative correlation (no uptake in tumor tissue, only uptake in healthy liver tissue)

Discussion
The importance of dosimetry in radioembolization has been emphasized after the negative results of large prospective trials [5][6][7][8][9][10]. In a post hoc dosimetric analysis of one of these trials, performed in a subset of patients, a clear tumor-absorbed dose-overall survival correlation was found [11,12]. Patients receiving a sufficient tumor-absorbed dose (≥ 100 Gy) had a significant longer median overall survival and had better objective response rates [11]. Thus, currently a clinical unmet need for improved dosimetry exists which is user/physician friendly. Unfortunately, current available methods or software packages are often very time consuming, incorporate errors by miss registration of different modalities and incorporate additional inter-observer variability/errors by the required manual delineations by the treating physicians.
Based on a previous publication by Lam et al. in 2013, the combination of a pretreatment simulation SPECT/CT with 99m Tc-macroaggregated albumin ( 99m Tc-MAA) and physiological healthy liver tissue delineation with 99m Tc-sulfur colloid SPECT/CT seemed feasible [13]. 99m Tc-labeled radiocolloids only accumulate in healthy liver tissue via phagocytosis of the colloid particle by Kupffer cells [1,13]. As Kupffer cells are absent in tumorous tissue, this allows an easy differentiation between tumorous and healthy liver tissue. In the study protocol by Lam et al, healthy liver tissue was delineated on the 99m Tc-sulfur colloid SPECT/CT using a 10% threshold of the maximum pixel value, and on the 99m Tc-MAA SPECT/CT liver tissue below a 10% threshold was simplified to 0 (i.e., "non-irradiated functional liver tissue"). By subtraction of both images with the according thresholds, 4 different compartments (based on physiological data) could be defined as follows: irradiated tumor, irradiated healthy liver tissue, nonirradiated healthy liver tissue, and tumor necrosis. In the population of 122 patients treated with 90 Y-loaded microspheres, clinical toxicity data was correlated to the absorbed dose in healthy liver tissue [13].
However, the combination of 99m Tc-MAA and 99m Tc sulfur colloid is impractical, mainly because of the separate acquisition of the two SPECT/CTs. Furthermore, 99m Tc-MAA is known to be a poor predictor of intrahepatic distribution of microspheres [14,15]. To bring pre-treatment dosimetry to a higher level, a more practical imaging protocol and a more predictive particle are needed. In previous studies, the use of a small amount of 166 Ho microspheres with an activity of 250 MBq, was determined to be safe and superior to 99m Tc-MAA [15][16][17][18][19]. Thus the main aim of this study was to investigate the technical feasibility of a dual-isotope protocol combining 166 Ho microspheres and 99m Tc sulfur colloid.
Based on this work, a quantitative reconstruction framework for dual-isotope scanning of a 250 MBq 166 Ho scout dose with 50 MBq 99m Tc-stannous phytate seems feasible for simultaneous treatment simulation and healthy liver tissue delineation. Both isotopes influence each other's image quality. The 50 MBq 99m Tc seems sufficient for the task of segmentation of the healthy liver, although this work focusses on the 166 Ho image quality rather than 99m Tc (in continuing research the 166 Ho/ 99m Tc activity ratio is evaluated in a clinical setting, where the quality of the 99m Tc reconstruction directly impacts dosimetry). The 166 Ho photopeak window is hampered by the downscatter from the high energetic (> 1.3 MeV) gamma emissions in the 166 Ho emission spectrum, bremsstrahlung and the generation of (K-shell) lead x-rays in the collimator. Addition of 99m Tc activity potentially degrades 166 Ho image quality further.
Phantom experiments revealed that, in absence of 99m Tc, UMCS overestimates the 166 Ho activity by approximately 10%, when compared with the activity reported by our dose calibrator. This difference can be readily compensated for by applying a crosscalibration factor to the reconstructed images. In the presence of 99m Tc however, the accuracy of 166 Ho activity recovery was reduced, especially when small volumes were considered. Figure 5 indicates that the 99m Tc background causes the reconstruction to converge more slowly, most notably in the small spheres. Consequently, for a given number of iterations, the resolution recovery is negatively impacted by 99m Tc background. However, when VOIs around the spheres were dilated in order to avoid spillout due to the limited resolution, the recovered activity was overestimated compared with the no-background case, which may partly be caused by counts from the 99m Tc background being attributed to 166 Ho, due to errors in downscatter correction. Experiments with the 99m Tc line source and cylindrical source (Fig. 3) indicate that 99m Tcdownscatter simulations contain inconsistencies with respect to the measured projections of up to 10%, which may be due to a combination of causes. One of which may be that material compositions cannot be discriminated from the attenuation correction CT, and a material is assumed by the system (i.e., soft tissue [water] or bone). In case of the line source, which was centered in PMMA scatter material, mass attenuation coefficients of water were applied by UMCS rather than those of PMMA. Other causes may be due to the use of a limited number of pre-simulated PSFs and potential inaccuracies in modeling of the gamma cameras' energy resolution, peaking imperfections, and dead time effects.
There are several limitations to our reconstruction protocol. In the current protocol, patient breathing was not accounted for. Patient breathing is known to result in an underestimation of actual activity depositions and blur SPECT/CT images [20]. This issue may be resolved by applying breath gating during image acquisition, although this feature is currently not supported on our SPECT/CT imaging devices.
The acquisition time is similar to the widely applied 99m Tc-MAA SPECT; however, the post-processing computation time is approximately 3 min per iteration on a regular desktop PC, resulting in a total reconstruction time of 2 h (including 2 reconstructions of 20 iterations and a downscatter simulation). The current implementation of the software lacks multi-processing support, which limits the computational speed, especially given the potential for parallelizable processing. A new version of the software will be developed which will allow for a much greater speedup. In previously conducted prospective studies on 166 Ho-radioembolization, all patients had both the treatment simulation procedure as well as the actual treatment on the same day (so-called 1-day treatment). Based on our experience with 1-day treatments and current findings, 20 iterations with the current software were clinically acceptable in a daily workflow.
The future potential of this imaging protocol is clear. First, the 99m Tc-colloid SPECT has the potential to get (semi-)automated image segmentation, thus avoiding interobserver differences caused by manual delineation, which is known to result in variations and additional errors in dosimetry [4]. Second, the combined acquisition avoids errors introduced by registration of different modalities.
The results of this study are promising; however, this dual-isotope protocol needs additional research on its applicability in the clinical setting.

Conclusion
A realistic quantitative reconstruction framework for dual-isotope scanning of 166 Ho and 99m Tc was successfully developed and seems feasible for clinical practice.
This dual-isotope protocol may resolve several technical issues in radioembolization dosimetry.