MEASUREMENT OF YTTRIUM-90 BIODISTRIBUTION IN SELECTIVE INTERNAL RADIATION THERAPY (SIRT): A COMPARISON BETWEEN PET AND SPECT IMAGING

Received: 14 August 2018 Received in revised form: 04 September 2019 Accepted: 30 September 2019 Measurement of radionuclides biodistribution in post-radioembolization 90Y SIRT is a part of treatment evaluation, in which the assessment of biodistribution is used to evaluate the possible extrahepatic presence and the absorbed dose estimation for the tumor cells, healthy liver cells, and critical organs. As the dose-response analysis is performed based on this evaluation, the biodistribution measurement coming from post-imaging modality has a crucial role in achieving these goals. The two devices, Single Photon Emission Tomography (SPECT) and Positron Emission Tomography are discussed in some aspects, including the quality of quantitative images, performance characteristics, and absorbed dose considerations.


Radioembolization by Selective Internal
Radiation Therapy (SIRT) using 90 Y is a preferable treatment option for inoperable primary liver malignancy and colorectal metastases in the liver (Murthy et al., 2005). In these malignancies, external radiotherapy is not favorable due to liver cell intolerance and complications (such as hepatitis and progressive liver failure) after 30 Gy dose for liver (Stubbs & Wickremesekera, 2004). Yttrium-90 is a beta emitter with high energy (Eβmean=0.94 MeV) which can deliver a significant amount of dose (up to 80 Gy) and thus increase treatment efficacy (Pasciak, Bradley, & McKinney, 2017). The 90 Y radionuclide along with its carrier substance is provided through catheters directly to the intrahepatic artery. The carrier element can be either in the form of resin or glass microsphere; each has its particular application. * Corresponding author. Tel.: +62 8995139341; E-mail address: sita.gandes.p@gmail.com Resin microsphere is aimed for primary colorectal cancer while glass microsphere is aimed for unresectable primary liver cancer (Giammarile et al., 2011). Since the liver is surrounded by many critical soft tissues and organs, a careful treatment planning is required to minimize potential complications. It was reported that extrahepatic accumulation of 90 Y could lead to several side effects, such as ulceration, bleeding of the gastrointestinal track, and radiation pneumonitis (Leung TW, 1995;Murthy et al., 2007;Riaz et al., 2009). Due to the nature of malignant cells, in which they require more blood supply compared to the normal cell, there is a unique process which can be used as a basis for malignant treatment based on SIRT. The distinct character of malignant hepatic cells is that they obtain their blood supply mainly from the hepatic artery, while the normal hepatic cells get their blood supply from the portal vein (Biermann HR, 1951). Based on this fact, a radioembolization method can be employed to target the malignant cells selectively.
The procedure of radioembolization includes the restriction of arterial bifurcation connecting the malignant site to the adjacent organs (Lau et al., 2012). In doing so, first, vascular mapping and radioembolization processes are done by using a microcatheter guided by fluoroscopy. Second, to estimate the potential spreading of the radionuclide to the lung, an assessment of lung shunting fraction (LSF) by using a surrogate radiopharmaceutical ( 99 Tc-MAA) is performed by using gamma camera for adjusting the activity administered to be given to the patient. There is a guideline from the manufacturer for each type of microsphere. For resin microspheres, there is no reduction of administered activity for LSF<10%, 20% reduction of 10-15% LSF, 40% reduction for 15-20% LSF, and no treatment if the amount of LSF found to be higher than 20% (SIR-Spheres®, Sirtex Medical, Lane Cove, Australia). For glass microspheres, there is no such reduction factor, but the activity is limited by the amount of lung dose, i.e. <30 Gy for a single treatment and <50 Gy for cumulative treatment (TeraSphere®, MDS Nordion, Canada). The third step is activity planning. There are several methods available, namely, the body surface area (BSA) method (Kao YH, 2011), the partition model (PM) (Ho S, 1997), and medical internal radiation dosimetry (MIRD) (Gulec, Mesoloras, & Stabin, 2006;Stabin, 2006) for the resin microspheres. Whereas, for glass microsphere, the compartmental MIRD microdosimetry (Doherty, 2015) is used. Once the amount of activity has been decided, the 90 Y microsphere is infused into the hepatic artery by microcatheter guided by fluoroscopy (Lau et al., 2012). Following these steps, the treatment evaluation is performed after one day for the sake of three purposes. First, to assess the extrahepatic transfer of the radionuclide, second, to know the prognosis for the response of the lesion, and the last, to estimate absorbed dose obtained by both the cancerous and normal liver sites as well as the critical organs (Giammarile et al., 2011).
A general formalism of the Medical Internal Radiation Dosimetry (MIRD) developed by the Society of Nuclear Medicine (SNM) takes into account several parameters, including the biodistribution of the radionuclide and the contributing physical factor, which are specified as time-integrated activity and S-value (Stabin, 2008). Nowadays, the measurement of biodistribution uses a quantitative imaging approach which can provide a more accurate yet non-invasive procedure. Our particular concern in this review was to discuss the available tools for measuring biodistribution of activity in the post-treatment of 90 Y SIRT based on quantitative imaging. The scope of the review was intentionally restricted only to Positron Emission Tomography (PET) and Single Photon Emission Tomography (SPECT) as these are the primary modalities which can quantify biodistribution of activity of 90 Y microspheres in multiple views, which cannot be obtained in planar imaging, such as the gamma camera. It is of our interest because the biodistribution of activity can affect absorbed dose accuracy estimation and will guide the assessment of treatment efficacy, the decision for additional dose for the patient, and dose-response analyses (Stabin, 2008).

Current Techniques For 90y Imaging
Yttrium-90 is a common therapeutic radionuclide which can be produced via 89 Y(n,γ) 90 Y reaction in a nuclear reactor, or via the 90 Sr decay. Figure 1 shows the decay of 90 Y to 90 Zr which emits beta particles as the major decay scheme. It has a half-life of 2.6684 days, beta energy of 2.28 MeV (max) and 0.937 MeV (mean), tissue penetration depth of 11 mm (max) and 2.5 mm (mean) which make it suitable to irradiate the size of either metastatic or primary tumor in the liver (Bé, 2006). Also, the 90 Y decays completely within 13 days, a desirable characteristic that provides the maximum balance between treatment time and  The decay scheme of 90 Y to 90 Zr. Although the majority of 90 Y decays via βemission, there is a small portion of decay via the excitation state of 90Zr (E=1.76 keV) with branching ratio of (3.186±0.047)×10 -5 . The picture was reproduced based on Selwyn, Nickles, Thomadsen, DeWerd, & Micka, 2007. At the beginning of the use of radioembolization 90 Y SIRT, there was common knowledge that 90 Y was a pure βemitter and that was the reason why SPECT is the only choice to measure the biodistribution of 90 Y through the bremsstrahlung photons produced by the interaction between βand tissue (Bailey & Willowson, 2013). Though people realized that the image quality is relatively poor for quantitative measurement, SPECT was still extensively used for post-treatment evaluation as no other option was available. Despite the fact that Ford had predicted the 0 + -0transition of 90 Zr producing β + and been experimentally discovered by Johnson, et al. in 1955(Ford, 1955Johnson O, 1955), it still needs a long way to go until the PET-based quantitative measurement was started to be carried out. The existence of β + decay even with a very small branching ratio (3.186±0.047)×10 -5 (Selwyn, Nickles, Thomadsen, DeWerd, & Micka, 2007) leads to internal pair production interaction between β + and the β -, thus producing two annihilation photons with the same energy (~511 keV) in the opposite direction. This process is the basis of PET imaging (Saha, 2016).
The discussion below elaborates general technical issues of SPECT and PET in the use of 90 Y biodistribution measurement, it highlights the active use of each, and reminds the possible remaining challenges to be improved in both systems.

Spect
Several investigations had been carried out to evaluate SPECT-based postradioembolization measurement which is still facing many problems (Elschot M, 2011;L. M. Minarik D, Segars P, Gleisner KS, 2009;S.-G. K. Minarik D, Linden O, Wingardh K, Tennvall J, et al, 2010). Since the indirect bremsstrahlung x rays is a continuous spectrum which lacks photopeaks (Elschot M, 2011), the discrimination of counts based on the energy of the photon becomes difficult. As an effort to improve the SPECT sensitivity, a large energy window is required (Elschot M, 2011), which at the same time, allows for more noise caused by background radiation and scatter photons (Cherry, Sorenson, & Phelps, 2012). The resulted contrast also get reduced with the use of collimator septa, a mechanical aperture made of lead. The collimator septa are needed for locating the counting events, but at the same time, its existence also allows for high energy photon penetration and thus degrades the image quality (Elschot M, 2011). All those factors that eventually influence the overall performance characteristics of SPECT are examined in detail in the next section.
Some studies tried to improve the image quality of SPECT by concerning the following ways. The first is to find the optimal energy windows, either using Monte Carlo (MC) simulation of the full energy spectrum (Heard, 2004;Rault, Staelens, Holen, Beenhouwer, & Vandenberghe, 2010) or phantom studies (Minarik, Gleisner, & Ljungberg, 2008). Second, by modeling of the image degrading factor, which can be done by employing the simulation of imaging nuclear detector (SIMIND) (Rong, Du, Ljungberg, et al., 2012) or again, the MC simulation-based modeling (Elschot, Lam, Bosch, Viergever, & Jong, 2013). Based on the studies previous studies, it can be figured out that the optimal energy windows for detection of 90 Y are in the middle energy range of approximately 100-150 keV, 100-160 keV, or 90-125 keV, respectively (Heard, 2004;Shen S, 1994). The lower energy range (70-100 keV) is mainly resulted by characteristics photon coming from the lead collimator, while the higher energy range (200-300 keV) and >300 keV mostly comes from backscattering and septal penetration, respectively (Heard, 2004). The latter approach by modeling of degrading factors yields a significantly better image contrast, from 25% to 88% for the 37 mm sphere and less count error detection from 73% to 15%, accompanied by the increase of noise (Mattijs . Though the SPECT system suffers from poor sensitivity and spatial resolution, there are some counter facts, which in turn makes SPECT is advantageous in some cases (Bailey & Willowson, 2013). As the SPECT radionuclides have longer half-lives, there is no urgency of placing a medical cyclotron near the SPECT facility. A simultaneous multi-tracer study is also possible when it is needed to assess two or more radionuclide biokinetics in a single session. Lastly, the SPECT system requires less investment as compared to PET and has the greater use worldwide.

PET
The current practice of postradioembolization 90 Y SIRT imaging was firstly initiated by Nickles et al. in 2004 by using phantom studies (Nickles et al., 2004), while the clinical study was firstly performed by Lhommel et al. in 2009(Lhommel et al., 2009). Later, a PET-based dosimetry was confirmed to be feasible with the additional Time of Flight (TOF)  or even without the TOF installation (Werner et al., 2009). Indeed, the utilization of TOF is likely producing a better estimation of dose as investigated by several groups (Attarwala et al., 2014;Carlier et al., 2015;Elmbt, Vandenberghe, Walrand, Pauwels, & Jamar, 2011;Martí-Climent et al., 2014;Willowson, Forwood, Jakoby, Smith, & Bailey, 2012;Willowson, Tapner, & Bailey, 2015). It is because the use of TOF improves the sensitivity and spatial resolution due to its ability to provide high coincidence time resolution and accurate coincidence location detection (Walrand, Hesse, Renaud, & Jamar, 2015). The TOF design itself has been much developed for the last three decades, where the coincidence time resolution gets reduced from 650 ps in the first prototype, down to 320 ps in the latest digital TOF-PET, each providing the annihilation location prediction within 9 cm and 4.5 cm, respectively (Surti & Karp, 2016). The improvement of TOF-PET system nowadays is the result of the utilization of lutetium-based crystal (Lutetium oxy-orthosilicate, LSO, and lutetium-yttrium oxy-orthosilicate, LYSO), which is better than bismuth germinate (BGO) crystal in term of faster decay time and higher photon to light conversion output (Surti & Karp, 2016).
There are several promising facts to support the capability of PET in the use of 90 Y imaging. First, as the annihilation photons have approximately uniform energy (~511 keV), the discrimination window may easier to distinguish the true coincidence, despite the possibility of scattering, random, and multiple coincidences, and thus producing better sensitivity and spatial resolution. Second, the maximum positron energy of 90 Y is 758 keV that is comparable to the 633 keV of 18 F (Bailey DL, 2003). Consequently, the correction techniques including scatter, random, and attenuation correction are expected to be clinically suitable for the well-established correction procedure for 18 F (van Elmbt L, 2010).
As a functional imaging tool, it is well understood that PET also suffers from several limitations which degrade the accuracy of its measurement (IAEA, 2014). Naturally, the positrons travel a certain distance (~4 mm) before the annihilation events occur (Cherry et al., 2012). It inevitably produces an uncertainty in events location detection. The other issue is the probability of block effect or depth of interaction (J., 2013). In this phenomena, the interaction of annihilation photon does not only take place in the corresponding two detectors but also penetrate in the neighboring detectors which end up in the wrong line of response (LOR) (J., 2013). Furthermore, there is also a possible non-collinearity detection between the two annihilation photons direction. This effect  (Cherry et al., 2012). The other case is that the spatial resolution is not uniform for the whole field of view (FOV). Instead, the highest resolution is achieved in the middle of FOV and getting worse towards the periphery (Cherry et al., 2012). As the spatial resolution is characterized by point spread function (PSF), there is a so-called PSF algorithm included in the image reconstruction process to improve the spatial resolution (Rahmim, Qi, & Sossi, 2013).

IMAGE QUALITY IN QUANTITATIVE IMAGING
Description of image quality in quantitative imaging can usually be made based on several aspects, including spatial resolution, contrast, noise, and detectability (M. . Representative images of SPECT and PET are illustrated in Fig. 2. In general, PET images can provide a better representation of the lesion shape and higher activity uptake as compared to SPECT images (Y.-H. Kao et al., 2013).

Spatial Resolution
The spatial resolution is defined as the ability to distinguish two adjacent points and characterized by the full width half maximum (FWHM). An insufficient spatial resolution produces the blurring and also lead to partial volume effect which may reduce quantitative accuracy (Soret, Bacharach, & Buvat, 2007). Investigation of image resolution performed based on 90Y SPECT yields the range of 11.4 -12.5 mm (for a source to detector distance of 6 -6.5 mm) (Elschot M, 2011;Shen S, 1994). Meanwhile, the image resolution of PET laid between 2.2 -12.1 mm (Attarwala et al., 2014;Carlier et al., 2015;Elmbt et al., 2011;Martí-Climent et al., 2014;Willowson et al., 2012;Willowson et al., 2015), respectively. However, this value may differ between each type of device, the type of microsphere used, the type of scintillator crystal, and the reconstruction parameter as shown in Table III. Nevertheless, in general, the spatial resolution of PET is still superior relatively compared to SPECT.

Contrast and Noise
Elschot et al. compared the contrast and noise in the two imaging modalities (Mattijs . The contrast recovery and noise are the parameters of interest because it affects the distinction of the lesion object and the adjacent background. They employed several methods to assess the influence of PSF algorithm and the additional TOF to the resulted images. For all methods, the iterative reconstruction was set. The results are shown in Fig. 3 and Fig. 4.  The picture was taken from  According to Fig. 3, the overall images showed that the contrast recovery was higher for the PET image, where the highest contrast was achieved by the iterative+PSF+TOF and followed by the iterative+PSF method. In contrary, the use of PSF algorithm did not give any difference to the contrast recovery of SPECT. In line with the contrast recovery parameter, noise levels are also significantly higher in the PET images where the incorporation of PSF algorithm and TOF notably reduced the noise level. Ideally, it is desired to have a sufficient contrast with less noise presence. Therefore, the postreconstruction processing was performed and obtained the PET image (iterative+PSF+TOF) with the SPECT-like noise level to establish better contrast and noise composition.

Detectability
The detectability comparison between SPECT and PET was demonstrated according to two criteria, i.e. Rose criterion for false negative detection and the method described by Elschot et al. (Mattijs Elschot et al., 2013) to analyze the detectability in the intra-and extrahepatic biodistribution accumulation simulated in a phantom. Though the Rose criterion results indicate that all sites in the intra-and extrahepatic are visible, the other method denoted different outcomes. While the extrahepatic site remained visible up to as small as 28 mm in the diameter in PET image, all spheres in SPECT did not. In the intrahepatic site, the detectability thresholds are 13 and 22 mm for PET and SPECT, respectively. After all, this analysis signifies the superiority of the PET over SPECT in detecting the intra-and extrahepatic arteries.

PERFORMANCE CHARACTERISTICS
The excellence of quantitative imaging system can also be evaluated based on their performance characteristics. There are several measures, including spatial resolution, sensitivity, recovery coefficient, image uniformity, count rate linearity, calibration factor, and convergence of the image reconstruction algorithm (Cherry et al., 2012;. This review will cover the comparison of SPECT and PET characteristics in term of spatial resolution and sensitivity.

SPECT Characteristics
Elschot et al. characterized the spatial measurement and sensitivity of SPECT and listed in Table I-II. In their research, a material called polymethylmethacrylate or PMMA was placed in the position between the source and detector to create the realistic condition of electron scattering. The data were taken based on several different object-to-source distances, namely, 2, 6, and 11 cm was set for 1 cm of scattering material, and 11 cm for 5 and 10 cm of scattering material. Two kinds of windows were used, which consisted of a broad (50-250 keV) and a smaller (120-250 keV) energy windows. There were also two types of collimator used, medium-energy (MEGP) and high-energy-multi-purpose (HEGP) collimators (Elschot M, 2011). The spatial resolution and sensitivity were then determined according to NEMA guideline (Association, 2007), where the sensitivity was obtained by the ratio of acquired counts and the true activity measured in a dose calibrator. The sensitivity was calculated for two positions, 10 cm and 40 cm from the detector (Elschot M, 2011).
In Table I, the number on the outside and inside the bracket showed the FWHM and FWTM, respectively. The spatial resolution is substantially better when the source-detectordistance is kept remained close. Also, the use of smaller energy window (120-250 keV) and the HEGP collimator notably produced preferable  (Elschot M, 2011). The result in this study was in line with another publication by Shen et al., in which they obtained FWHM of 12.5 mm measured at 6.8 mm source-object-distance in the energy window of 55-285 keV (Shen S, 1994). Meanwhile, the nearest setting in the Elschot et al.'s work provided FWHM of 11.4 mm (Elschot M, 2011). The spatial resolution is written in mm unit.
S01D02 corresponds to the measurement with 1 cm of scatter material and line-source to collimator distance of 2 cm, S01D06 to the measurement with 1 cm of scatter material and line-source to collimator distance of 6 cm, etc.
This table was reproduced based on Elschot, et al., 2011. The sensitivity obtained here was also consistent with another publication where they obtained 11.6 and 6.75 cps MBq -1 for MEGP and HEGP, in the broad energy window (Shen S, 1994). As may be expected, the broad energy window yielded higher sensitivity as consequence of involvement of scattering fraction and background counts. For the same energy window, the MEGP collimator produced higher sensitivity than the HEGP (Elschot M, 2011;Shen S, 1994).

PET Characteristics
Many authors have recently reported investigations on PET spatial resolution with different modification factors as summarized in Table III. The assessment of spatial resolution in PET is commonly performed by using NEMA guideline NU 2-2007 (Association, 2007). The choice of the scintillation crystal, the utilization of TOF, reconstruction parameter seemed to be a matter of the final spatial resolution results. As stated before, the LSO is a good material for the PET detector design (Surti & Karp, 2016). Accompanied by a high number of iteration and subsets, though without TOF, the resolution of 90Y PET imaging can reach 2.5 to 4 mm resolution (Gates, Esmail, Marshall, Spies, & Salem, 2010).
The sensitivity determination also takes into account the true activity from dose calibrator measurement. Later, the counts reported in the PET system is divided by the true activity to obtain the sensitivity value, the same calculation performed in SPECT. Some factors are influencing the determination of sensitivity, such as detection efficiency, source-detector geometrical position, and acquisition mode (2D, with septa, or 3D, without septa). The detection efficiency concerns on conversion from photon to light, and is affected by some parameters, including scintillation decay time, atomic number, density, and the thickness of the scintillation material (Cherry et al., 2012).
The sensitivity of 90 Y PET was measured by several authors Ng et al., 2013;Werner et al., 2009). The absolute sensitivity measured using Biograph mCT-TrueV with TOF was 0.403 and 0.388 cps MBq -1 in the center of FOV and 10 cm of the FOV, respectively (Martí-Climent et al., 2014). Another work by Bagni et al. performed by using GE Discovery ST PET/CT provided similar results, which are 0.409 and 0.577 cps MBq -1 for the same position setting . In comparison with the well-known 18 F, the sensitivity of 90 Y is much smaller, about four order below the sensitivity of 18 F (  0.5 versus 9 cps MBq -1 ) (D'Arienzo et al., 2012). It can be explained by the fact that the positron abundance of 90 Y is only 32 per one million decays, whereas the 18 F gains almost 967 per 1000 decays (Pasciak et al., 2017).

ABSORBED DOSE CONSIDERATION
Image-based quantitative dosimetry was performed for both SPECT (iterative+PSF) and PET (iterative+PSF+TOF) images in both phantom and patient studies (M. . For phantom dosimetry, due to the small size of ROIs, they encountered PVE, and thus, a corrected dose was also estimated. The results were compared to the true value which was determined based on high-resolution CT data. The results were shown in Fig. 5-6 and Table  IV-V. Among all cumulative dose volume histograms (CDVHs) in the phantom study, the PET sphere CDVH appeared to be the closest to the true CDVH. The PET sphere CDVH was a bit differ to the true sphere CDVH in the form of the flat curve because it was influenced by the substantially higher noise in the PET image. Also, the overall dose measured based on PET (corrected and uncorrected dose) consistently showed the least difference to the true value. For instance, the smallest diameter measured by SPECT and PET achieved -75% (+69%) and -45% (+40%) dose difference to the true value, respectively. As the PVE did not strongly affect large diameter, the dose difference tends to be lower, which are -58% (+18%) and -11% (+6%) for SPECT and PET, respectively. Analog to CDVH results, the patient dosimetry showed more significant dose difference between the low dose (LD) and high dose (HD) in the PETbased dose estimation. For example, the dose difference in the SPECT-dose map (Fig. 5B) and PET-dose map (Fig. 5C) are 39 and 73 Gy. It was also supported by the substantial gap between the PET HD and SPECT HD CDVHs curves. According to this result, it was concluded that the high-resolution PET with the TOF installation and PSF reconstruction algorithm likely produced a better image of 90 Y distribution in post-radioembolization imaging.   The mean dose (Gy) in the low-dose ROI (LD) and the high-dose ROI (HD) are given for all patients. This table was repreoduced based on .

CONCLUSION AND REMARKS
The emerging use of 90 Y Radioembolization SIRT needs a reliable posttreatment imaging modality because of its function to produce a quantitative image as the basis of dose estimation and dose-response analyses. In term of image quality, considering the spatial resolution, noise, contrast, and detectability, the high-resolution PET-based image well demonstrates these parameters relatively compared to the SPECT-based image. The high-resolution PET may be achieved by the installation of TOF, the use of the iterative algorithm with a high number of subsets and iterations, the use of PSF algorithm, and the utilization of lutetium-based scintillation crystal. In contrast to the large underestimation or overestimation of dose in SPECT-based dosimetry, the PET dosimetry seemed to be closer to the true value. When the highresolution PET is used, the dose estimation still needs to consider the correction of PVE which may exist for the small lesion site.