PET/CT and SPECT/CT imaging of 90Y hepatic radioembolization at therapeutic and diagnostic activity levels: anthropomorphic phantom study

Purpose: Prior to 90Y radioembolization procedure a pretherapy simulation using 99mTc-MAA is performed. Alternatively, a small dosage of 90Y microspheres could be used. We aimed to assess the accuracy of lung shunt fraction (LSF) estimation in both high activity 90Y posttreatment and pretreatment scans with isotope activity of ~100 MBq, using different imaging techniques. Additionally, we assessed the feasibility of visualising hot and cold hepatic tumours in PET/CT and Bremsstrahlung SPECT/CT images. Materials and Methods: Anthropomorphic phantom including liver (with two spherical tumours) and lung inserts was filled with 90Y chloride to simulate an LSF of 9.8%. The total initial activity in the liver was 1451 MBq, including 19.4 MBq in the hot sphere. Nine measurement sessions including PET/CT, SPECT/CT, and planar images were acquired at activities in the whole phantom ranging from 1618 MBq down to 43 MBq. The visibility of the tumours was appraised based on independent observers scores. Quantitatively, contrast-to-noise ratio (CNR) was calculated for both spheres in all images. Results: LSF estimation: For high activity in the phantom, PET reconstructions slightly underestimated the LSF; absolute difference was <1.5pp (percent point). For activity <100 MBq, the LSF was overestimated. Both SPECT and planar scintigraphy overestimated the LSF for all activities. Foci visibility: For SPECT/CT the cold tumour proved too small to be discernible (CNR <0.5) regardless of the 90Y activity in the liver, while hot sphere was visible for activity >200 MBq (CNR>4). For PET/CT, the cold tumour was only visible with the highest 90Y activity (CNR>4), whereas the hot one was seen for activity >100 MBq (CNR>5). Conclusions: PET/CT may accurately estimate the LSF in a 90Y posttreatment procedure. However, at low activities of about 100 MBq it seems to provide unreliable estimations. PET imaging provided better visualisation of both hot and cold tumours.

Purpose: Prior to 90 Y radioembolization procedure a pretherapy simulation using 99m Tc-MAA 23 is performed. Alternatively, a small dosage of 90 Y microspheres could be used. We aimed to 24 assess the accuracy of lung shunt fraction (LSF) estimation in both high activity 90 Y 25 posttreatment and pretreatment scans with isotope activity of ~100 MBq, using different 26 imaging techniques. Additionally, we assessed the feasibility of visualising hot and cold 27 hepatic tumours in PET/CT and Bremsstrahlung SPECT/CT images. 28 Materials and Methods: Anthropomorphic phantom including liver (with two spherical 29 tumours) and lung inserts was filled with 90 Y chloride to simulate an LSF of 9.8%. The total 30 initial activity in the liver was 1451 MBq, including 19.4 MBq in the hot sphere. Nine 31 measurement sessions including PET/CT, SPECT/CT, and planar images were acquired at 32 activities in the whole phantom ranging from 1618 MBq down to 43 MBq. 33 The visibility of the tumours was appraised based on independent observers' scores. 34 Quantitatively, contrast-to-noise ratio (CNR) was calculated for both spheres in all images. 35 Results: 36 LSF estimation: For high activity in the phantom, PET reconstructions slightly underestimated 37 the LSF; absolute difference was <1.5pp (percent point). For activity <100 MBq, the LSF was 38 overestimated. Both SPECT and planar scintigraphy overestimated the LSF for all activities. 39 Foci visibility: For SPECT/CT the cold tumour proved too small to be discernible (CNR <0.5) 40 regardless of the 90 Y activity in the liver, while hot sphere was visible for activity >200 MBq 41 (CNR>4). For PET/CT, the cold tumour was only visible with the highest 90 Y activity (CNR>4), 42 whereas the hot one was seen for activity >100 MBq (CNR>5).
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 10, 2022. ; https://doi.org/10.1101/2022.07.07.22277361 doi: medRxiv preprint Introduction 48 Radioembolization is a method of hepatic tumours treatment where microspheres 49 containing Yttrium-90 are administered into the arterial vasculature of the liver, to be 50 delivered into close proximity of the tumour. The tumour is then irradiated by β  particles 51 emitted in 90 Y decay (1). 52 Prior to radiation microsphere therapy, patients undergo relevant planning studies, including 53 mapping angiography and 99m Tc-labeled macroaggregated albumin ( 99m Tc-MAA) imaging (2). 54 A diagnostic dose of 99m Tc-MAA is injected through hepatic arteries, just like 90 Y 55 microspheres. This pretherapy simulation is performed to predict 90 Y microsphere 56 distribution (3). One of the main goals of the 99m Tc-MAA examination concerns the issue of 57 the safety of the planned therapy: estimation of the lung shunt fraction (LSF) as well as 58 detection of potential extrahepatic depositions (4,5). This is because radiation induced 59 pneumonitis and sclerosis due to hepatopulmonary shunting of 90 Y microspheres is a major 60 toxicity concern in radioembolization procedure (6). 61 Currently, in most of the nuclear medicine facilities that provide radioembolization 62 treatment, LSF is routinely estimated by 99m Tc-MAA planar scintigraphy performed without 63 accounting for attenuation or scatter effects. Because lung and liver have different tissue 64 densities, the LSF will be overestimated when attenuation correction is not applied (6,7).
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted July 10, 2022. ; https://doi.org/10.1101/2022.07.07.22277361 doi: medRxiv preprint In contrast, SPECT/CT imaging may improve the accuracy of the LSF assessment. According 66 to the EANM standard operational procedure on "a unified methodology for 99m Tc-MAA pre-67 and 90 Y peri-therapy dosimetry in liver radioembolization with 90 Y microspheres" patients 68 with substantial lung shunt visible in planar imaging are recommended to have an additional 69 SPECT/CT scan performed to ensure a more accurate quantification of the LSF (3). 70 Regardless of which 99m Tc-MAA imaging method (planar or SPECT) is used to predict the 71 distribution of 90 Y microspheres, there are a number of factors affecting the mismatch 72 between the simulation with MAA and the actual therapy distribution, the most important 73 of which are: differences in shape and size distribution between MAA particles and 74 therapeutic ones, different number of injected particles between the 99m Tc-MAA simulation 75 and the therapy session, and uncertainty about the stability of MAA after labelling (3,8,9). 76 Another option that has been proposed is to inject the patient with the so called scout dose 77 consisting of a small batch of microspheres, identical to those used for treatment. The goal is 78 to better simulate the treatment (10). However, the pretreatment activity of β  emitters, 79 used normally to destroy diseased tissue, must be limited. For 90 Y the estimated safety 80 threshold is about 100 MBq (5,8,10). Of course, such low activity makes imaging even more 81 challenging. 82 In this study we aimed to assess the accuracy of LSF estimation by means of phantom 83 imaging. To this end we analysed both high activity 90 Y posttreatment scans and 84 pretreatment scans with the isotope activity of ~100 MBq, using different nuclear imaging 85 techniques: PET/CT, Bremsstrahlung SPECT/CT as well as planar imaging. 86 90 Y imaging using gamma camera (both planar and SPECT) is based on registration of 87 Bremsstrahlung in a wide energy window. Therefore, it poses a major problem when exact . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted July 10, 2022. For PET, the very low positron emission probability is the main fundamental limit to image 96 quality (11). The small branching ratio related to the internal pair production during 90 Y 97 decay, directly related to the acquired number of coincidences, may be insufficient to 98 accurately estimate lung shunting, especially at low activities of 90 Y (8). However, even in 99 high activity 90 Y posttreatment scans, activity concentration in the lung area may be very 100 low, which makes its quantification challenging due to scarcity of registered counts. 101 Based on our previous experiences with 90 Y phantom studies (11,15), we continued the topic 102 of visualization of hot and cold foci in the phantom for different isotope concentrations. The 103 use of an anthropomorphic phantom (instead of Jaszczak or NEMA phantoms) with liver 104 insert containing fillable spheres allowed us to approximate conditions similar to clinical 105 ones. As in our previous study, tumour visibility analysis based on PET/CT and SPECT/CT 106 imaging data was performed both qualitatively and quantitatively using contrast-to-noise 107 ratio as a quantitative parameter (11). Calculated quantitative measure was then compared 108 to the results of qualitative assessments. With the clinical aspect of this current work in 109 mind, we also simulated an extrahepatic lesion in order to investigate whether it can be 110 detected using hybrid nuclear imaging techniques, in both 90 Y post-and pretreatment scans.
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted July 10, 2022.  Table 1. Because both . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 10, 2022.  . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 10, 2022. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 10, 2022. ; https://doi.org/10.1101/2022.07.07.22277361 doi: medRxiv preprint  For PET imaging the liver and lung VOI were manually delineated on one CT scan, referred to 198 as "reference scan", which in turn was rigidly registered to all other CT scans. All VOIs were 199 transformed accordingly from the reference scan to the other CT scans (5,8). The LSF was 200 calculated directly as the activity in the lungs divided by the activity in the liver and lungs. 201 Additionally, for PET imaging an extra region without activity within the phantom was 202 delineated to simulate an LSF of 0% (LSF simulated ). Since its volume was smaller than that of 203 the lungs, the activity in the cold region was re-scaled to avoid underestimation of the 204 LSF simulated . The VOI of the cold region was transformed as described above. 205 To include a correction for the natural background in PET imaging, the mean activity 206 concentration expressed in MBq/ml for the whole phantom volume was first determined, 207 and then the background activity both for liver and lung volume was calculated. The 208 LSF BKG_corrected was then calculated as the activity in the lung VOI divided by the total activity 209 in the liver and lung VOI, whereby the 90 Y activity in the VOI was reduced by the value of 210 background activity in the corresponding VOI.
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 10, 2022. To calculate the LSF based on SPECT images there was no need to convert "counts per pixel" 215 to the activity concentration. Since counts are assumed to be proportional to activity, the 216 LSF can be calculated as the counts from the lung VOI divided by the total counts for the liver 217 and lung VOIs. 218 The methods for organ segmentation as well as VOI transformation between different CT 219 scans were similar to those used for PET. The LSF of 0% was simulated in a similar manner as 220 in PET.  231 We used two approaches to calculate the LSF value. The first was to directly calculate the LSF 232 as defined: the counts from the lung ROI divided by the total counts for the liver and lung 233 ROIs.
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. CNR is over a certain threshold, which is usually between 3 and 5 (16).
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 10, 2022.  The LSF simulated , based on the VOI with no activity, was stable over the range of high activities 302 down to approximately 500 MBq, and its value was then less than 1% (when the background 303 correction was applied). However, along with the decrease in phantom activity, the  . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 10, 2022. The estimated LSF for a true LSF of 0% was greatly overestimated. Mean LSF simulated values 319 from the first seven measurements were: 10.9%, 10.3%, and 9.9% for energy windows of 320 W1, W2, and W3, respectively (Fig 3B) 321 Planar imaging 322 Planar images greatly overestimated the LSF (up to 17.6pp for W1 and 15.9pp for W2). The 323 background correction reduced the LSF overestimation by approximately 5.7pp and 5.5pp 324 for the W1 and W2 energy window, respectively (Fig 4). 325 Foci visibility 326 The reconstructed images (both PET and SPECT) were used for qualitative and quantitative 327 analysis. The observers used both coronal and axial views to determine the visibility of the 328 foci. They could review single modality as well as fused images. On the other hand, the CNR 329 calculations were performed on the axial views of the phantom (Fig 5). We have also qualitatively analysed the visibility of the extrahepatic concentration (Fig 5). It 358 remained visible up to the 3 rd day after phantom filling, when activity in the whole phantom 359 was 747 MBq (in the extrahepatic deposition the activity was then 4.5 MBq). However, it 360 needs to be noted that the last dataset with discernible activity concentration outside of the 361 liver required a careful review of all available cross sections.
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 10, 2022. . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 10, 2022. For PET/CT images we are able to visually distinguish the organs of interest (liver and lungs) 384 for activities in the phantom over 200 MBq. For lower activities the PET reconstructions are 385 very noisy leading to poor discernibility of the liver, while the lungs cannot be identified at 386 all (Fig 1). On the other hand, in SPECT images (both in MIP and cross-sectional planes) the 387 lung shunt is not visible, even for high activities (Figs 1 and 5). The liver is easily identifiable 388 on almost all scans. However, the spill-over effect of activity outside of the liver cannot be 389 ignored, as it adversely affects the possibility of accurate estimation of activity in the 390 background and neighbouring organs. This effect is also connected to lower spatial 391 resolution than in PET technique. Similar spill-over effect could be observed in the planar 392 images. In those images the liver was as distinguishable as in SPECT, however the lungs were 393 identifiable in images obtained with high activities. It has to be noted, that the visibility of 394 lungs was worse than in PET reconstructions. Both SPECT and planar imaging was strongly 395 influenced by scattered radiation, due to registration of photons in wide energy windows. 396 Neither of those techniques included corrections for this effect, which adversely affects the 397 quality of those images. These qualitative findings are in good agreement with results of 398 study by Kunnen et al. (8), even though they used a higher LSF value of 15%. 399 We chose the lung shunt value of 10% because for resin microspheres (SIR-Sphere) it is the 400 experts' recommended lung shunt threshold beyond which a reduction in isotope activity is 401 strongly advised (1,17,18). Therefore, the LSF of about 10% needs to be detectable and 402 accurately estimated. 403 The PET data yielded the most accurate LSF estimation as presented in Fig 3A. In our study 404 the absolute differences of calculated and true LSF values for total activities in the phantom 405 over 200 MBq were less than 1.5pp. In comparison, Kunnen et al. report differences of less . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 10, 2022.  (Fig 3B). This suggests that the obtained results are much more reliable than the values 422 calculated from SPECT data. 423 In planar imaging we observed a gross overestimation of calculated LSF (Fig 4). The 424 background correction performed according to EANM recommendations (3), allowed for its 425 partial reduction. Nevertheless, the obtained results were definitively worse than those from 426 PET or SPECT. In part, it can be explained by the lack of scatter and attenuation correction, 427 as well as overlapping of different structures.
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 10, 2022. ; https://doi.org/10.1101/2022.07.07.22277361 doi: medRxiv preprint 21 428 As a continuation of our previous research we have analysed the visibility of hot and cold 429 foci in the liver. We have found that PET acquisitions provided better images for 430 distinguishing the tumours from the liver. The hot lesion was visible even for low activities, 431 while the cold lesion was reported to be visible both qualitatively and quantitatively only for 432 the highest activity of 90 Y (Fig 5A). 433 Previously we have demonstrated that the smallest cold foci visible in our SPECT acquisitions 434 of the Jaszczak phantom had a diameter of 25.4 mm (11). Since we wanted to analyse 435 tumours' visibility at higher concentrations of 90 Y activity in surrounding area closer to 436 clinical settings, we have chosen to use a slightly smaller sphere (diameter of 22 mm). The 437 cold tumour was not visible in any of the acquired SPECT scans (Fig 5B). This might be 438 explained by poor spatial resolution and spill-over activity from the liver. The acquired CNR 439 values are consistent with the lesion's discernibility and our previous findings. 440 We have found good correlation between CNR and hot tumour visibility for energy window 441 W1 and W2, with higher values calculated for the latter.  CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted July 10, 2022. consuming and are not easily accessible for many smaller nuclear medicine centres. 453 Our study suggests that PET/CT provides better images for high activity posttreatment 454 imaging of patients undergoing therapy with 90 Y microspheres than Bremsstrahlung based 455 SPECT/CT. Lowering the activity to levels acceptable in pretreatment scans (about 100 MBq), 456 poses a great challenge in imaging. Its main goal is to reliably assess the LSF. Our work 457 demonstrates that PET/CT yields the best results among the tested methods. However, it 458 seems that it may have some inherent limitations in imaging (low counts in the image), 459 which explains the difficulties with image interpretation and analysis at the lowest activities. 460 SPECT/CT could be considered for this purpose, if suitable corrections can be applied. As for 461 planar imaging it did not prove to provide enough information for it to be clinically useful for 462 pretreatment 90 Y scans. 463