Clinical Impact of 99m Tc-MAA SPECT/CT-based Personalized Predictive Dosimetry in Selective Internal Radiotherapy: a Retrospective, Single-center Study for Unresectable HCC Patients

Recent data indicates that personalized dosimetry-based selective internal radiotherapy (SIRT) be We to evaluate the contribution of personalized predictive dosimetry (performed with Simplicity90® software) in HCC patients by comparing them to our historical cohort whose activity was determined by standard dosimetry.


Introduction
Hepatocellular carcinoma (HCC) accounts for 90% of all primary liver cancers, with the majority of cases being associated to cirrhosis 1 . Therefore choosing the most suitable treatment option depends not only on the tumor stage, but also on the severity of the underlying liver disease. Current guidelines 1,2 consider the Barcelona Clinic Liver Cancer (BCLC) staging system as the algorithm of choice for tumor staging and therapeutic options 3 , taking into account tumor burden, Child-Pugh classi cation and performance status. Locally advanced HCC is suitable for transarterial locoregional therapies, conducted mostly in a palliative setting.
Selective Internal RadioTherapy (SIRT), known as TransArterial RadioEmbolization (TARE) in the context of liver disease, consists here of the intra-arterial infusion of small beads that are loaded with a radioactive isotope, generally Yttrium-90 ( 90 Y) or more recently Holmium-166 ( 166 Ho), and it relies on the beta radiation emitted by the isotope to induce tumor necrosis, with a minor contribution from microembolization 4 . Its e cacy was reported in several large patient series [5][6][7][8] . Two products are commercially available for 90 Y-based SIRT. 90 Y-labelled resin microspheres (SIR-Spheres; Sirtex Medical, Sydney) and 90 Y glass microspheres (TheraSphere®; Boston Scienti c, Marlborough, MA, USA) are approved for the treatment of unresectable liver tumors, including HCC. Currently, new 166 Ho-PLLA microspheres (QuiremSpheres®), aiming to improve work-up (or simulation) by the use of 166 Ho-PLLA "scout" activity, have been developed and showed safety and e cacy in unresectable, chemorefractory liver metastases 9,10 .
Nevertheless, for advanced tumors or chemoembolization refractory HCC, two RCT phase III, randomized controlled trials, comparing SIRT to Sorafenib, showed no difference in terms of overall survival (OS) 11,12 . Moreover, the addition of SIRT to Sorafenib, evaluated in a recent phase II study, did not show bene t in terms of survival 13 .
Usually, the prescribed activity in these trials was based on the body surface area (BSA). This represents a kind of standard calculation method for the activity of 90 Y to be administered and is assumed to correlate with the patient's liver volume, adjusted by the percentage of tumor involvement and the magnitude of the lung-shunt fraction (i.e. the fraction of injected microspheres lodged within the precapillary of the lungs) 14,15 .
Nevertheless, the BSA method may lead to underdosing (e.g., large livers) due to its moderate correlation to the liver volume and the percentage of tumor involvement adds little value when adjusting the activity to administer.
In order to optimize the administered activity, the concept of a more personalized dosimetry has been advocated, making use of patient-speci c parameters and a multi-compartmental modeling that estimates the dose to the tumor, based on the Medical Internal Radiation Dose (MIRD) model. The capacity to predict the distribution of 90 Y-microspheres has been shown to be a factor of improvement of SIRT e cacy [16][17][18] . The similar distribution of 99m Tc-macroaggregated albumin ( 99m Tc-MAA) and 90 Y-microspheres allows dosimetry simulation using 99m Tc-MAA single-photon emission computed tomography ( 99m Tc-MAA SPECT /CT) co-registered with MRI, or contrast-enhanced CT (CE-CT).
In our institution glass microspheres (TheraSphere®, Boston Scienti c) have been used since 2016 at rst using a standard dosimetry-based simulation, and, more recently, with a personalized dosimetry to prescribe a more accurate activity (i.e.: calculated using Simplicit 90 Y TM , a software able to do multicompartmental MIRD dosimetry). According to recent publications, a (mean) absorbed dose of, at least, 205 Gy delivered to the lesion is required to achieve an optimal response without exceeding 70 to 75 Gy to the non-tumoral liver 19,20 .
Hereby, we report our data regarding the use and potential bene ts of personalized dosimetry for SIRT in our tertiary Hospital, by retrospectively comparing the response rate of SIRT obtained using standard dosimetry-based simulation to this obtained with SIRT using personalized dosimetry-based simulation. In addition, in patients treated after standard-dosimetry-based simulation, we compared the activity administered and the activity that would have been administered if personalized dosimetry had been applied. Finally, we analyzed the safety of SIRT in these two groups as well as the impact of personalized dosimetry had on survival.
Material And Methods

a. Study design and patient allocation
This is a retrospective study conducted in a tertiary health-center in Belgium, from February 2016 to December 2020, that enrolled 66 consecutive HCC patients who underwent at least one work-up (based on 99m Tc-MAA scintigraphy) for radioembolization and received or not the treatment by SIRT (by Y90loaded glass microspheres). Patients were divided in two consecutive groups. Twenty-nine patients underwent work-up with standard dosimetry-based simulation (group A) from February 2016 to November 2017 and 37 patients underwent work-up with personalized dosimetry-based simulation (group B) from December 2017 to last enrolment at the end of 2020 ( Figure 1).  21 , de ned by the proportion of treated nodules that presented complete or partial response, as well as the best overall response (BOR) de ned as the best recorded response per patient from the start of the study treatment until disease progression between the two groups of patients (i.e., group A treated by 90 Y with an activity calculated using standard predictive dosimetry and group B using personalized predictive dosimetry).
The secondary objectives were: (1) the comparison of the two treatment groups in terms of progression free survival (PFS), de ned as the time from treatment to the rst observation of progressive disease or death, and overall survival (OS), de ned as the time from treatment to death of any cause; (2) safety and toxicity pro les in the two groups evaluated according to the Common terminology Criteria of Adverse Events Version 5.0 (CTCAE V5.0) 22 ; safety was evaluated clinically and biologically at 24 hours, 1 and 3months and radiological adverse events (AEs) were recorded at 3 months after treatment; (3) for group A: (i) dose-response link investigation using mRECIST criteria on target lesions at 3 months, (ii) comparison between the activity to be administered by SIRT determined using a personalized dosimetry software with multicompartment MIRD technique (Simplicit 90 Y®) to the activity administered determined by the classical non-compartmental dosimetry planning, using a target volume based on the 99m Tc-MAA SPECT only (for patients treated before the acquisition of Simplicit 90 Y®); (4) for group B, dose-response link investigation using mRECIST (target) criteria at 3 months.

d. Study procedures, activity calculation and dosimetry
The radioembolization procedure was performed over two different sessions: work-up session and treatment session, by the same two interventional radiologists with 5-10 years' experience (between 50-100 procedures per year). This work-up evaluation started with an angiography in order to obtain a precise map of the patients' abdominal vascular anatomy and coil embolization was performed if gastrointestinal branches arising from the hepatic arteries were found.
Patients in group A received treatment with 90 Y activity calculation based on the classical noncompartmental MIRD dosimetry planning following standard guidelines, using a volume based on the 99m Tc-MAA SPECT/CT. 148 to 185 MBq of 99m Tc-MAA were injected after diagnostic angiography in the hepatic artery. Images were performed on a Philips BrightView XCT gamma-camera with a Low Energy High Resolution (LEHR) collimator. First, a whole-body scan was acquired to determine lung shunting (140 ± 14 keV, 18 cm/min, 256 pixels wide). Then, a liver-centered SPECT/CT was conducted to estimate the spatial distribution of the 99m Tc-MAA (140 ± 14 keV, 32 projections, 30 s/projection, 360 °, 128 x 128 pixels). The iterative reconstruction method commercially available used is Astonish® (3 iterations, 8 subsets). A calculation sheet from the 90 Y provider gave us the activity knowing the volume segmented from the 99m Tc-MAA SPECT data (with thresholding based on a maximum intensity percentage), the lung shunt fraction determined on the whole-body scan (with manual segmentation), and the desired dose to the targeted volume (based on standard guidelines). 99m Tc-MAA lung shunt fraction did not exceed 30 Gy in a single treatment or 50 Gy in case of multiple treatments. In case on an unfavorable 99m Tc-MAA workup, the procedure was repeated and a solution was searched for (e.g., more selective placement of the catheter during injection to improve the targeting of the lesion). If 90 Y-based SIRT could not be performed, the patient was treated according to best medical practice. If the work-up had a favorable outcome, the patients were re-admitted for treatment within 15 days. 90 Y Bremsstrahlung Emission Computed Tomography ( 90 Y BECT/CT) post-treatment images used for dosimetry purposes were acquired on the same Philips BrightView XCT gamma-camera. 0,42 to 5,1 GBq were injected regarding this standard predictive dosimetry group (A). An energy window around 120 keV ± 24 keV was chosen to avoid the lead uorescence X-rays around 80 keV and more energetic photons, eventually passing through the collimator because Medium Energy General Purpose (MEGP) collimator was used. First, a whole-body scan was acquired to visually con rm the absence of lung shunting (120 kev ± 24 keV, 12 cm/min, 256 pixels wide). Then, a liver-centred BECT/CT was conducted to estimate the spatial distribution of 90 Y-TheraSphere® (120 kev ± 24 keV, 64 projections, 30 s/projection, 360 °, 64 x 64 pixels). The same commercially available algorithm as previously described was used to reconstruct the data.
In order to compare the received activity according to standard predictive dosimetry and the activity that would have been recommended by the Simplicit 90 Y® software, patients in group A for which we disposed of imaging at 3 months after treatment, were included. For all of them, CE-MRI, or CECT, 99m Tc-MAA SPECT/CT and 90 Y BECT/CT were available. Radiological tumor response, evaluated following the mRECIST target criteria at 3 months, was correlated with the perfused tumor dose and the perfused fraction of the total tumor volume, determined using 90 Y BECT/CT and Simplicit 90 Y®. We also calculated the activity needed to reach 205 Gy in the perfused tumor, using 99m Tc-MAA SPECT/CT and Simplicit 90 Y®. The activity to be administered based on this minimal tumor absorbed dose criteria was compared to the activity actually administered. We plotted the relative differences, and correlated these differences with the radiological tumor response.
Patients in group B received treatment after personalized predictive dosimetry that was performed using Simplicit 90 Y® software. MRI or CECT were used for the segmentation of the liver, the tumor, and the nontumoral liver (manually on MRI, automatically on CE-CT with corrections when needed). Then, 99m Tc-MAA SPECT/CT was co-registered (rigid or non-rigid coregistration when needed) and the perfused volume determined (with thresholding based on a maximum intensity percentage), after lung shunt fraction evaluation on the whole-body scan. Acquisition parameters are the same as described above.
Regarding post-treatments imaging, in group B, 90 Y PET/CT were available for 9 patients. 90 Y PET/CT post-treatment images used here for dosimetry purposes were acquired on a digital Philips Vereos PET/CT scanner (20 minutes per bed position for a total of 40 minutes or 2 bed positions, 288 x 288 pixels of 2 x 2 mm with a slice thickness of 2 mm). The iterative reconstruction algorithm is an Ordered Subset Expectation Maximization (OSEM, 3 iterations, 17 subsets, with Point Spread Function option applied). Personalized dosimetry was performed using Simplicit 90 Y®. MRI or CECT were used to do the segmentation of the liver, the tumor, and the non-tumoral liver (manually on MRI, automatically on CECT with corrections when needed). Then, 90 Y PET/CT was co-registered (rigid or non-rigid coregistration when needed) and the perfused volume determined (with thresholding based on a maximum intensity percentage). Lung shunt fraction was evaluated on a Bremsstrahlung Emission Whole Body scan as described above.
In this group B, the radiological tumour response, evaluated following the mRECIST target criteria at 3 months, was correlated with the perfused tumour dose and the perfused fraction of the total tumour In group B, a total of 47 work-ups were performed, with several patients undergoing more than one evaluation. Thirty failures were registered, mainly due to poor tumor targeting. Only 16 patients received treatment (43.24%), with also one patient undergoing two radioembolizations on the same tumor.
The mains reasons for work-up failure are noted in the Flowchart (Figure 1).
Baseline characteristics of patients in groups A and B receiving SIRT are listed in Table 1. Demographics, tumor burden and liver function were similar for the two groups. ) was observed in 31.58% in group A and 12.5% in group B. Nevertheless, 1 patient that presented partial response was shown to have hepatic progression in a non-targeted area. The same situation was showed at 6 months evaluation (Table 2A). Table 2. Overall response rate on target areas (A) and best overall response (B -detailed response and C -CR+PR and SD+PD) for the standard dosimetry group at 3 and 6 months, according to mRECIST.
CR-complete response, PR-partial response, SD-stable disease, PD -progressive disease according to mRECIST DCR is de ned as the composite of ORR and stable disease between patients in group A and B.

c. Follow-up and survival
The median time of follow-up was the same in the two groups, 21 months (range 3-55) in group A and 21 months (range 4-39) for the ones in group B.
The Kaplan-Meier curves of OS and PFS are depicted in Figure 2. When compared, the two groups showed no statistical difference in terms of survival (p=0.17). The same trend was noticed for PFS with a median time to progression in group A of 6.14 months (95% CI: 3.94 -9.23) and 6.34 months (95% CI: 2.92 -21.52) in group B.

d. Safety and complications
In the safety analysis, no mortality was registered at 30 days from treatment.
In group A, 6 out of 20 patients (30%) evaluated at 1 months presented grade 1 or 2 clinical AEs (fatigue 33.33%, abdominal pain 50% and nausea/vomiting 33.33%). At three months evaluation, one patient also presented gastrointestinal bleeding due to newly appeared angiomas due to portal hypertension.
In group B, 7 patients out of 15 (46.67%) that were evaluated at one month presented grade 1 or 2 clinical AEs (nausea 13.33%, fatigue 13.33%, abdominal pain 13.33%, ascites 6.67%, hemorrhagic duodenal ulcer 6.67%). The patient that received two treatments presented a hemorrhagic duodenal ulcer one month after his second treatment. At 3 and 6 months only 3 and 1 patients respectively, still presented clinical toxicities. The latter had grade 3 ascites that required repeated paracentesis.
The most common biochemical toxicity at 1 month in group A was hyperbilirubinemia (20%). This persisted at 3 months, with one patient presenting Grade 3 toxicity.
However, the most common biochemical toxicity for group B was decrease albumin at 1 month BECT/CT were available. Among these 19 treatments evaluated by mRECIST (target) criteria at 3 months, 11 treatments induced a PR and 2 a CR. In these responders, the 90 Y BECT/CT-based dosimetry showed 7 patients receiving less than 205 Gy and 6 patients receiving more than 205 Gy in the perfused tumour (188 Gy on average) (Figure 3). For these latter 6 patients, the comparison between the administered activity and the recommended activity estimated retrospectively using Simplicit 90 Y® showed that 5 patients received an activity that was higher than recommended (79% mean difference) (Figure 4), but, with a dose to the non-tumoral liver staying below 70 Gy except for 1 patient (44 Gy on average) 20 .
Among the 7 responding patients receiving less than 205 Gy in the perfused tumor, 2 patients showed visible areas of necrosis before treatment on the anatomical images used.
Among the 6 treatments inducing no response, 3 patients progressed and 3 patients were stable. All of these patients received less than 205 Gy to the perfused tumor, based on a 90 Y BECT/CT-based dosimetry, except one ( Figure 3). In the latter patient, microspheres were concentrated non-homogenously with a well-de ned hot spot and undertreated parts ( Figure 5). The comparison between the administered activity and the recommended activity calculated using Simplicit 90 Y® showed that the other 2 patients who progressed received less activity than recommended (51% mean difference). The results are the same for 2 out of 3 patients who were stable (47% mean difference for these 2 out of 3) (Figure 4).
Ten patients of the group B underwent post treatment 90 Y PET/CT, thus allowing dose-response analysis to be conducted. At the 3 months imaging evaluation 8 treatments induced a response (5 PR and 3 CR). Among the 2 treatments inducing no response, 1 patient progressed and 1 patient was stable ( Figure 6).
In the group A (with BECT/CT-based dosimetry), a threshold of 175 Gy appears to be the cut-off value to reach response, at least partially (Figure 3), except for 1 patient who progressed with a well, but nonhomogenously perfused tumor ( Figure 5). In the group B (with PET-CT based dosimetry), the perfused fraction of the total tumor volume seems to play a role too, because we encountered a patient in progression after 3 months (regarding the targeted tumor following mRECIST criteria) in whom the covering fraction was 80%. Reaching more than 90% for the covering fraction and 400 Gy for the entire tumor dose ensure response with an important proportion of total responses in this group (60%) ( Figure   6).

Discussion
This retrospective trial compared the objective response rate for treated nodules as well as the best overall response in our population of HCC patients that underwent treatment with radioembolization using standard predictive dosimetry versus personalized predictive dosimetry. For patients treated after standard-dosimetry-based simulation, the activity that would have been administered by SIRT determined using a personalized dosimetry software with multicompartment MIRD technique (Simplicit 90 Y®) was compared to the activity actually administered determined by the classical non-compartmental dosimetry planning, using a target volume based on the 99m Tc-MAA SPECT (patients treated before the acquisition of Simplicit 90 Y®). To the best of our knowledge, this is the rst study that addressed the potential changes due to personalized predictive dosimetry in treatment administration in a retrospective cohort.
In our study, there was no signi cant statistical difference when comparing the two groups in terms of response rate per treated nodule (ORR), despite a trend in favor of personalized dosimetry (87.5% for personalized dosimetry versus 68.4% for standard dosimetry at 3 months, p= 0.24). These discordant results are probably due to the small size of our two groups population. By contrast, in the per patient analysis, there was a clear bene t of personalized dosimetry in terms of BOR at 3 months (80% versus 33.3%, p= 0.007) and 6 months post-treatment (77.8% versus 22.2%, p= 0.06).
The response rate we report is slightly higher than the results published recently in the randomized, multicenter DOSISPHERE-01 study 23  In group A, the comparison between the administered activity and the recommended activity calculated using a personalized dosimetry software with multicompartmental (MIRD) approach showed that the patients who progressed received less activity than recommended ( Figure 4) or that this activity was not adequately distributed ( Figure 5). Our results indicate that reaching a minimal absorbed dose criteria increases the e cacy of SIRT. They support the use of a more personalized predictive dosimetry instead of the classical non-compartmental MIRD dosimetry for treatment planning. This is in line with the notion that a minimal absorbed dose of a minimum 205 Gy to the lesion is required to achieve an optimal response for SIRT with glass microspheres 19,20 .
Note that taking into account areas of necrosis in the tumor absorbed dose calculation does not provide an estimate of the dose to the viable tumor, as mentioned by Garin et al.
Dose-response link evaluations, in both groups, shows clearly that a total response cannot be reached if the perfused fraction of the total tumor volume is not enough and about 100%. Giving enough dose to a su cient tumor volume fraction ensures at least a partial response.
Our study has several limitations, the most important being the small number of patients and its retrospective nature. Furthermore, a bias is to be taken into account due to the fact that some patients had already undergone previous treatments on the treated tumors. Moreover, the groups have been treated in two successive periods. Therefore, the bene t of the experience acquired may have in uenced our results for the patients in group B. This effect is mitigated by the fact that the most sensitive part of the protocol that is subjected to a learning curve, that is the angiographic procedure has been performed by experienced radiologists (between 5-10 years' experience of hepatic angiography practice).

Conclusion
Our data contributed to display that multi-compartmental dosimetry, more personalized dosimetry, plays a key role to estimate in advance the e cacy of SIRT during simulations. A higher BOR can be reached by (more) personalized dosimetry over standard one and this difference may explain disappointing data previously reported, indicating that a (more) personalized dosimetry is needed for SIRT planning in the treatment of HCC. Figure 1 Flowchart of the study (Group A-patients that received SIRT after standard-based-dosimetry simulation;

Figures
Group B-patients that received SIRT after personalized-based-dosimetry simulation). Box plot showing the distribution of activity relative differences (between activity to be administered following personalized dosimetry and activity actually administered following classical standard approach) in 2 subgroups of group A (complete or partial response evaluated at 3 months following mRECIST target criteria with perfused tumor dose above 205 Gy versus stable or progression disease with perfused tumor dose under 205 Gy).
Page 25/26 Figure 5 Isodose curves from 90Y BECT on MRI images (T2 sequence) in a 61-year-old male patient suffering from an HCC well perfused but non-homogeneously with undertreated parts (under 205 Gy) Figure 6 Dot plot of perfused fraction of the total tumor volume according to perfused tumor adsorbed dose and tumor response evaluated following mRECIST (target) criteria at 3 months, in patients from group B with PET-CT-based dosimetry