Response to Trans-arterial Radioembolization Therapy for Liver Cancer According to Voxel-Based Dosimetry: A Retrospective Study

Usual clinical dosimetry models for trans-arterial radioembolization (TARE) is based on assumption of uniform dose distribution in each tissue compartment. We performed a simple voxel-based dosimetry using post-treatment yttrium-90 ( 90 Y) microsphere PET in TARE and investigated the prognostic value of dosimetry parameters from the voxel-based dosimetry. Twenty-eight patients with hepatocellular carcinoma who underwent TARE using 90 Y-microsphere were retrospectively included. Mean absorbed dose of each lesion (TDv) was analyzed using voxel-based dose maps derived from posttreatment 90 Y-microsphere PET and voxel-wise S-value kernels. Heterogeneity of intra-tumoral absorbed dose was investigated using standard deviation and coecient of variation of voxel doses in a tumor. The response of each lesion was classied as local control success (LCS) and local control failure (LCF) based on follow-up MRI or CT. Prognostic values of dosimetry parameters and clinicopathologic factors were evaluated using survival analysis for progression-free survival.


Background
Hepatocellular carcinoma (HCC) is a highly lethal malignancy prevalent worldwide [1]. Among the treatment options for HCC, complete resection or liver transplantation are the only curative treatments.
Trans-arterial radioembolization (TARE) is currently used as a bridging treatment or a local control method for patients with intermediate-risk or advanced HCC who are not eligible for curative treatments [2,3]. TARE has shown better survival outcomes than other treatment options, such as trans-arterial chemoembolization (TACE) and systemic chemotherapy including sorafenib [4]. However, local control failure (LCF) is observed in approximately 20-50% of cases [4,5], probably because of insu cient radiation dose to tumors.
Usual dosimetry models for TARE estimate the mean absorbed radiation doses of tumors, normal liver tissues supplied by tumor-feeding artery (in-target normal liver), and normal liver tissues supplied by nontumor-feeding arteries (out-target normal liver) assuming a uniform dose distribution in each tissue compartment [6,7]. The radiation dose delivered by TARE is determined using a partition model of dosimetry based on pretreatment technetium-99m ( 99m Tc) macroaggregated albumin (MAA) scan or simply based on tumor volume. However, there is a considerable intra-tumoral heterogeneity in the actual dose distributions depending on vessel density, thrombosis, and tumor necrosis. Hence, intra-tumoral dose heterogeneity is considered an underlying cause of LCF [8].
Voxel-based dosimetry was developed as a method to overcome the issue of intra-tumoral dose heterogeneity. Dieudonné et al. reported the feasibility of voxel-based three-dimensional (3D) dose mapping based on 99m Tc-MAA single-photon emission computed tomography (SPECT)/computed tomography (CT) performed for pretreatment planning of yttrium-90 ( 90 Y) microsphere TARE [9]. Kafrouni et al. demonstrated the correlation between treatment response and mean absorbed radiation dose derived from voxel-based dosimetry based on post-treatment 90 Y-microsphere positron emission tomography (PET) [10].
In this study, voxel-based dosimetry was performed using posttreatment 90 Y-microsphere PET in those who underwent TARE for HCC. This study aimed to investigate the prognostic value of radiation dose parameters obtained from voxel-based dosimetry and other clinical factors in predicting the treatment outcomes of TARE.

Patients
This study retrospectively included patients who underwent TARE from July 2012 to September 2014 based on the following inclusion criteria: (1) con rmed HCC diagnosis, (2) presence of post-TARE 90 Ymicrosphere PET data, and (3) presence of follow-up records and imaging studies for evaluating treatment response. The clinicopathologic information of each patient was obtained by reviewing their electronic medical records. The Institutional Review Board of Seoul National University Hospital approved the study design and waived the requirement for informed consent (H-2009-114-1158).
TARE and PET Image Acquisition TARE was performed as previously described [11]. During planning angiography, 185 MBq of 99m Tc-MAA was injected into the tumor-feeding artery. Subsequently, planar scan and SPECT/CT were performed to calculate the liver-to-lung shunt fraction (LSF) and dose absorbed by normal organs and tumors. TARE was performed when the LSF on the planar scan was < 20%. The dose of injected radioactivity was determined using partition model-based dosimetry with 99m Tc-MAA SPECT/CT [6], so that estimated tumor dose is > 120 Gy. Radioactivity was adjusted, if necessary, according to the patient's condition and the operator's decision. TARE was performed by a single experienced interventional radiologist (H.C.K.) using 90 Y-labeled resin microspheres (SIR-Spheres®; Sirtex Medical Ltd., Lane Cove, Australia). 90 Y-microsphere PET/CT was performed using a large eld-of-view PET/CT scanner (Biograph mCT64, Siemens Healthineers, Germany; axial eld-of-view 216 mm) immediately after TARE. CT images were acquired rst (slice thickness, 5 mm; pitch, 1.2, 120 kVp; and 35 mAs) for attenuation correction and lesion localization. PET images were acquired using a 3D mode for one bed position to cover the lower chest and upper abdomen for 10 min. The images were reconstructed on 200 ⋅ 200 matrices using an iterative method (2 iterations, 21 subsets) with an algorithms for point-spread function recovery and timeof-ight estimation.

Voxel-based Dosimetry
The voxel S-value (VSV) kernel [Gy/MBq•sec] convolution approach was used, where VSV is a voxel-level Medical Internal Radiation Dose (MIRD) schema, de ned as the mean absorbed dose to a target voxel per radioactive decay in a source voxel [12]. The voxel dose (D) was calculated by convolving the cumulated activity (Ã) with VSV using the equation below.
Voxel-based dosimetry was conducted using VSV the kernel reported in a public database [13]. The 6 ⋅ 6 ⋅ 6 VSV kernel of 90 Y was used with the voxel size of 3 ⋅ 3 ⋅ 3 mm 3 . The PET images were re-sized such that their voxel size is same as that of the VSV kernel to avoid false dose calculation. Accumulated activity maps were generated from the re-sized PET images by calculating the time-integrated activity from the injected time point to in nity. The 3D cumulated activity maps [MBq•sec] were convoluted with the 3D VSV [Gy/MBq•sec] to yield absorbed dose maps [Gy] using MATLAB software.
These absorbed dose maps were fused with enhanced magnetic resonance (MR) or CT images obtained immediately before TARE, using a vendor-supplied software package (Syngo.via, Siemens Healthineers, Germany). The tumor regions-of-interest were drawn manually based on enhanced MR or CT images on every image slice, and the tumor volume-of-interest (VOI) was obtained by stacking all tumor regions. From the VOI on voxel-based dose map, the mean absorbed tumor dose (TDv), standard deviation (SD), and coe cient of variation (CV) of absorbed doses were measured for each tumor. Tumors smaller than 1 cm were excluded because of possible partial-volume effects. A representative image of a tumor VOI is shown in Fig. 1.

Response Evaluation and Statistical Analysis
Tumor response was determined during follow-up imaging studies, including enhanced MR imaging (MRI) and/or CT. The local response of each lesion was determined according to the modi ed Response Evaluation Criteria in Solid Tumors (mRECIST) criteria based on the European Association for the Study of the Liver (EASL) guidelines [14][15][16]. Complete remission was de ned as local control success (LCS), with other responses de ned as local control failure (LCF). Progression was analyzed per-lesion and progression-free survival (PFS) was de ned as the time from the date of TARE to the date of the imaging study that showed local control failure or progression.
Values are expressed as means ± SDs. The median values of the longest diameter and TDv of each tumor were used as the cutoff values for survival analysis. For laboratory data, the upper normal limit of each parameter was used as the cutoff value. Univariate and multivariate survival analyses were performed using the Kaplan-Meier method and Cox regression analysis, respectively. The hazard ratio (HRs) and 95% con dence intervals (CIs) were also calculated. Group comparisons of values were performed using Mann-Whitney U tests. SD and CV were used as heterogeneity indices in the VOI. Correlations between factors were assessed using Spearman's correlation analysis. P-values less than 0.05 were considered statistically signi cant.

Patients and Treatment Responses
During the study period, 28 patients underwent TARE in our institution. Among them, 13 had HCC, and 10 (all men; age, 59 ± 10 years; range, 48-84 years) had available data and were nally included in the analysis. The patient characteristics are summarized in Table 1  Dose maps were successfully generated using the VSV kernel convolution (Fig. 1) 2B, P = 0.021). There were no signi cant difference in the heterogeneity indices of absorbed doses between the two groups (SD: P = 0.266; CV: P = 1.000) (Supplemental Fig. 1).

Voxel-based Dosimetry and Survival
Univariate survival analyses conducted by including tumor factors and clinicopathologic factors showed that high serum aspartate transaminase (AST) level (> 40 IU/L), large tumor size (> 66 mm), and low TDv (< 81 Gy) were signi cant prognostic factors for poor PFS (P = 0.019, 0.042, and 0.026, respectively; Table 2). The cutoff values for tumor size and TDv were 66 mm and 81 Gy, respectively. Kaplan-Meier survival curves of these factors are shown in Fig. 3. Multivariate analysis conducted by including these factors identi ed only TDv as an independent predictive factor for PFS (P = 0.022; HR, 21.018; 95% CI: 1.549-285.204). TDv and PFS showed signi cant correlations (r = 0.669, P = 0.009) and ve of six patients (83%) with TDv > 81 Gy showed PFS > 20 months (Fig. 4).  [17][18][19][20]. In TARE, radiopharmaceuticals are directly injected into the tumor-feeding arteries and do not show redistribution after initial embolization. Thus, calculation of its treatment dose is easier than that of other systemically administered radioactive drugs. Pretreatment planning angiography is currently performed using 99m Tc-MAA to simulate radiation doses to the tumor, normal liver tissues, and lungs.
In addition to pretreatment scans, many institutions perform a post-treatment scan to assess the results of TARE and identify possible complications occurring because of exposure of unexpected organs to radioactivity. Because 90 Y is a pure beta-emitter, the post-treatment scans are performed using highly sensitive scanners to acquire bremsstrahlung gamma images or PET images. The radiation dosimetry methods using post-treatment images are based on the MIRD schema. Several studies have applied the MIRD schema to bremsstrahlung imaging [21,22], and Gulec et al. reported the use of this schema for the dosimetry of 90 Y-microspheres con ned to the liver [23]. The usual MIRD schema assumes a uniform distribution of radioactivity in a certain organ or tissues and its use is very simple.
However, in the real world, therapeutic radiopharmaceuticals show heterogeneous distributions in-target tissues or organs because of vasculature heterogeneity, anatomical variation, and tissue necrosis. This study adopted voxel-based dosimetry, which considered intra-tumoral heterogeneity of 90 Y-microsphere distribution [24]. In this voxel-based dosimetry study, the VSV was calculated for each voxel to create an absorbed dose map. This is an easier approach than that involving patient-speci c Monte Carlo simulations and showed accurate dosimetry results in uniform-density organs such as the liver [24]. We applied a VSV from a 90 Y dataset published previously [13].
Based on the voxel-based dose map, the TDv of each lesion was successfully calculated. The considerable variation in TDv, from 40 Gy to 177 Gy, may be the main reason behind the difference in treatment outcomes. In survival analysis, small tumor size, low AST level, and low TDv (median, ≥81 Gy) were signi cant prognostic factors for successful treatment. The multivariate analysis con rmed TDv as the only independent prognostic factor. Previous studies reported clinical factors related to hepatic function (total bilirubin, albumin) and tumor aggressiveness (alpha-fetoprotein level, portal vein thrombosis, and tumor size) as signi cant prognostic factors in patients treated with TARE [25,26]. In our study, these factors were not signi cant, probably because of the small sample size. Further investigation in a larger cohort is needed to determine the role of these factors.
When applying a partition model of dosimetry to pretreatment simulation scans, 120 Gy is deemed a target dose for effective treatment. In our study, some lesions were outside this target range. Kao [28]. We observed an obvious difference in PFS between groups showing high and low absorbed doses. In Fig. 4, the cutoff of the TDv was set at 80-120 Gy. We also observed signi cant correlations between the absorbed dose and PFS in correlation analysis; this nding is consistent with that of a previous study [29].
Because this study performed voxel-based dosimetry, it was possible to calculate the voxel-wise intratumoral heterogeneity. High intra-tumoral heterogeneity was assumed to be related to LCF, despite the high average tumor dose. The SD and CV of dose distributions in a tumor were measured as simple indices for heterogeneity [30]. Although there was no signi cant difference in these values between the LCS and LCF groups in our study, further studies including more cases are warranted to investigate the effects of intra-tumoral dose distributions.
This study has several limitations. First, a small number of patients with inoperable HCC who underwent a single TARE session with 90 Y-resin microspheres were included in this study. Further studies with large cohorts are required to investigate the role of voxel-based dosimetry in clinical practice. Second, we used only simple indices for intra-tumoral heterogeneity. However, the results of this study may be used as preliminary data for determining the role of intra-tumoral heterogeneity of absorbed doses in treatment response.

Conclusions
In TARE using 90 Y-microsphere, the voxel-wise absorbed dose can be easily estimated using posttreatment 90 Y PET with a simple voxel-based dosimetry method conducted using VSV kernels. The TDv calculated by voxel-based dosimetry was a signi cant prognostic factor for the outcome of TARE in patients with HCC, with a cutoff value of 80-120 Gy. Further studies are required to determine the roles of tumor dose and intra-tumoral heterogeneity in treatment response.