Analysis of differences between ^99mTc-MAA SPECT- and ⁹⁰Y-microsphere PET-based dosimetry for hepatocellular carcinoma selective internal radiation therapy

Twenty-three unresectable HCC patients treated at our institution by SIRT using ⁹⁰Y glass microspheres from October 2015 to September 2018 were considered for this retrospective study. Among them, nine and four patients were included in DOSISPHERE and STOPHCC trials respectively [8, 9]. Authorization for the ancillary study was obtained from the principal investigators.

All the patients received the injection of microspheres in a single session except for one patient who underwent two sequential SIRT due to a reflux during injection at the first session. For this ^99mTc-MAA/⁹⁰Y-microsphere dosimetry comparison, these two sessions were considered as distinct procedures. All selected patients were at an intermediate or advanced stage of the disease. Baseline characteristics of the patients included are summarized in Table 1.

SIRT was applied following the general procedure described in the literature [10]. Prior to the treatment, an angiography was performed for hepatic vasculature mapping, potential coil embolization of extrahepatic vessels, and determination of optimal catheter position. This was followed by injection of a standardized activity of 185 MBq of ^99mTc-MAA and acquisition within an hour of planar images for assessment of lung shunting and SPECT/CT for visualization of potential extrahepatic microsphere deposition and tumor targeting and for activity planning. For patients included in trials, the activity to be delivered was planned according to the protocol instructions, i.e., for DOSISPHERE trial, delivering 120 ± 20 Gy to the treated liver or more than 205 Gy to the tumor (standard vs. optimized dosimetry arm [8]), and for STOPHCC trial, delivering 120 Gy ± 10% to the treated lobe of the liver [9]. For patients not included in any trials, activity was planned using personalized dosimetry with a 205 Gy minimum mean dose objective to the tumor [2] and a maximum mean dose of 50 Gy and 30 Gy to the NL and to the lungs respectively. Doses were assessed on the ^99mTc-MAA SPECT images using a voxel-based approach detailed in the dosimetry paragraph.⁹⁰Y glass microspheres (TheraSphere®, BTG Biocompatibles Ltd., Farnham, UK) were ordered through the form provided by the manufacturer including an estimated 2% residual activity as a preventive measure. ⁹⁰Y-microspheres were administered 18 ± 7 days after the simulation stage (range 12–37 days) according to the manufacturer’s instructions. Before injection, the ⁹⁰Y activity in the microsphere vial was systematically measured with the dose calibrator. Residual activity in the vial and the radiology material used was systematically assessed following the manufacturer recommendations [11], i.e., considering the ratio of the mean dose rate measured at four 90°-spaced points after and before injection. This ratio was then applied to the vial activity measured before injection to deduce the residual activity. The delivered activity, defined as the subtraction between the vial activity before injection and the residual activity, was 3.6 ± 1.2 GBq with a range of 0.9–6.6 GBq. ⁹⁰Y PET/CT images were acquired on the following day.

SPECT/CT data were acquired on a Symbia Intevo system (Siemens Healthcare, Erlangen, Germany) with the following parameters: window of 140 keV ± 7.5%, 32 projections per head, 25 s/projection, matrix 128 × 128, voxel size 4.79 mm × 4.79 mm × 4.79 mm, and low energy collimator. SPECT/CT data were reconstructed with Flash 3D Iterative Reconstruction was applied using 5 iterations/8 subsets, 6 mm Gaussian filter, with attenuation correction using a CT attenuation map, and scatter correction applying the Jaszczak method (dual-energy-window scatter correction, with a scatter window of 120 keV ± 7.5%, weighting factor of 0.5).

PET/CT data were obtained on a Biograph mCT flow (Siemens) with liver-centered continuous bed motion image acquisitions (bed speed of 0.2 mm/s). The PET/CT reconstruction parameters used for SIRT dosimetry were TrueX + time of flight reconstruction algorithm (Siemens), all-pass filter, 2 iterations, 21 subsets, and matrix 200 × 200 with voxel size 4.07 × 4.07 × 2.03 mm³. ⁹⁰Y PET data were corrected for attenuation and scatter using the single scatter simulation method. The PET scanner was calibrated by the manufacturer to measure ⁹⁰Y emission quantitatively. Moreover, this was verified beforehand applying the QUEST study by Willowson et al. using a NEMA 2007/IEC 2008 PET Body Phantom (Data Spectrum Corporation, NC) [12].

Predictive and post-treatment dose calculations were carried out based on the ^99mTc-MAA SPECT/CT and ⁹⁰Y-microsphere PET/CT images respectively using a dedicated software (PLANET® Dose, DOSIsoft SA, Cachan, France). The general workflow applied was similar to the one described in a previous study [13]. Briefly, tumor and NL were defined manually by an expert radiologist using prior morphologic imaging data (contrast-enhanced CT or magnetic resonance imaging). Only lesions larger than 2 cm located in the targeted lobe were considered in order to limit bias induced by partial-volume effect. The number of lesions was 1 for all treatments except for one treatment that concerned 2 lesions. ^99mTc-MAA SPECT/CT and ⁹⁰Y-microsphere PET/CT were rigidly co-registered with the imaging exam used for volume delineation. Thus, the same tumor and NL contours were used for both ^99mTc-MAA SPECT/CT- and ⁹⁰Y-microsphere PET/CT-based dose calculations. Three-dimensional dose maps at the voxel level were calculated for predictive and post-treatment dosimetry using a Voxel S-Values dose kernel convolution algorithm. Post-treatment dosimetry was performed using the activity concentration directly quantified on ⁹⁰Y PET data; no other calibration factor was applied.

On the one hand, the ⁹⁰Y planned activity was compared to the delivered activity to include all the clinical hazards: vial selection, actual time of injection vs. expected time, and residual activity. On the other hand, the activities measured in the whole field of view (FOV) and in the anatomically segmented liver on PET images were compared to the planned and delivered activity.

In order to assess dosimetric discrepancies related to differences in dose spatial distribution, ^99mTc-MAA SPECT was normalized so that the liver activity corresponds to that quantified inside the liver on ⁹⁰Y PET (designated as normalized ^99mTc-MAA SPECT).

For each treatment, data related to radiological gesture were compared by a single expert radiologist between simulation and treatment stages using patient records and angiographic images: operator, radiology material used, catheter position, distance to major bifurcation, volumes of injection, and potential vascularization modifications. The difference of catheter tip position was considered when a deviation > 5 mm was measured between simulation and therapy angiographic data. Distance to major bifurcation was estimated on dynamic planar (11/24) or CT (13/24) angiographic data when available and classified as follows: ≤ 5 mm, 10 ± 2 mm, 15 ± 2 mm, 20 ± 2 mm, > 22 mm considering the uncertainty of measurement.

The following metrics extracted from ⁹⁰Y-microsphere PET and normalized ^99mTc-MAA SPECT dosimetry were used for comparison: mean dose to the tumor (D_T) and to the NL (D_NL), as well as dose-volume histogram-based minimal dose to 70%, 50%, and 20% of the tumor volume (D₇₀, D₅₀, and D₂₀ respectively) and percentage of the tumor volume receiving at least 205 Gy (V₂₀₅).

Isodose volumes from normalized predictive and post-treatment dosimetry were compared using the Dice similarity coefficient [14]. Assessed isodoses corresponded to 50, 100, and 150 Gy referred as DC₅₀, DC₁₀₀, and DC₁₅₀ respectively.

Dose metrics based on ^99mTc-MAA SPECT and ⁹⁰Y-microsphere PET were compared using paired Student’s t tests. Pearson’s correlation coefficient (ρ), Bland-Altman analysis, and Lin’s concordance coefficient (ρ_c) were used to evaluate the agreement between predictive and post-treatment dosimetry. The normality of the data distributions was checked using the Kolmogorov-Smirnov test. Pearson’s correlation coefficient (ρ) was also used to evaluate the correlation between ⁹⁰Y PET activity recovery and patient’s BMI.

Predictive vs. post-treatment dose disparity was measured through the absolute difference for dose metrics (D_T, D_NL, D₂₀, D₅₀, D₇₀, and V₂₀₅) and isodose Dice similarity (DC₅₀, DC₁₀₀, DC₁₅₀). The following parameters were investigated as potential determinants of predictive vs. post-treatment dose disparity in univariate and multivariate analysis: age, body mass index (BMI), Child-Pugh class, BCLC stage, delay between simulation and treatment, type of tumor (infiltrative vs. nodular), portal vein thrombosis, tumor volume, liver volume, type of targeting (segmental, sectorial, lobar, or whole liver), lung shunt fraction, administered activity, difference between delivered and planned activity, and radiological gesture data (including operator identity, type of material, difference in catheter position, and distance from major bifurcation at treatment). Univariate analysis was performed by testing Pearson’s correlation between the dose disparity metric and the potential explanatory variable. Multivariate analysis was conducted using a forward-stepwise linear regression with an entry criterion of P ≤ 0.1 and a removal criterion of P > 0.05. Overall, a P value of 0.05 or less was considered significant.

⁹⁰Y delivered activity was lower than expected in 67% (16/24) of the cases. The difference between the planned activity and the one ordered selecting the closest vial was − 69 ± 133 MBq (− 2 ± 4%). Injection time was always later than expected except for two treatments. The delay between the expected and actual time of injection was 89 ± 55 min, resulting in a difference of activity of − 1.6 ± 1.0%. The residual activity measured after ⁹⁰Y-microsphere injection was 6 ± 7% including three cases with a substantial (> 10%) residual activity. The relative deviation between the planned activity and the delivered activity was − 148 ± 491 MBq (− 3 ± 9%).

The planned and delivered therapeutic activities, as well as the activities measured from PET acquisition in the whole FOV (covering mainly the liver) and inside the anatomic liver, are given in Table 2.

Difference between the delivered activity and the PET’s FOV activity was − 6 ± 11%. Besides, a substantial difference of − 20 ± 8% could be noticed between the activity measured in the segmented liver on the PET and the total activity in the PET’s FOV. This result was supported by the measurement made using an anthropomorphic phantom [15] (experiments not detailed here) where a deviation of − 15% between the activity in the liver of the phantom and in the total PET’s FOV was observed.

In addition to these results, the two types of differences (activity in the PET’s FOV vs. delivered activity and activity in the segmented liver vs. activity in the PET’s FOV) were correlated to the patient’s BMI (ρ = 0.41, P = 0.05 and ρ = − 0.55, P = 0.006 respectively) (Fig. 1).

After normalization, ^99mTc-MAA SPECT dosimetry was highly correlated and concordant with ⁹⁰Y-microsphere PET dosimetry for D_T (ρ = 0.87, ρ_c = 0.86, P = 3.10⁻⁸) and for D_NL (ρ = 0.91, ρ_c = 0.90, P = 7.10⁻¹⁰). Bland-Altman plot did not show any correlation between dose deviation and dose value (Fig. 2). No significant difference was found between ^99mTc-MAA SPECT and ⁹⁰Y-microsphere PET dose metrics except a minimal difference in D₇₀ (15 Gy, P = 0.04) (Table 3).

Details regarding radiological gesture are summarized in Table 4. In four scenarios, a 5-French catheter was used for simulation and a 4-French was used for therapy or vice versa, but the same microcatheter was used for both procedures. In two scenarios, both the catheter and the microcatheter differed between simulation and therapy. Catheter details were missing for one patient. No modification of tumor vascularization was observed between simulation and therapy for the patients included in this study. The injected volume was always different between simulation and therapy: 5 mL of ^99mTc-MAA solution vs. 60 mL of ⁹⁰Y-microspheres including rinsing.

Regarding tumor dosimetry, the sole parameter that was found to be significantly associated with D_T difference in both univariate and multivariate analyses was catheter position between simulation and therapy. Mean dose deviation was lower when the catheter position was identical than when it differed (16 Gy vs. 37 Gy, P = 0.007) (Fig. 3a). Besides, D_T and D₅₀ difference were negatively correlated with the catheter tip distance from major artery bifurcation (P = 0.03 and 0.01 respectively).

Dice coefficients calculated on the volumes extracted from 50 Gy and 100 Gy isodoses were significantly higher when catheter tip position was identical for simulation and therapy than when it differed (0.77 vs. 0.56; P = 0.001, and 0.68 vs. 0.53; P = 0.01 for DC₅₀ and DC₁₀₀ respectively) (Fig. 3b and c). Spatial concordance between predictive and post-treatment dosimetry also significantly correlated with the distance of the catheter position from artery bifurcation (P = 0.04, 0.0004, and 0.05, for DC₅₀, DC₁₀₀, and DC₁₅₀ respectively) (Fig. 4).