ScannerAll patients underwent a scan on Biograph Vision Quadra (Siemens Healthineers, Knoxville, TN, USA) PET/CT scanner which employs silicon photomultiplier-based detectors with 3.2 3.2 20 mm3 lutetium-oxoorthosilicate crystals [24]. Quadra is comprised of 32 detector rings, each with 38 detector blocks, which provide an AFOV of 106 cm. The data were acquired using the complete AFOV with a maximum ring distance of 322 crystals (MRD 322). The reconstruction algorithm using the data acquired with the full FOV (MRD 322) is called ultra-high sensitivity (UHS). The overall system sensitivity is 176 cps/kBq for the UHS, whereas the TOF is 230 ps [24]. The attenuation correction was based on the CT data. Randoms correction was applied employing the delayed event subtraction method. For the scatter correction a novel 3D correction algorithm was used, which estimates the full 3D scatter profile from the residual between measured and modeled data [25]. More specifically, the 2D measured data together with the 2D single scatter simulation (2D-SSS) model-based scatter provide the non-scattered true image estimate. In addition the patients were scanned also on Siemens Symbia Intevo Bold dual-head SPECT/CT (Siemens Healthineers, Knoxville, TN, USA) after the PET/CT scan. All necessary corrections were applied.
Patient Population The patient scans were performed between March 2021 and June 2024. 32 patients (median age, 71 years; interquartile range (IQR) 64-76) were included in this retrospective study. From them, 3 had a second treatment session, totaling in 35 therapies. Glass microspheres (TheraSphere®; Boston Scientific, Marlborough, MA, USA) were used for the transarterial radioembolization (TARE) [26]. The average time between the intervention and imaging scan was 2.5 ± 0.5 hours. There were 24 male and 8 female patients: 21 diagnosed with HCC, 6 with CCC, and 8 with hepatic metastasis from neuroendocrine tumors. The tumors were localized in the right lobe in 29 cases and in the left lobe in 4 cases. All patients had sufficiently large tumors (well over 1 cm). For 28 patients, there was one solid tumor on one location whereas for 5 patients there were at least 2 tumor sites treated.
Pre-treatment SPECT plan Forthe pre-treatment dosimetry plan, in accordance with the EANM guidelines [27], a 99mTc-MAA scan is utilized to visualize tumor sites in the liver based on percentage uptake. The LSF is calculated as the geometric mean of the counts measured in the whole-liver and the counts measured in the lung. As one of the most radiosensitive organs, the lung’s absorbed dose must be kept below 30 Gy (the single treatment dose limit) to avoid potential radio-pneumonitis. The MAA scan predicts the distribution of 90Y microspheres, though variations, particularly with resin microspheres, are a topic of ongoing discussion. Despite these differences, this method remains the current clinical standard of care.
The protocol for the technetium scan includes ventral and dorsal planar imaging, totaling 15 minutes. The quantitative SPECT scan lasts 20 minutes and is reconstructed with 8 iterations 4 subsets, using a 256 256 matrix. The CT for attenuation correction has the following parameters: 110 kV, modulated mA, a slice thickness 2 mm, and a 256 256 matrix. Both SPECT and CT images, along with the planar images, are loaded in Simplicit90YTM (Mirada Medical Ltd, Oxford, UK; Boston Scientific Corporation, Marlborough, MA, USA) incorporating segmentation of the liver and lung to calculate the LSF. The responsible physician performs segmentations of the entire liver (including right and left lobes), the perfused tumor, and the viable perfused tumor sites on the CT images.
Preparation and measurement of the 90Y microsphere activity The activity was measured using a well-type dose calibrator (ISOMED 2010). The calibrator satisfied the Swiss regulatory requirements and was calibrated by the Swiss Federal Institute of Metrology (METAS) [28]. The application took place in the presence of a radiologist/angiologist and a nuclear medicine physician. Once the intervention was completed, the patients were transported to the Nuclear Medicine Department for the PET/CT and SPECT scans as part of routine clinical practice.
Post-treatment LAFOV PET scan & reconstruction A helical CT scan was acquired for the PET attenuation correction with the following parameters: 80 kV tube voltage, 39 modulated mAs tube current, 38.4 mm total collimation width, 5 mm slice thickness and a pitch 0.8. The CT images were reconstructed using a 512 512 matrix, producing 644 slices with voxel dimensions of 1.523 1.523 1.600 mm3. Following the CT scan, PET acquisition was performed with a total acquisition time of 20 minutes, covering the body area from the head to the thighs. The scanner was calibrated for 90Y by the manufacturer, and a calibration factor was used to normalize the data, which were decay-corrected to the injection time. All patient data were reconstructed using the standard clinical protocol with 3D ordered subset expectation maximization (3D-OSEM), enabled TOF and point spread function (PSF) recovery. The scan time was 20 minutes. The images were reconstructed using 2 iterations and 5 subsets, with a gaussian filter of 2 mm full width at half maximum (FWHM), and a matrix size of 220 220 with UHS, based on the results of a phantom evaluation by another study [29]. The voxel dimensions for the PET images were 3.3 3.3 1.65 mm3. The images were also corrected for respiratory motion using Oncofreeze AI (Siemens Healthineers, Knoxville, TN, USA).
Post-treatment SPECT scan & reconstruction An HEGP collimator was used. The images were reconstructed with a matrix of 256 x 256, 3D OSEM, 8 iterations and 6 subsets together with a gaussian filtering of 6 mm. The total scan time was 30 minutes. The CT parameters for AC were the same as for the CT of the pre-treatment SPECT.
Dosimetry Analysis
The pre-treatment dosimetry plan was developed using Simplicit90YTM, a commercial dosimetry software (Mirada Medical Ltd, Oxford, UK; Boston Scientific Corporation, Marlborough, MA, USA) [30]. Simplicit90YTM employs a local deposition model for dose calculations (organ-wise dosimetry) [31]. The software’s multi-compartment algorithm estimates the relative absorbed dose for the perfused volume, tumor, and whole-liver normal tissue by measuring the counts within each segmented anatomical volume on the SPECT/CT image. The mean lung dose was calculated based on the geometric mean method, factoring in the LSF between the lungs and liver from anterior and posterior planar images. For post-treatment dosimetry validation, the PET images were loaded on both an organ-wise approach (Simplicit90YTM) and a voxel-wise approach software (HERMIA GOLD Smart Workstation 2.17, Hermes Medical Solutions AB, Stockholm, Sweden with the Voxel Dosimetry 1.1 toolbox) [32, 33]. The voxel-wise approach utilizes a voxel dosimetry approach, specifically the semi-Monte Carlo (sMC) algorithm, which requires quantitative PET images as input. The sMC algorithm considers the individual transport of electrons and photons. Electron energy is assumed to be absorbed locally, i.e. the same voxel where the decay happens, within the same voxel where the decay occurs, while photon energy deposition is calculated using a point-wise transport method. This dual consideration of electron and photon transport allows for efficient and accurate dose calculations. By utilizing these two software platforms with different approaches, we were able to independently evaluate their performance in dosimetry planning and validation. The comparisons are described in terms of predicted (SPECTpre) and actual (post-treatment SPECTpost and PETpost) absorbed doses. The whole-liver normal tissue absorbed dose and mean lung dose (based only on the voxel-wise approach) were also extracted. The D90, indicating the absorbed dose delivered to 90% of the tumor volume, was extracted from the DVH.
Lung shunt fraction & lung mean dose
The LMD could be measured using the HERMIA toolbox. Both lungs were manually segmented on the CT image corresponding to the PET and visually checked to eliminate any errors. As an inclusion criterion for the radioembolization treatment, the LMD should not exceed 30 Gy for a single treatment and/or 50 Gy in case of multiple treatments [27]. Following the consensus, the measured LMD was based on the activity in the left lung only, to achieve higher accuracy and reliability since it is less susceptible to mis-registration of liver counts inside the lung VOI as the right lung [34, 35]. The right lung was affected by scatter from the liver moving in the craniocaudal direction due to breathing motion. As literature has already shown, the LSF based on 99mTc-MAA is overestimated especially when planar imaging is used [34-37]. The LSFMAA based on 99mTc-MAA is a poor predictor the for actual lung shunting, in this work we evaluate also the LSF based on the 90Y PET/CT. LSFPET was measured on the PET images based on the methodology already proposed by Stella et. al. [34] and mathematically described as:
where, the mean activity concentration in left lung was computed as the average of the voxel values within the lung mask, lung volume is the segmented volume in mL, and the activityprescribed is the injected activity at radioembolization. The mean activity concentration was corrected for 90Y decay between application and scanning time.
It is necessary to note, that the maximum range of the betas emitted by 90Y in tissue in 12 mm, which is in the same order of magnitude as the PET resolution. Therefore, it is assumed that the total energy of the beta particle is deposited within the voxel of origin [38]. In addition, the distribution of 90Y is uniform in cases of lung shunting and last the lung density is considered to be the same for all patients.
Statistical analysis
The non-parametric Wilcoxon signed rank test for paired data was used because data normality could not be assumed due to the small size of the sample to assess the difference of statistical significance of the mean values of all evaluated metrics. A p-value < 0.05 was considered as statistical significant. The statistical testing was performed using the statsmodels package of Python (Python, version 3.12). As a convention, from here throughout the manuscript, SPECTpre denotes the planned absorbed dose based on the 99mTc-MAA scan, SPECTpost denotes the post-radioembolization measured absorbed dose with SPECT bremsstrahlung and PETpost denotes the post-radioembolization measured absorbed dose with PET. The median is given together with the IQR in parenthesis while the mean is given as mean ± standard deviation (SD) or/with range.