The Jaszczak Pro-NM Performance, complying with National Electrical Manufacturers Association (NEMA) standards publication (NU 1-2001), consists of a cylindrical phantom with an inside cylinder diameter of 206 mm, an inside cylinder height of 186 mm, and a cylinder wall thickness of 7 mm. The cold rods and cold spheres inserts were removed. Instead, a complementary set of fillable spheres was used (with inner diameters of 9.9 mm, 12.4 mm, 15.4 mm, 19.8 mm, 24.8 mm, and 31.3 mm, and a wall thickness of 2 mm), as illustrated on the Fig. 1.

The Jaszczak Pro-NM Performance, complying with NEMA standards publication (NU 1-2001), cold rods and cold spheres inserts being removed and replaced by fillable spheres

The phantom and the six spheres were filled with a holmium-166 chloride (166HoCl) solution mixed with Diethylene Triamine Penta Acetic (DTPA) acid to avoid inhomogeneous distribution of the radioisotope due to stickiness to the plastic walls of the phantom and specifically of the spheres. The activity ratio between spheres and background was 18:1. The goal was to respect the well-known limit of 400 MBq [6], [11] for the total activity, and to approach the real activity used for a typical 166Ho-PLLA simulation when 50–200 MBq are administered to the patient at our institution in a typical tumor volume of 100 ml (calculated as the median size of hepatocellular carcinomas (HCC) treated with 166Ho-PLLA in our center). We finally filled the hollow spheres with 29 MBq homogeneously distributed in the 31.5 ml of the total spheres volume, to approach a 1 MBq/ml ratio mimicking what was experienced with tumors, and filled the background with 312 MBq homogeneously distributed in the 6200 ml of the total background volume, for a total of 341 MBq at the beginning of the first acquisition. The phantom was acquired 4 times to a final total activity of 321.5 MBq at the beginning of the last acquisition, therefore staying under the recommended 400 MBq [6], [11].

Nineteen patients have been included, for a total of 21 treatments, 2 patients being treated twice. All the patients except one (cholangiocarcinoma), presented unresectable hepatocellular carcinoma (HCC), diagnosed according to European Association for the Study of the Liver and European Organisation for Research and Treatment of Cancer (EASL-EORTC) guidelines [14]. Our workflow was performed over two separate sessions: the simulation and the treatment (TARE).

Simulation evaluation started with an angiography in order to obtain a precise map of the patients’ abdominal vascular anatomy. Then, the simulation included the administration of 40 to 179 MBq of 166Ho-PLLA, so called “scout” (QuiremScout®), following the tumor size and perfusion, in order to predict the distribution pattern expected after treatment. 166Ho-PLLA TARE was performed (with QuiremSpheres®) within 14 days after the simulation. The amount of 166Ho activity administered to a patient for TARE purposes depended on the tumor perfusion and the predicted tumor absorbed dose. As previously mentioned, recent data (not focusing on HCC) showed a dose–response link, i.e. an estimated tumor dose of 168 Gy needed for a partial response and of 232 Gy for a total response [3]. Following these requirements, the compartmental dosimetry software Q-Suite™ 2.0, was used to calculate the activity that needed to be injected, in order to expect tumor doses between 169 and 300 Gy for the 21 treatments, leading to administered activities from 1.86 to 13 GBq. The absorbed dose to the lung never exceeded 30 Gy in a single treatment or 50 Gy in multiple treatments.

One post-treatment dosimetry was excluded because the threshold of 400 MBq [6], [11] was overcome during the imaging, leading to a final dataset of 21 simulations and 20 treatments.

The activity uptake is visualized by a whole body planar imaging and SPECT imaging of the abdomen, including a low dose computed tomography (CT).

Imaging was performed using the same Philips BrightView XCT SPECT/CT system (Philips Medical Systems, Cleveland, Ohio, USA) for every session. SPECT/CT acquisitions of the phantom were realized with the same acquisition parameters used for the SPECT/CT imaging of patients, except the distance from collimators being fixed at 30 cm from the phantom center. This distance was as close as possible for patients (using autobody contouring).

The thicker collimator available was used: the Medium Energy General Purpose (MEGP) collimator, with 0.86 mm septa thickness of lead and a length of 58.4 mm [15]. This was the best possible choice knowing its intrinsic limitations studied by Bayouth et al. and reminded in the introduction [11]. A main acquisition energy window around 80.6 keV with 15% width (equivalent to 12 keV) was chosen [4] (“window 2” or “W2” in Table 1). For patients’ acquisitions, firstly a whole-body scan was acquired to assess the absence of extra hepatic deposition (18 cm/min, 256 pixels wide). Then, a SPECT imaging (lung-liver centered) was performed to estimate the spatial distribution of 166Ho-PLLA (120 projections, 30 s/projection, 360°, 128 × 128 pixels matrix size) [4], together with a cone-beam (CB) CT imaging (for attenuation correction). Concerning the scatter correction, 2 extra acquisition energy windows were set, the camera being limited to 3 concurrent imaging. The first one was set around 118 keV [4, 11, 13] with 10,2% width (12 keV) for DEW method purposes (“window 4” or “W4” in Table 1). We conceived an alternative window, corresponding to the sum of 2 narrow windows around the main window, which are then considered as a unique acquisition window: 71.56 keV with 8.39% width (6 keV) (“window 1” or “W1” in Table 1) and 89.65 keV with 6.69% width (6 keV) (“window 3” or “W3” in Table 1). The latter acquisition window (W1 + W3) was used for the TEW method correction. Every count collected in one of those 3 acquisition windows (W2, W1 + W3 and W4) lead to a different image. The sizes of those acquisition energy windows (illustrated on the Fig. 2) were obviously chosen to easily apply the scatter correction by a user-friendly mathematical operation consisting in the subtraction of the acquisition image (projection) corresponding to one of the extra windows from the acquisition image (projection) corresponding to the main window.

Decomposed energy spectrum simulated using GATE. Spherical source in cylindrical water phantom with MEGP collimator. The yellow energy window 1 (W1) is 6 keV wide, the red energy window 2 (W2) is 12 keV wide, the green energy window 3 (W3) is 6 keV wide, and the purple energy window 4 (W4) is 12 keV wide

Every projection image (uncorrected, DEW-corrected, TEW-corrected, or numerically simulated) used in this work was reconstructed with the same method using the Ordered Subset Expectation Maximization (OSEM) algorithm with 10 iterations, and 8 subsets. An additional Butterworth filter (cutoff = 0.25 and order = 1.5) was applied for visualization purpose.

The numerical simulations were performed using the GATE 7.0 software which is an open-source simulation software for medical imaging and radiotherapy purposes [16]. It has been developed by the international OpenGATE collaboration which regroups 18 worldwide institutions. GATE uses the well-known Geant4 software to simulate the particles production and transport. It can manage by itself the detector and signal-processing chain. The particle transport is based on a Monte-Carlo method which allows to reproduce numerically the underlying physics that are the cause of the corresponding image output from a SPECT acquisition. In practice, firstly, the different components of the SPECT device used in this work were reproduced in GATE considering their material and shape, adapted from the script from OpenGATE collaboration and University Hospital Carl Gustav Carus in Dresden [17]. The Jaszczak Pro-NM Performance phantom was also coded in GATE. Then, the physical interactions of interest were activated considering for this purpose a full standard physics list (available in Geant4).

At the end of the simulation process, we obtained two major outputs from GATE: first, a characterization of the spectrum components considering real acquisition conditions for a sphere centered in a water cylinder mimicking the size of the Jaszczak phantom, and second, a Jaszczak phantom scatter-free and attenuation-free projections set considering a vacuum environment in place of real matter for phantom and air.

To compare scatter correction methods, we chose to evaluate Contrast Recovery Coefficients (CRC) on Jaszczak phantom reconstructed images. CRC were calculated for a sphere i following the Eq. 1 [18]:

$$_=\frac__}/__}}_/_}$$

(1)

The volume of interest (VOI) for a sphere is defined as all the voxels inside the physical (CT) volume avoiding border voxels affected by partial volume effect. The background VOI is a cylinder (with a 25 mm radius) centered in the transverse plan which includes the centers of the Jaszczak spheres. As explained by Stam et al. [18], pixcountssphere,i represents the activity concentration measured on the image in sphere i, activitysphere,i represents the actual activity concentration in sphere i, pixcountsbackground is the activity concentration measured on the image in the background volume, and activitybackground is the actual activity concentration in the background volume.

The calculated CRC were fitted using a sigmoid interpolation [19] following the Eq. 2:

$$S\left(x\right)=\frac^}-d$$

(2)

In this equation, a = 1 because CRC tends to 1 after a certain sphere volume, as showed by Cherry et al. [19]. The other parameters are determined using a least square method. The variable x is the ratio sphere diameter / Full Width at Half Maximum (FWHM) reflecting the spatial resolution.

To compare scatter correction methods, we also evaluated Contrast to Noise Ratio (CNR) on Jaszczak phantom reconstructed images. The CNR for a sphere j is calculated following the Eq. 3 [20]:

$$_=\left|\frac__}-_}_}\right|$$

(3)

As explained by van Gils et al. [20], Csphere,j is the average number of counts in the sphere j, Cbackground is the average number of counts in the background VOI and σbackground is the standard deviation in the background VOI.

Other metrics such as the homogeneity and/or coefficient of variation of axial and radial profiles are available, but those metrics are better suited to a homogeneous phantom, which was not what was used in our study.

The compartmental dosimetry software Q-Suite™ 2.0 was used to assess predictive dosimetry and post-treatment dosimetry for every treatment. For predictive dosimetry only, Q-Suite™ 2.0 at first predicts the lung dose after contouring the lungs and the whole liver on the CT images associated to the SPECT and specifying the planned activity to be administered. This method is indeed a volumetric SPECT-CT evaluation. Then, CT, T1- or T2-weighted MRI can be used to define compartments in the liver, as tumors and non-tumoral liver (NTL) tissue. A manual rigid registration is available to co-register the SPECT-CT images used to generate the dose map.

In Q-Suite™ 2.0, Dose Point Kernel model [21] is only available for post-treatment purposes, whereas Local Dose Deposition model [22] is available for predictive and post-treatment dosimetry. We therefore used the latter model for both dosimetries.

The TEW method was chosen to correct the simulation images used to calculate (in a predictive way) the activity to administer for TARE treatments.

To evaluate the impact of the choice of the scatter correction on personalized dosimetry, we use paired t-tests between parameters’ distributions obtained for the 21 pre-treatment dosimetries conducted on DEW and TEW method corrected images, as well as for the 20 post-treatment dosimetries also realized on DEW and TEW method corrected images. The parameters taken into account are the following outputs of Q-Suite™ 2.0: tumor dose, non-tumoral liver dose, tumor fraction receiving at least 150 Gy, and non-tumoral liver fraction receiving between 0 and 50 Gy. These analyses led us to 8 statistical comparisons of 16 datasets.