All animal experiments were conducted in compliance with European Union Directive 2010/63/EU on the protection of animals used for scientific purposes. The protocol was approved by the local animal research ethics committee. All surgeries were performed under general anesthesia and aseptic conditions and were supplemented by appropriate analgesic programs.

The VX2 rabbit tumor is a commonly used animal model for translational research on HCC in interventional radiology [18]. Implantation of a VX2 fragment was performed in healthy New Zealand white rabbits (Charles River Laboratories, Saint-Germain-Nuelles, France).

VX2 well-vascularized tumor fragments (25 mg) were sampled from a carrier animal and immediately implanted in the left median lobe of the exposed liver of the recipient rabbits. One donor was used for 3–6 receivers. Tumor growth lasted at least 19 days after implantation. Ultrasound imaging was performed to ensure that the tumor had reached a length of at least 10 mm (major axis); otherwise, the animal was kept until the tumor was workable. Nineteen to twenty days after tumor induction, the population was divided into 3 groups: L for Lipiodol®, M for microspheres, and C for control.

The rabbits of the L and M groups received buprenorphine (Buprecare® 0.14 mL/kg) 1 h before surgery and were hydrated with 50 mL of saline subcutaneously in the flank. Then, they received an intravenous injection of heparin diluted to 1/10 at a dose of 50 IU/kg in the ear. A pediatric valve introducer 4F (Radifocus® TERUMO™) was inserted into the femoral vein and a 1.7F catheter (Microcatheter 1.7F angle 90° - ECHELON™ - MEDTRONIC EV3) was guided under x-ray angiography (Philips Veradius®) to the feeding artery of the tumor at the level of the left hepatic artery. After removal of the catheter, the skin and muscle planes were sutured at the paw level.

The L group received an adjusted dose of Lipiodol® ultra-fluid into the left common hepatic artery up to reflux or pulmonary passage and to a maximum volume of 0.4 mL. The Lipiodol® ultra-fluid (Guerbet) injection liquid contains per 1 ampoule of 10 mL ethyl esters of iodized fatty acids of poppy seed oil, equivalent to 4.8 g of iodine (480 mgI/mL).

The M group received a fixed volume of 0.3 mL of microspheres in the same injection site. The radiopaque microspheres used in this study were made polyethylene glycol methacrylate (PEGMA) resin microspheres and were sieved to obtain an average diameter of 33 µm. They were made by Guerbet Research representative of approved microspheres in terms of size, which have been customized to make them radiopaque for the purpose of the study. Just before injection, 300 µL of microspheres were taken from the vial and suspended in 3 mL of saline water. The total amount of this suspension was injected slowly (about 0.1 mL∙min−1).

The C group received nothing.

Different time intervals were studied to investigate the distribution kinetics of the products. Because of its ability to extravasate leading to a possible modification of distribution during the first hours after injection, the pharmacokinetics of Lipiodol® ultra-fluid (L group) was studied at different timepoints (15 min (D0), 1, 2, 6, 9 and 12 days). For microspheres (M group) which are known to stay several months in the intravascular compartment, only the following delays were studied: 15 min (D0) and 12 days (D12) after injection. The C group was imaged at 15 min, 6 days, and 9 days. At studied time-points, the rabbits were euthanized by an intravenous injection of pentobarbital at a dose of 1 mL/kg under general anesthesia. The liver was explanted, and the tumor was isolated for high resolution 3D X-ray micro-computerized tomography (µCT). A Quantum GX2 (Perkin-Elmer) was used with the following parameters 90 kV, 88 µA, and a CuAl filter, and an acquisition time of 14 min. The field of view diameter was 72 mm or 86 mm depending on the size of the tumor, leading to a voxel side of 0.144 mm or 0.172 mm.

For the L group, as soon as the µCT image was acquired, the tumor was cut into slices of up to one centimeter, frozen (− 80 °C) and sent for analysis to Oncovet Clinical Research (Clinical Research, Loos, France). Frozen samples of liver with tumor were cut into sections of 12 µm thick. The sections were stained with Hemalum-Eosin after a previous silver staining (2.5%, 60 min, 4 °C) allowing the detection of Lipiodol® ultra-fluid. Assessments from the resulting histologic slides were performed by a veterinary pathologist blinded to sample. The Lipiodol® ultra-fluid and microspheres distributions were studied in the vascular network and in the parenchyma of the tumors.

To compare Lipiodol® ultra-fluid and microspheres capabilities to penetrate into tumor tissues, we applied a set of first-order radiomic features on the µCT images. To do so, the tumors were segmented manually using the software tool 3DSlicer [19]. The radiomics features were extracted using the SlicerRadiomics extension based on PyRadiomics [20]. A Spearman correlation test was done between time delay, tumor volume, and each radiomic feature. For these variables, the 3 groups were compared using the non-parametric Wilcoxon test. The statistical significance was considered to be achieved for a p value below 0.05.

Tri-dimensional (3D) dosimetry was modelized based on the Lipiodol® and microspheres distribution deduced from the µCT images. The tumor contours previously defined for the radiomic analysis were used. The distribution volume of iodine was segmented by manual thresholding. All voxels belonging to this structure were scaled so that the values were ranging from 0 to 1. The resulting image templates were then used to generate the activity maps so that the total activity within the tumors was equal to 1 MBq.

The activity in voxels was converted to time-integrated activity, which is also referred as the total number of disintegrations over the course of the treatment. In radioembolization, the calculation is simplified by the fact that the biological half-life is far greater than the physical half-life of the radionuclides used. Thus, time-integrated activity Ã(s) in each source voxels was calculated as

$$\widetilde(s)=\frac$$

(2)

with A(s,t = 0) being the initial activity in the voxel and λ the decay constant of the radionuclide.

The absorbed dose was calculated in water using dose-point kernel (DPK) convolution implemented in a previous study [21, 22]. Water DPKs had a resolution of 0.1 mm. The dose D(x) at position x was calculated as

$$D\left(x\right)=\iiint \widetilde\left(s\right)_\left(\left|s-x\right|\right)\mathrms$$

(3)

with s being the position of the source, \(\widetilde\left(s\right)\) the time-integrated activity, and kw the kernel in water.

The absorbed dose by tumor was calculated for each radionuclide in Gy per MBq administered to the tumor, which equals the ratio of S factor over the radionuclide decay constant λ. Indeed, according to the medical internal radiation dose (MIRD) formalism [23, p. 21], the tumor-absorbed dose is expressed as:

$$D=\widetilde\times S$$

(4)

Hence, knowing that for radioembolization \(\widetilde=\frac\), the tumor absorbed dose over the administered activity within the tumor can expressed as:

To compare the biological efficacy between absorbed dose distributions, the biological effective dose (BED) was calculated according to the linear-quadratic model applied to radioembolization [24] as:

$$\mathrm=D\left(1+\frac\times \frac\times D\right)$$

(6)

with μ the DNA repair constant, α and β are the linear and quadratic cell killing constants. We set the value of μ to 0.46 h−1 as reported by Cremonesi et al. for tumors [25], and α and β values to, respectively, 0.037 Gy−1 and 0.0028 Gy−2, as reported by van Leeuwen et al. [26].

To consider the heterogeneity of absorbed dose distribution, we implemented the equivalent uniform dose (EUD) concept of Jones and Hoban [27] to the BED leading to the EUBED:

$$\mathrm=-\frac\mathrm\left(_^^}_}\times _\right)$$

(7)

with BEDi being the histogram ith bin, vi the volume fraction, and N the number of histogram bins. EUBEDs were calculated for absorbed doses ranging from 1 to 1000.

Mean and standard deviation were calculated for the following variables for each radionuclide: tumor volume V, S/λ, EUBED(D = 100 Gy). The L and M groups were compared for each variable using the Kruskal–Wallis test by ranks. The statistical significance was set for a p value < 0.05. All statistics and graphics were processed using RStudio 2022.12.0.353 [28] and R 4.2.2 [29].