General workflow
The quantification of Gd mass in organ from SPECT images is an indirect process, see Fig. 1. First, the 111In concentration in a volume of interest (in red in Fig. 1) was estimated from images that were calibrated and corrected by scatter, attenuation, and partial volume effect. An 111In to Gd factor αInGd was estimated and applied to derive the corresponding Gd mass. Obtained values were compared for activity to reference gamma counter and for Gd mass to ICP-MS measurements. Studies were performed on both phantom and rat images acquired on a preclinical SPECT/CT device.
Nanoparticle radiolabeling
The AGuIX nanoparticles used in this study were obtained from NHTheraguix (Crolles, France). They are composed of a polysiloxane matrix bearing DOTA chelators on the surface able to chelate Gd3+ ions and 111In for SPECT experiments. The AGuIX NP hydrodynamic diameter is under 6 nm. The nanoparticles (50µL, 100 mM) were radiolabeled by adding 300 µL of citrate buffer 50 mM pH5 and 40–80 MBq of high purity 111In-chloride (Mallinckrodt, Petten, Netherlands). The mixture was incubated for 30 min at 40 ∘C. A diethylenetriaminepentaacetic acid (DTPA) was added at the end of radiolabeling after incubation for free 111In evacuation. Radiochemical purity of AGuIX-111In was over 97%.
For stability testing, an aliquot of the radiolabelled AGuIX-111In was incubated at 37 ∘C in 2 mL phosphate buffer saline (pH 7.4) and in rat serum, and radiochemical purity (RCP) was evaluated using ITLC-SG and citrate buffer 0.1M pH5 as mobile phase. This test showed that at 48h after incubation, RCP was still greater than 96% in phosphate buffer saline (pH 7.4) and in rat serum indicating a suitable kinetic stability to perform in vitro and in vivo experiments.
Phantoms and animals
For the in vitro studies, two sets of data were analyzed:
-
111In + Gd-NPs. Four tubes were used, containing 500 µL of saline solution with different concentrations of Gd-NPs labeled with 111In corresponding to 7.46 MBq, 4.20 MBq, 2.16 MBq, and 1.23 MBq at the SPECT imaging time. They were used to measure the Gd mass (described below).
-
111In without Gd-NPs. Six tubes were imaged, containing 250 µL of 111In corresponding to 9.85 MBq, 4.55 MBq, 2.50 MBq, 1.27 MBq, 0.67 MBq, and 0.37 MBq at the SPECT imaging time. The aim of this experiment was to evaluate the linearity of the image-based quantification for different activities.
For in vivo imaging, 9 OFA (Oncins France Strain A) male rats with chondrosarcomas were used, four weeks after tumor placement. The animals were injected both intratumorally or intravenously with an 111In radiolabeled Gd-NPs in the presence of DTPA, with activities ranging from 5 to 25 MBq measured for each injection with a dose calibrator (Capintec Inc., Florham Park, USA). Animals were sacrificed at 5 min and 30 min intratumorally and 1, 2, and 4 h for intravenously injected animal. The original study plan included measurements at 6H and at 24H. Even if the injected activity was not low (10–20 MBq), only the image obtained at 4H were usable while the signal-to-noise ratio of the others was insufficient. We also stop at 4H point because the full irradiation study, that is under publication, showed that the treatment is effective at this time point. Images were acquired post mortem in order to have reference measurements on extracted organs. Kidneys and tumors were removed and placed in formol in plastic tubes adapted for a gamma counter measurement of the activity. Supplementary images were also acquired on these tubes for ex vivo studies.
SPECT/CT image acquisition
We used a nanoSPECT/CT (Bioscan Inc., Washington D.C., USA) for preclinical imaging with multiplexing multipinhole apertures. It has four detection heads allowing the acquisition of four projections simultaneously. The pinhole collimator for rat imaging used in the experiments, named APT2, has 9 cone shape pinholes drilled from both sides of the collimator giving an opening diameter of 2.5 mm. The field of view of the camera is a cylinder with a diameter of 65 mm, and the axial scan length of 25 mm. We used 111In radionuclide emitting 171.3 keV (90.61%) and 245.4 keV (94.12%) gamma rays. Therefore, the projections were acquired for two energy windows of 10% around the peaks and one additional energy window of 209 keV ± 10% used for scatter correction.
The SPECT device acquired 24 projections (6 projections × 4 heads) of 256 ×256 pixels for every 15 degrees. The scan duration was 100 seconds per projection. The reconstruction was performed with the manufacturer software, HiSPECT, using an ordered subsets expectation maximization (OSEM) algorithm with 9 iterations and 4 subsets with an image voxel size of 0.6 mm.
The cone-beam CT scans contained 180 projections for the full coverage with a duration of 1 s/projection acquired with a beam voltage of 55 kV. The images were reconstructed with Feldkamp’s filtered backprojection reconstruction algorithm [15] with a voxel size of 0.4 mm. The reconstructed CT images were registered and resampled in order to match the sampling of the SPECT images.
SPECT image corrections
Scatter
The dual-energy correction method (DEW) [16] was used for the scatter corrections. This method consists in subtracting an estimate of the scatter component from the peaks. It is based on a measurement of the number of counts in one energy region near the peak. The two peaks, 171.3 keV and 245.5 keV, were corrected for the scatter component based on the number of counts in the scatter window between these peaks.
Attenuation
The reconstructed SPECT images were corrected with Chang’s multiplicative method [17] on a voxel-by-voxel basis. The linear attenuation coefficient images were recalculated from CT images using a bilinear model (see [18]) and based on the NIST tables of mass attenuation coefficients. The attenuation correction factors (ACF) for each voxel of the reconstructed image were obtained taking into account the gamma path through the tissue.
The 111In isotope has two photopeaks which means that the attenuation correction should be done for these two peaks separately. However, the comparison between the correction for two peaks simultaneously and separately gave a difference of ∼ 1%. Therefore, we applied the attenuation corrections in the two energy windows by calculating a weighted combination of the ACFs with experimentally defined weights for each peak component as w171keV=0.635 and w245keV=0.365.
Partial volume effect
SPECT images suffer from PVEs due to limited spatial sampling and a finite spatial resolution [19]. Therefore, a region of high activity tends to be underestimated and neighboring voxels overestimated. This means that if the VOI is selected from an anatomic CT image, the measured activity will be biased (underestimated in this case). In this study, we used post-reconstruction Müller-Gärtner method (MGM) [20] for partial volume effect correction in its generalization to two regions (see [21] for the detailed workflow).
Absolute calibration
For NanoSPECT/CT image calibration, we followed the NEMA standard protocol [22]. We used a cylindrical phantom with 5-mL volume containing the activity with a concentration of cVol=0.72±0.01 kBq/mL measured with a gamma counter (Wallac Wizard 1470 Gamma Counter, GMI inc.) and recalculated for the acquisition time. The calibration system volume sensitivity, SVol [23] (in cps/Bq), was calculated as:
$${} S_{\text{Vol}}\,=\,\frac{R}{c_{\text{Vol}}\cdot V_{\text{VOI}}}\times \text{exp} \left(\frac{T_{0}-T_{\text{cal}}}{T_{1/2}}\cdot \mathrm{ln2}\right)\\\!\times\!\left(\frac{T_{\text{acq}}}{T_{1/2}}\cdot \mathrm{ln2}\right)\times \left(1-\text{exp}\left(-\frac{T_{\text{acq}}}{T_{1/2}}\cdot \mathrm{ln2}\right)\right)^{-1},$$
where VVOI (in mL) is a volume of interest (VOI) placed in the reconstructed image, T0 is the start time, Tacq is the duration of the acquisition, T1/2 is the half-time of the radionuclide used, Tcal is the time of the activity calibration, and R (in cps) represents the counting rate measured in the VOI.
Volume of interest selection
For phantom and ex vivo studies, we used CT images for VOI selection. The difference of HU values for water and soft tissue and air provided an opportunity to obtain VOI by binarization of CT images with an adapted threshold.
For in vivo studies, the separation of kidneys and tumors from surrounding soft tissues was difficult in CT images. Therefore, the SPECT images were used for the VOI selection. For kidney studies, we first applied a threshold in SPECT images, and then, as a spill-out from the PVE would bias this selection, we eroded the VOIs in order to match the borders in CT images. For the tumor analysis, the threshold in SPECT images cannot be used in the same manner as the activity distribution was heterogeneous and its border mismatched the actual tumor borders in anatomical image (Fig. 2). The whole tumor VOI is presented in Fig. 2 in red and the VOI selected for activity above certain threshold is shown in green. Such VOIs were selected in SPECT images following these steps: (i) image with Gaussian filtering with σ of 1.5×(voxel size), (ii) define the threshold as 8% of a maximum value in a local region in the blurred image, and (iii) dilate the obtained VOI by 2 voxels. This approach has been approved in kidney images before using it in the final tumor image analysis.
Gadolinium quantity calculation
The obtained 111In activity measurement from SPECT images was used to find the equivalent Gd-NP quantity. The coefficient of proportionality between these two values, αInGd(t), was defined as
$$\alpha_{InGd}(t)=\frac{m_{Gd}}{A_{In}(t)},$$
where mGd is the mass of Gd-NP and AIn(t) is the 111In activity at time t. It could either be measured or calculated for each injection preparation. In these studies in case of measured αInGd(t), the activity, AIn(t), was obtained with the Wallac Wizard gamma counter and the Gd mass, mGd, was determined by ICP-MS. We also used calculated \(\alpha ^{calc}_{InGd}(t)\) coming from the same formula as above from initial proportions of gadolinium mass, mGd, and 111In activity at preparation time.
In the experiments on phantom tubes, αInGd(t) was obtained on the sample with the highest concentration and used to calculate the Gd-NP concentration for the three other tubes. In in vivo kidney studies, several independent samples from the same preparation as in the main analysis were used to determine αInGd(t) by fitting the linear proportionality between AIn(t) and mGd. In in vivo tumor studies, we also used the calculated \(\alpha ^{calc}_{InGd}(t)\).
Reference measurements for 111In and Gd-NPs
In order to evaluate the SPECT image-based activity quantification, we measured the reference values on the same Wallac Wizard gamma counter as for the calibration. The reference values of the Gd masses were obtained with the ICP-MC measurements. This study was originally made on post mortem animals because the reference values of the activity could only be measured on extracted organs.
Uncertainties estimation
In order to estimate the uncertainty on the activity measurements we summed up in quadrature the following individual uncertainties:
-
Standard deviation of a count rate in VOI
-
Uncertainty on mask selection was taken of 10%
-
Uncertainty on activity reference measurement with a gamma counter (2%)