Energy windows setting
An Infinia Hawkeye 4 gamma camera (General Electric, USA), equipped with a 5/8-in. crystal and a medium-energy general-purpose (MEGP) collimator, was used. Note that Owaki et al. [24] proved in a clinical study that a high-energy general-purpose (HEGP) collimator gives a higher image contrast than a MEGP collimator, yet an HEGP collimator was not available in our study. The energy window setting was chosen based on the maximum intensities of photons emitted by 223Ra. Therefore, three photopeak energy windows were chosen: 85.0 keV ± 20%, 154.0 keV ± 10%, and 270.0 keV ± 10. This choice includes several photon emission ranges of 223Ra and the 271.0 keV photons emitted from daughter products 219Rn. Then, three energy windows for scatter correction were chosen: 47.0 keV to 67.0 keV, 103.0 keV to 123.0 keV, and 210.2 keV to 242.8 keV.
In order to quantify the contribution of each of these energy windows, we used first the NEMA phantom (Data Spectrum™, USA). This phantom contained six fillable spheres of different diameters: 10, 13, 17, 22, 28, and 37 mm (0.5, 1.1, 2.6, 5.6, 11.5, 26.5 ml). We filled each of these spheres with a solution of 2.7 kBq/ml of 223Ra. In order to model the attenuation and scatter, the remaining portion of the phantom was filled with water. We placed the phantom on the table in the center of the field of view and positioned the gamma camera heads at 10 cm in front of it. We acquired SPECT/CT images. The acquisition parameters were 6° between each projection, 128 × 128 matrix, pixel size 4.4 mm, and circular orbit. For each energy window, the images were reconstructed on the clinical workstation XELERIS (General Electric, USA) using the OSEM algorithm with the following parameters: 2 iterations, 10 subsets, Butterworth filter with fcut = 0.48 cycle/cm and p = 10. During the reconstructions, each image was corrected for attenuation using CT-based attenuation maps and for scatter using the Jaszczak method [25]. Indeed, during the SPECT/CT acquisition, three images corresponding to the three emission energy windows and three images corresponding to the scatter windows were created. The software also generates three CT-based attenuation maps with attenuation coefficients corresponding to each photopeak energy. So, the projections corresponding to the 85 keV ± 20% emission window were corrected for attenuation with its corresponding attenuation map (μ = 0.179 cm−1 in water) and for scatter with the 47–67 keV scatter acquisition. The projections corresponding to the 154 keV ± 10% emission window were corrected for attenuation with its corresponding attenuation map (μ = 0.147 cm−1 in water) and for scatter with the 103–123 keV scatter acquisition. Finally, the projections corresponding to the 270 keV ± 10% emission window were corrected for attenuation with its attenuation map (μ = 0.121 cm−1 in water) and for scatter with the 210.2–242.9 keV scatter acquisition. Finally, the reconstructed images were summed to evaluate every combination: 85 keV versus 85 keV + 154 keV versus 85 keV + 154 keV + 270 keV.
For the analysis of the background and of the reconstructed signal, we manually delineated six spherical regions of interest (ROIs) using the PLANETOnco software (Dosisoft, France) on the merged SPECT/CT images covering the six hot spheres. The ROI diameters were equal to the physical inner diameters of the six hot spheres. To compare a background noise metric common to all the six hot spheres, we selected as target slice for background analysis an imaging slice containing all the six hot spheres. For each size of the hot spheres, we positioned 12 background 3D ROIs on the target slice. Six different ROI sizes were used, and the ROI diameters were equal to the physical inner diameters of the hot spheres. Figure 1 shows the 37-mm-diameter ROIs (corresponding to the biggest sphere of the NEMA phantom) overlaid on the target slice.
From these measurements, we computed the sensitivity and the signal-to-noise ratio (SNR) for each hot sphere. The sensitivity was calculated as S = Counts/(t A0), where Counts is the total number of counts measured in a ROI within the radioactive volume; t is the acquisition duration (seconds); and A0 is the activity in the sphere (MBq). The SNR is defined for each hot sphere as
$$ \mathrm{SNR}=\frac{C_{\mathrm{hot}}}{C_{\mathrm{background}}} $$
where Chot is the counts measured in the hot sphere and Cbackground is the mean of the counts measured in every background ROIs of the same size as the considered hot sphere.
Second, in order to evaluate the contribution of each emission window on the spatial resolution, we used a Triple Line phantom (Data Spectrum™, USA). This phantom contained three parallel linear capillaries (1 mm diameter): a central one and two lateral ones. These three linear capillaries were filled with 10.8 MBq/ml of 223Ra. The phantom was positioned in the center of the field of view at a distance of 10 cm from the heads of the gamma camera. Acquisition and reconstruction parameters were identical to those used for the NEMA phantom. The acquisitions were performed with and without an attenuation medium (water) in the Triple Line phantom. For each acquisition, we fitted the profile of each linear source on both the sagittal and axial views with a Gaussian function. The spatial resolution was calculated as the full width at half maximum (FWHM) of the fitted Gaussian function.
Reconstruction parameters for SPECT imaging
First, the NEMA phantom was used to assess the best reconstruction for quantification purposes. Each sphere of the phantom was filled with a solution of 20 kBq/ml of 223Ra. The remaining portion of the phantom was filled with water. We placed the phantom on the table in the center of the field of view at a distance of 10 cm from the gamma camera heads. The acquisition parameters were 30 s/projections, 6° between each projection (128 × 128 matrix, pixel size 4.4 mm, and circular orbit).
In order to assess the best filter and number of iterations of the OSEM algorithm, we performed several reconstructions of the SPECT/CT acquisition on the clinic workstation. We tested several configurations, namely, the use of 1 to 10 iterations, with Hann (fcut = 1.56 cycle/cm) filter, Butterworth (fcut = 0.48 cycle/cm and p = 10) filter, Gauss (FWHM = 4 mm) filter, and with no filter (none). The scatter was compensated in each emission windows using the Jaszczak method [25]. Compensation for attenuation was performed using the attenuation maps from the CT acquisition.
In order to study the reconstructed signal, we used the manually delineated ROIs as described in the previous section. The sensitivity and the signal-to-noise ratio (SNR) were calculated for each hot sphere.
Second, the spatial resolution was evaluated with a variable number of iteration (n = 1–9). The same Triple Line phantom with the background filled with water as described in Section 2.2 was used. Image acquisition for SPECT was the same as those for the NEMA phantom.
Calibration factors
We performed SPECT/CT acquisitions on the NEMA phantom over a 223Ra concentration range from 1.8 to 22.8 kBq/ml in each sphere. The acquisitions and reconstructions were carried out following the acquisition and reconstruction parameters determined above.
For each 223Ra concentration, the calibration factor (CF) (cts/s/MBq) was calculated for each sphere with the following equation [26]:
$$ \mathrm{CF}=\frac{C_{\mathrm{measured}}}{A\times t} $$
where Cmeasured is the measured number of counts in the delineated 3D ROI surrounding each hot sphere; A denotes the activity in each sphere; and t is the acquisition duration.
Validation with an anthropomorphic phantom
Once the optimal acquisition and reconstruction parameters were established, the accuracy of quantitative 223Ra SPECT imaging was investigated using an anthropomorphic TORSO® phantom (Orion, France), which is designed to mimic as close as possible clinical conditions [27]. This phantom contained a liver insert, lung inserts, and a cylindrical insert of 156 ml nominal volume. Two spheres of 0.5 ml and 5.6 ml were placed in the cylindrical insert and another 5.6 ml sphere was fixed on it (Fig. 2).
In order to mimic 223Ra uptakes in the healthy bone and lesions, we used the following tumor to normal tissue (TNT) ratios between the spheres and the cylindrical insert: 6, 10, and 30. The liver insert and the phantom background were filled with water. We filled the lung inserts with Styrofoam® beads and water to mimic lung tissues. Several concentration ranges were used in the spheres: from 2.3 kBq/ml to 8.1 kBq/ml for a TNT = 30, from 8.7 kBq/ml to 21.5 kBq/ml for a TNT = 10, and from 22.8 kBq/ml to 64.0 kBq/ml for a TNT = 6.
To study the accuracy of activity quantification in SPECT/CT images, we used the recovery factor (RF) as described in the MIRD pamphlet 23 [26]. This factor is defined as the ratio between 223Ra activity estimated from the image and the true activity in the object. The activity estimated from the image (Aexpected) was assessed for each sphere using the following equation:
$$ {A}_{\mathrm{expected}}=\frac{C_{\mathrm{measured}}}{\mathrm{CF}\times t} $$
where Cmeasured is the number of counts measured in each spherical ROI corresponding to a hot sphere, CF is the calibration factor, which depends on the sphere sizes and was established on the NEMA phantom, and t is the acquisition duration. The recovery factor was analyzed as a function of 223Ra concentrations and TNT ratios.