Phantom characteristics

An anthropomorphic phantom (model ECT/TOR/P), including lungs and liver, was used to mimic patient anatomy. Two fillable spheres (S1 and S2) were placed in the liver to resemble tumors of different sizes.

For the 166Ho–99mTc dual-isotope measurements, the healthy-liver compartment (1205 mL) and spheres (S1: volume = 24.2 mL, radius = 1.79 cm and S2: volume = 15.7 mL and radius = 1.55 cm) were filled with 166Ho-chloride with a sphere to healthy-liver compartment concentration ratio of 10:1, resembling a high tumor-to-non-tumor uptake, typically reported for large, highly vascularized tumors [7, 8]. This resulted in an activity percentage, with respect to the total activity in the liver, equal to 15.10% for S1 and 9.76% for S2. In order to consistently achieve an equivalent of 250 MBq of 166Ho across the various measurements, which resembles the prescribed scout activity as used in the clinical studies [9], the imaging time for each scan was adjusted (17.8–38.2 s per projection) with respect to the clinical protocol (20 s per projection) to compensate for 166Ho decay between scans. The validity of this approach was based on the assumption that the difference in dead time had little effect on count statistics. The measured dead time rate shifted on average from 4.5% at the higher activities to 2.5% at lower activities compensated by longer scanning time. 99mTc activity was injected multiple times in the healthy-liver compartment leading to various 166Ho:99mTc ratios, ranging from 2 to 10. Effective 99mTc activity, modified to correct for the varying imaging times, ranged from 25 to 126 MBq.

In a separate series of measurements, the anthropomorphic phantom was filled only with 99mTc in the healthy-liver compartment (activity ranging from 8 to 240 MBq). These 99mTc-only measurements were used as reference for comparison of the 99mTc images acquired in presence of 166Ho.

No radioactivity was injected into the lung compartment of the phantom, nor the torso compartment.

Phantom data acquisition

All images were obtained using a Symbia T SPECT/CT scanner (Siemens, Erlangen, Germany), using medium-energy collimators. Projections were recorded on a 128 × 128 matrix (pixel spacing, 4.8 × 4.8 mm), with 120 angles, over a non-circular 360° orbit using step-and-shoot mode. Energy windows used for image acquisition are summarized in Table2

Table 2 Energy window characteristicsPhantom data reconstruction

All images were reconstructed using commercially available software (Siemens Flash3D), with 10 iterations, 8 subsets, incorporating scatter and attenuation correction. No post-reconstruction filtering was applied.

166Ho-99mTc dual-isotope downscatter correction

166Ho images were reconstructed with window-based scatter correction, using projections acquired in the 118 keV energy window (scaled by a k-factor) as an estimate for downscatter in the 81 keV photopeak window originating from both 99mTc and higher-energy 166Ho gamma emissions and bremsstrahlung.

Starting from the k-factor value previously computed by dividing the counts in the 81 keV and 118 keV energy window of 166Ho–99mTc dual-isotope projections [6], the k-factor for different 99mTc activities was empirically investigated by reconstructing 166Ho images for a variety of k-factors ranging from 0.65 to 1.30 with a 0.05 interval. The optimal value of the k-factor was tuned by measuring, and minimizing, the impact of 99mTc activity on the 166Ho count density measured on 166Ho image reconstruction. Photopeak scatter, i.e., scattered photons originating from the 81 keV 166Ho photopeak, was not accounted for.

99mTc images were reconstructed using the 118 keV and 170 keV windows for triple-energy-window scatter correction, and the scatter was estimated as:

$$_= \left(\frac_}_}+\frac_}_}\right)\times \frac\times _=\left(\frac_}_\times 2}\right)\times _+\left(\frac_}_\times 2}\right)\times _$$

where \(_\) and \(_\) are the recorded projections for the lower (Scatter_118) and upper scatter (Scatter_170) windows, respectively, and \(_\), \(_\) and \(_\) are the widths of the lower, upper and main photopeak energy windows.

For consistency, this method was also applied when no 166Ho activity was present in the phantom.

Phantom data analysisVOI definition

VOIs matching the phantoms’ liver compartment and sphere inserts were defined on a high resolution CT. The sphere VOIs were subtracted from the liver mask to produce the healthy-liver compartment VOI. These pre-defined VOIs were registered to each SPECT/CT reconstruction using Elastix [10, 11]. Grid matrices were super-sampled to allow partial voxels to be included within the VOIs.

Uniformity

The healthy-liver uniformity for different 99mTc activities was quantified by the coefficient of variation (COV), defined as the ratio of the standard deviation to the mean, computed within the healthy-liver compartment VOI, for both 166Ho and 99mTc reconstructions. The COV was computed for each 99mTc image, acquired either in presence or not of 166Ho in the phantom. A binary erosion of 1 cm [12] on the healthy-liver mask was applied to avoid edge effects.

Contrast recovery

Image quality can be assessed by analyzing the contrast recovery coefficient for either hot or cold spheres (\(_\) or \(_,\) respectively), generally defined as:

$$Q = \frac /C_ - 1}}} \times 100\%$$

where \(_\) is the mean intensity measured in the sphere VOI, \(_\) is the mean intensity measured in the healthy-liver compartment VOI, and R is the nominal activity concentration ratio between spheres and healthy-liver compartment. However, for cold spheres, R is zero by definition.

The effect of adding 99mTc activity to the healthy-liver compartment was assessed by measuring the contrast recovery coefficients on both 166Ho and 99mTc reconstructions (\(_\) or \(_,\) respectively). The nominal activity concentration ratio between spheres and healthy-liver compartment, R, was 10 for 166Ho reconstructions, but since only 166Ho was present in the spheres, and not 99mTc, R was zero for all 99mTc reconstructions.

Healthy-liver segmentation

The usability of the 99mTc scans for the purpose of automatic segmentation of the healthy-liver was investigated by analyzing the overlap between the segmentations and the pre-defined healthy-liver compartment VOI for images acquired at different 99mTc activities. The segmentations were obtained using a thresholding procedure. The accuracy of a standard thresholding procedure relies on the choice of the threshold value, which is typically defined as a percentage of the maximum image intensity. This, however, implies that the segmentation relies on a single voxel value, the maximum, which is prone to inaccuracy due to noise. To reduce this dependency, the threshold value was instead based on a percentage (α) of the maximum value after having smoothened the image using a 3D Gaussian filter. The threshold was then applied back to the original, un-smoothened, image to produce the segmentation.

For every individual scan, an optimal threshold percentage α could be determined by applying an optimization routine which varied α to correctly recover the volume of the healthy-liver in the phantom. However, as these values for α may be different between scans, a single value to apply to all scans was defined as the average of all individual optimal values.

The accuracy of the healthy-liver segmentation using 99mTc images was evaluated by assessing both the recovered healthy-liver compartment and the resulting cold spheres (i.e., the tumors). The ratio between the segmented volumes and the nominal volumes was computed for the three VOIs (cold sphere S1 and S2, and healthy-liver compartment). In addition, the overlap between the segmentations and the nominal VOIs was assessed through the Sørensen–Dice index [13].

Statistical analysis

For the above-mentioned metrics (uniformity, cold-sphere contrast recovery coefficients and healthy-liver segmentation), a t test was used to determine if there was a significant difference between the measurements acquired in presence or not of 166Ho. P-values were reported only if a statistically significant difference was found.

Impact of k-factor on 166Ho phantom reconstructions

SPECT reconstructions of 166Ho images suffer from downscatter induced by higher-energy gamma emissions and bremsstrahlung, detected in the 81 keV 166Ho photopeak window. The 118 keV energy window, scaled with a k-factor, is used as an estimate for these downscatter contributions. In case of dual-isotope 166Ho–99mTc imaging, there is an additional downscatter contribution arising from the 99mTc photopeak at 140 keV. Ideally, however, with a well-chosen k-factor, the 166Ho reconstructions are independent of 99mTc activity.

The impact of 99mTc on 166Ho images can be assessed by the COV and the contrast recovery coefficients (of hot spheres), similarly to the 99mTc analysis. However, both these metrics strongly depend on the 166Ho count density in the healthy-liver compartment.

To determine the optimal k-factor, the count density in the healthy-liver compartment VOI was measured for all 166Ho images, reconstructed for a range of k-factors (0.65–1.30 with a 0.05 step interval). For each k-factor, the relative change in 166Ho count density was determined as a function of 99mTc activity.

Patient data

To clinically evaluate the findings regarding the k-factor obtained using the phantom scans, a similar analysis was applied to images from patient procedures, for which both a 166Ho–99mTc dual-isotope and a 166Ho-only acquisition was available.

For all patient SPECT/CT acquisitions used in this study, informed consent was obtained as part of the HEPAR PLuS study [14]. Twenty-six scout (pre-treatment) procedures performed on patients with liver metastases of neuroendocrine tumors were analyzed (median administered activity (and interquartile range): 224 (35) MBq of 166Ho).

According to the HEPAR PLuS study protocol, for each scout procedure, two SPECT/CT images were acquired: a 166Ho-only SPECT/CT and, after administration of 99mTc-stannous phytate, a 166Ho dual-isotope SPECT/CT. All scans were acquired and reconstructed using the same protocols as those adopted for the 166Ho–99mTc dual-isotope phantom data. Similar to the 166Ho phantom scans, 166Ho patient images were reconstructed using multiple k-factors ranging from 0.65 to 1.30 with step 0.05.

To assess the impact of the k-factor on 166Ho patient reconstructions, a volume of interest was defined containing the healthy-liver (by thresholding the 99mTc image). Within this healthy-liver VOI, the 166Ho count density was determined for both acquisitions, 166Ho-only and 166Ho dual isotope, using the same k-factor for both reconstructions. The percentage difference in 166Ho count density between the 166Ho dual isotope and 166Ho-only acquisition was computed for each k-factor (ranging from 0.65 to 1.30).