¹⁶⁶Holmium–^99mTechnetium dual-isotope imaging: scatter compensation and automatic healthy-liver segmentation for ¹⁶⁶Holmium radioembolization dosimetry

Abstract

Background

Partition modeling allows personalized activity calculation for holmium-166 (¹⁶⁶Ho) radioembolization. However, it requires the definition of tumor and non-tumorous liver, by segmentation and registration of a separately acquired CT, which is time-consuming and prone to error. A protocol including ¹⁶⁶Ho-scout, for treatment simulation, and technetium-99m (^99mTc) stannous phytate for healthy-liver delineation was proposed. This study assessed the accuracy of automatic healthy-liver segmentation using ^99mTc images derived from a phantom experiment. In addition, together with data from a patient study, the effect of different ^99mTc activities on the ¹⁶⁶Ho-scout images was investigated. To reproduce a typical scout procedure, the liver compartment, including two tumors, of an anthropomorphic phantom was filled with 250 MBq of ¹⁶⁶Ho-chloride, with a tumor to non-tumorous liver activity concentration ratio of 10. Eight SPECT/CT scans were acquired, with varying levels of ^99mTc added to the non-tumorous liver compartment (ranging from 25 to 126 MBq). For comparison, forty-two scans were performed in presence of only ^99mTc from 8 to 240 MBq. ^99mTc image quality was assessed by cold-sphere (tumor) contrast recovery coefficients. Automatic healthy-liver segmentation, obtained by thresholding ^99mTc images, was evaluated by recovered volume and Sørensen–Dice index. The impact of ^99mTc on ¹⁶⁶Ho images and the role of the downscatter correction were evaluated on phantom scans and twenty-six patients’ scans by considering the reconstructed ¹⁶⁶Ho count density in the healthy-liver.

Results

All ^99mTc image reconstructions were found to be independent of the ¹⁶⁶Ho activity present during the acquisition. In addition, cold-sphere contrast recovery coefficients were independent of ^99mTc activity. The segmented healthy-liver volume was recovered fully, independent of ^99mTc activity as well. The reconstructed ¹⁶⁶Ho count density was not influenced by ^99mTc activity, as long as an adequate downscatter correction was applied.

Conclusion

The ^99mTc image reconstructions of the phantom scans all performed equally well for the purpose of automatic healthy-liver segmentation, for activities down to 8 MBq. Furthermore, ^99mTc could be injected up to at least 126 MBq without compromising ¹⁶⁶Ho image quality.

Clinical trials The clinical study mentioned is registered with Clinicaltrials.gov (NCT02067988) on February 20, 2014.

Background

Holmium-166 (¹⁶⁶Ho) radioembolization is an established treatment for liver malignancies [1]. The current clinical practice for ¹⁶⁶Ho radioembolization includes pre-treatment planning and treatment, together with post-treatment verification. The pre-treatment phase can be performed using QuiremScout™ (Quirem Medical BV, Deventer, NL), particles which are shaped identical to the microspheres used for the treatment, ¹⁶⁶Ho-microspheres (QuiremSpheres™, Quirem Medical BV, Deventer, NL). Using the ¹⁶⁶Ho-microspheres for both procedures (pre-treatment and treatment) has the benefit of improving the intrahepatic distribution prediction in comparison with current clinical standard (technetium-99m macroaggregated albumin or ^99mTc-MAA) [2]. The activity distribution imaged in this phase serves as a predictor for the radiation dose distribution during the treatment and can be used to avert a potential extrahepatic deposition. In addition, it enables partition modeling [3] which allows a personalized activity calculation for ¹⁶⁶Ho radioembolization. However, this requires segmentation of tumors and non-tumorous tissue on anatomical images, which is typically performed using a complementing contrast enhanced CT, usually acquired up to weeks before the treatment. Consequently, these segmented volumes of interest (VOIs), tumors and non-tumorous area, should be registered to the SPECT/CT to perform dosimetry. A similar workflow applies to the post-treatment phase, to assess treatment outcome and response. So far, manual segmentation and manual image registration are currently most commonly applied in clinical practice. These manual processes are time-consuming, prone to error and introduce inter-observer variability. Therefore, a protocol to automatically segment and register the VOIs would tackle these drawbacks, allowing an automatic workflow for planning and evaluation of the treatment. To this purpose, a dual-isotope protocol was suggested by Lam et al. [4]. It is based on ¹⁶⁶Ho microspheres, which serve as treatment simulation, and technetium-99m (^99mTc) stannous phytate, which accumulates in the healthy-liver tissue and provides a healthy-liver demarcation (¹⁶⁶Ho–^99mTc dual-isotope protocol). These two compounds can be imaged simultaneously with a single SPECT acquisition and then be reconstructed into two images: ¹⁶⁶Ho and ^99mTc, avoiding any registration procedure.

However, the presence of the two radionuclides leads to a reciprocal influence between the two (depicted in Fig. 1), which has to be taken into account during the image reconstruction phase. In particular, the ¹⁶⁶Ho main photopeak, at 81 keV, is affected by the downscatter from ^99mTc, which has its main photopeak at 140 keV. Vice versa, ^99mTc is contaminated by downscatter from the ¹⁶⁶Ho high-energy gamma emissions and bremsstrahlung.

Fig. 1

Recorded ¹⁶⁶Ho–^99mTc spectra. Spectra recorded by the SPECT scanner of an anthropomorphic phantom filled with ¹⁶⁶Ho and ^99mTc. Curves are depicted for ¹⁶⁶Ho:^99mTc activity combinations (in MBq) of: 282:0 (solid line), 268:25 (dashed line), and 250:125 (dotted line). The curves were scaled such that the tails of the spectra overlap for visual comparison. The relative difference between the spectra is attributed to ^99mTc. Orange and blue areas represent the ¹⁶⁶Ho and ^99mTc photopeak windows, respectively, while the gray areas depict the scatter windows (see Table 2)

The ¹⁶⁶Ho–^99mTc dual-isotope protocol is currently implemented in our institute using 50 MBq of ^99mTc administered right after 250 MBq ¹⁶⁶Ho scout, leading to a ¹⁶⁶Ho:^99mTc activity ratio of 5. The decision to use this particular ratio was based on the results of a phantom study together with visual interpretation and consensus reading by two nuclear medicine physicians and a medical physicist [5]. Within the clinical practice, ¹⁶⁶Ho–^99mTc dual-isotope scans are reconstructed using 3D OSEM algorithm (Flash 3D; Siemens). The impact of this protocol on ¹⁶⁶Ho image quality was investigated through a comparison between patient scans, acquired before and after the additional ^99mTc injection [6].

This current study expands on the previous work by van Rooij et al. [5] who used a Monte Carlo-based reconstruction method to correct for the crosstalk interactions between the two isotopes. However, because this reconstruction method is not available for everyday clinical practice, the current study focuses on the applicability of the ¹⁶⁶Ho–^99mTc dual-isotope protocol in the clinical scenario, using commercially available reconstruction software. The impact of different ^99mTc activities on the ¹⁶⁶Ho-scout image quality and, vice versa, the impact of ¹⁶⁶Ho scout on ^99mTc image quality and the accuracy of healthy-liver segmentation were investigated using scans reconstructed according to the clinical protocol. To understand the impact of ^99mTc on ¹⁶⁶Ho image quality and its associated effect on downscatter correction during image reconstruction, ¹⁶⁶Ho image quality has been assessed by multiple metrics. Image uniformity, contrast recovery coefficients and ¹⁶⁶Ho count density, i.e., the voxel value in ¹⁶⁶Ho reconstructions, were measured for various amounts of ^99mTc activity, investigating their dependency on the k-factor used for ¹⁶⁶Ho image reconstruction. This will make the ¹⁶⁶Ho–^99mTc dual-isotope protocol more suitable for its adoption in ¹⁶⁶Ho radioembolization practice, independent of the treatment institution.

Materials and methods

To investigate ^99mTc image reconstructions for the purpose of healthy-liver segmentation during a ¹⁶⁶Ho-scout procedure, an anthropomorphic phantom using various concentrations of ^99mTc in the healthy-liver compartment in presence or not of ¹⁶⁶Ho was used. To mimic patient tumor/liver uptake, ¹⁶⁶Ho was injected in the healthy-liver compartment. Two ¹⁶⁶Ho-filled spheres with a higher concentration resembling tumors with a high uptake were placed within the healthy-liver compartment. ^99mTc was added only to the healthy-liver compartment and not to the tumors, simulating the expected distribution of ^99mTc stannous phytate. For the ¹⁶⁶Ho–^99mTc dual-isotope study, the ¹⁶⁶Ho activity was held constant at 250 MBq (as per scout prescription), while ^99mTc was varied across a range of activities.

^99mTc image quality for varying levels of ^99mTc activity was evaluated though several metrics: image uniformity, cold-sphere contrast recovery and the accuracy of healthy-liver segmentations (obtained by thresholding). For comparison, these acquisitions were repeated without ¹⁶⁶Ho in the phantom in order to investigate the influence of ¹⁶⁶Ho on ^99mTc reconstructions.

The effect that the addition of ^99mTc has on ¹⁶⁶Ho images was investigated by comparing ¹⁶⁶Ho reconstructions from similar scans, performed with various amounts of ^99mTc. Specifically, ¹⁶⁶Ho count densities (the voxel values in the reconstructions) in the healthy-liver compartment were compared between reconstructions. To this end, phantom measurements and multiple patient scans were analyzed.

Details regarding the measurements and the corresponding metrics are summarized in Table1

Table 1 Summary of the experiments

Phantom

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 characteristics

Phantom 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:

$${S}_{E}= \left(\frac{{C}_{L}}{{W}_{L}}+\frac{{C}_{U}}{{W}_{U}}\right)\times \frac{1}{2}\times {W}_{PP}=\left(\frac{{W}_{PP}}{{W}_{L}\times 2}\right)\times {C}_{L}+\left(\frac{{W}_{PP}}{{W}_{U}\times 2}\right)\times {C}_{U}$$

where ${C}_{L}$ and ${C}_{U}$ are the recorded projections for the lower (Scatter_{_118}) and upper scatter (Scatter_{_170}) windows, respectively, and ${W}_{L}$, ${W}_{U}$ and ${W}_{PP}$ 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 analysis

VOI 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 (${Q}_{H}$ or ${Q}_{C},$ respectively), generally defined as:

$$Q = \frac{{C_{S} /C_{B} - 1}}{{(R - 1)}} \times 100\%$$

where ${C}_{S}$ is the mean intensity measured in the sphere VOI, ${C}_{B}$ 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 (${Q}_{H}$ or ${Q}_{C},$ 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).

Results

Phantom

Uniformity

The ^99mTc image uniformity, measured as the coefficient of variation within the healthy-liver compartment, is displayed in Fig. 2. The COV decreased with the square root of ^99mTc activity, but was independent on the presence of 250 MBq of ¹⁶⁶Ho in the phantom. No statistically significant difference was found between ^99mTc images acquired in presence or not of ¹⁶⁶Ho.

Fig. 2

Coefficient of variation computed on the ^99mTc images within the healthy-liver compartment VOI. The coefficient of variation, measuring the level of inhomogeneity in the healthy-liver compartment VOI, is displayed as function of effective ^99mTc activity in the phantom. Blue circles refer to measurements acquired with ^99mTc only, while green triangles represent measurements acquired with ^99mTc in presence of ¹⁶⁶Ho (dual isotope). The COV decreases with the square root of ^99mTc activity, down to an offset which is assumed be present due to limited imaging resolution (independent of ^99mTc activity). The gray dashed line depicts a fit to the data, where the offset was found to be 3.8%

The ^99mTc image uniformity, measured as the coefficient of variation within the healthy-liver compartment, is displayed in Fig. 2. The COV decreased with the square root of ^99mTc activity, converging to a minimum of approximately 3.8% (likely limited by the imaging resolution), but was independent on the presence of 250 MBq of ¹⁶⁶Ho in the phantom. No statistically significant difference was found between ^99mTc images acquired in presence or not of ¹⁶⁶Ho.

Contrast recovery of cold spheres

Cold-sphere contrast recovery coefficients as function of effective ^99mTc activity in the phantom are reported in Fig. 3A, B, for sphere S1 and sphere S2, respectively. Mean ± standard deviation of ${Q}_{C}$ was 65.7% ± 1.6% for sphere S1 and 57.8% ± 2.1% for sphere S2. At low ^99mTc activities (< 50 MBq), a higher spread in ${Q}_{C}$ (Fig. 3) was visually noticeable. No statistically significant difference was found between ^99mTc images acquired either in presence or not of ¹⁶⁶Ho.

Fig. 3

Contrast recovery coefficient of cold spheres (${Q}_{c}$) as function of total effective ^99mTc activity. ${Q}_{c}$ computed on ^99mTc images acquired in presence of different ^99mTc activities. Results are displayed as function of the total effective ^99mTc activity in the phantom. Panel A refers to cold sphere S1 and panel B to cold sphere S2. Blue circles refer to measurements acquired in presence of ^99mTc only, while green triangles represent measurements acquired in presence of both ^99mTc and ¹⁶⁶Ho

Usability of ^99mTc images for automatic healthy-liver segmentation

The procedure to determine the optimal threshold percentage α was repeated for a range of Gaussian filter sizes, where a width (sigma) of 15 mm was found to produce the smallest variance among segmented volumes. For this filter size, the averaged optimal threshold percentage α adopted for the automatic segmentation process was 40%.

Figure 4 shows the axial view of six phantom SPECT/CT scans with the healthy-liver contour overlapped resulting from the automatic segmentation process.

Fig. 4

Axial view of ^99mTc reconstructions and healthy-liver segmentations. Healthy-liver segmentations of ^99mTc SPECT/CTs, acquired in presence of 250 MBq¹⁶⁶Ho (top panel) and without ¹⁶⁶Ho (bottom panel). Three ^99mTc activity levels are presented to indicate the effect on image quality and segmentation accuracy

Results for the volume recovery percentage and Sørensen–Dice index as function of the effective ^99mTc activity are depicted in Fig. 5A, D which refer to cold sphere S1, B and E to cold sphere S2 and panel C and F to the healthy-liver compartment. Mean ± standard deviation for the volume recovery percentage was 88.8% ± 5.3%, 118.2% ± 12.0% and 99.6% ± 3.8% for cold sphere S1, S2 and healthy-liver compartment, respectively. Mean ± standard deviation for the Sørensen–Dice index was 0.79 ± 0.02, 0.58 ± 0.03 and 0.93 ± 0.01 for cold sphere S1, S2 and healthy-liver compartment, respectively. No statistically significant difference was found between ^99mTc images acquired either in presence or not of ¹⁶⁶Ho.

Fig. 5

Nominal volume recovery and Sørensen–Dice index obtained from segmentation of the ^99mTc images using a threshold approach. On the left, the percentage of recovered nominal volume using the threshold approach for the three VOIs under investigation is shown: sphere S1 (panel A), sphere S2 (panel B) and healthy-liver compartment (panel C). Blue circles refer to measurements acquired in presence of ^99mTc only, while green triangles represent measurements acquired in presence of both ^99mTc and ¹⁶⁶Ho. The results are displayed as function of the total effective ^99mTc activity in the phantom. On the right (panel D, E and F), the corresponding Sørensen–Dice index is shown

Impact of k-factor on ¹⁶⁶Ho phantom reconstructions

The coefficient of variation and the contrast recovery coefficients for the hot spheres, as function of the effective ^99mTc activity in the phantom, are depicted in Fig. 6 for several k-factors.

Fig. 6

Quality analysis for the ¹⁶⁶Ho images, reconstructed using several k-factors, as function of the ^99mTc activity in the phantom during the ¹⁶⁶Ho–^99mTc dual-isotope acquisition. Coefficient of variation and contrast recovery coefficients for hot sphere S1 and S2 are depicted as function of effective ^99mTc activity in the phantom, for ¹⁶⁶Ho reconstructions using three different k-factors. Panel A shows the coefficient of variation, measuring the level of inhomogeneity, computed on the ¹⁶⁶Ho images within the healthy-liver compartment VOI. Panels B and C depict the contrast recovery coefficients for sphere S1 and S2, respectively. Ideally, these image-quality metrics for ¹⁶⁶Ho are independent of ^99mTc activity if scatter from ^99mTc is sufficiently corrected for. The impact of the choice of k-factor can be observed by the dependence on ^99mTc activity, causing an offset of counts in the phantom background. The depicted k-factors are the lowest, the highest and the k-factor leading to the lowest dependency on ^99mTc for the considered metric

A poor choice for the k-factor will under- or over-correct ^99mTc scatter in the ¹⁶⁶Ho image, causing an increase or decrease in apparent ¹⁶⁶Ho signal in the phantom, approximately linear with ^99mTc activity. This dependency is illustrated in Fig. 6 where results are shown for the lowest and the highest k-factor. However, tuning the k-factor can reduce the impact of ^99mTc, as shown by the dashed lines in Fig. 6.

The percentage error in ¹⁶⁶Ho count density, as a result of adding ^99mTc (evaluated at 50 MBq for consistency with the current clinical protocol), is plotted as a function of k-factor in Fig. 7 (blue crosses) along with the results from patient scans. It ranged from 14.3 to − 18.6%. The k-factor for which the ¹⁶⁶Ho count density in the phantom was least dependent of ^99mTc activity was 0.95.

Fig. 7

Dependence of ^99mTc scatter correction on ¹⁶⁶Ho reconstructions. Scatter contributions from ^99mTc on ¹⁶⁶Ho image reconstructions are corrected for using a dual-energy-window scatter correction. The scatter is estimated from the 118-keV energy window, positioned in between the ¹⁶⁶Ho and ^99mTc photopeak, scaled by a k-factor. This graph shows the measured count density difference in ¹⁶⁶ Ho reconstructions from ¹⁶⁶Ho to ^99mTc dual-isotope acquisitions relative to ¹⁶⁶Ho-only acquisitions, as function of the k-factor. Each boxplot represents the results for the 26 patient procedures. Count density was evaluated within the healthy-liver volume. Outlier values were due to a higher-than-average ¹⁶⁶Ho concentration in the liver (due to either higher activity or relatively small liver volume). The blue crosses show the equivalent metric for the phantom study, where the count density difference was evaluated for a ^99mTc activity of 50 MBq

Patient data

Impact of k-factor on ¹⁶⁶Ho patient reconstructions

For the patient scans, the count density in ¹⁶⁶Ho reconstructions was measured in the healthy-liver volume, for both ¹⁶⁶Ho-only scans and ¹⁶⁶Ho dual-isotope scans. The percentage difference between these measurements, attributed to downscatter from the additional ^99mTc in the liver, is plotted in Fig. 7 as a function of k-factor. For every k-factor, the data of the 26 patient procedures are summarized in a boxplot. A k-factor of 1.05 resulted in the smallest overall impact of ^99mTc on ¹⁶⁶Ho image reconstructions with a median percentage difference (and interquartile range) of 1.7% (6.4%), ranging from -12.9% to 13.3% (Tables 1, 2).

Discussions

Accurate image segmentation and registration are paramount to provide personalized dosimetry to radioembolization patients. However, these tasks are time-consuming and user-dependent, being currently performed manually.

The registration step could be avoided by adding a contrast enhanced CT acquisition to the SPECT/CT procedure. However, some limitations prevent this approach to be used in clinical practice. Typically, a contrast enhanced CT is already acquired prior to the scout procedure. Additionally, many SPECT/CT systems lack a contrast injector. Moreover, unless an algorithm for liver and tumor delineation on CT is clinically available, this approach would still require the manual segmentation of the volumes of interest.

The ¹⁶⁶Ho-^99mTc dual-isotope protocol is a viable option to provide an automatic segmentation of the healthy-liver within the ¹⁶⁶Ho radioembolization context. This study demonstrates that the reciprocal interaction between the two isotopes can mostly be negated, provided a proper k-factor for downscatter correction is applied for ¹⁶⁶Ho reconstructions and a triple-energy-window scatter correction is applied for ^99mTc reconstructions.

^99mTc reconstructions

From the measurements performed using an anthropomorphic phantom, it was shown that ^99mTc image uniformity improved only little for higher ^99mTc activities (> 100 MBq), but that it was independent of ¹⁶⁶Ho activity in the phantom. On average, the contrast recovery coefficients of the cold spheres were not dependent on ^99mTc activity, regardless of the presence of ¹⁶⁶Ho. Nonetheless at lower ^99mTc activities (< 50 MBq) a higher spread for ${Q}_{C}$ was found (Fig. 2). The partial contrast recovery can be attributed to spill-in from activity in the healthy-liver compartment due to the limited SPECT system resolution.

Within the investigated range, the amount of ^99mTc activity did not have a significant effect on the accuracy of the healthy-liver segmentations. Using the same method for each segmentation, irrespective of ^99mTc activity or in presence or not of ¹⁶⁶Ho, the healthy-liver compartment was recovered to 100% of its nominal volume, with a Sørensen–Dice index > 0.9. However, the cold spheres within the liver (representing tumors), were segmented to a lower accuracy as was evident from the reduced Sørensen–Dice indices.

¹⁶⁶Ho reconstructions

For phantom images acquired at an effective ¹⁶⁶Ho activity of 250 MBq in presence of ^99mTc, a major role was played by the choice of k-factor used for scatter correction, particularly for increasing ^99mTc activities. When reconstructed with adequate k-factors, the coefficient of variation, contrast recovery coefficients for the hot spheres and count density in the ¹⁶⁶Ho main photopeak were, on average, not dependent on effective ^99mTc activity in the phantom. However, there is no single k-factor which is optimal to fully compensate for scatter across the whole ¹⁶⁶Ho image. This emphasizes how the application of window-based scatter correction is only an approximation to correct a complex, nonlinear, phenomenon.

Patient scans showed the same k-factor dependence as demonstrated in the phantom study. Earlier work resulted in a k-factor of 1.15, which was based on simulations for ¹⁶⁶Ho (by ignoring the 81-keV emission line) and by comparing counts in the projection windows for ^99mTc. In the current work, however, the effect of the k-factor used to correct for ^99mTc downscatter was measured directly on the resulting ¹⁶⁶Ho reconstructions by considering the count density in the healthy-liver compartment (similar to patient dosimetry) depending on the presence of ^99mTc. A combined analysis of all patient and phantom scans (shown in Fig. 7) indicated that a k-factor of 1.05 resulted in the lowest impact of ^99mTc on ¹⁶⁶Ho reconstructions overall.

The direct effect of inadequate scatter correction on ¹⁶⁶Ho image reconstruction is an over- or under estimation of the count density, primarily in the healthy-liver compartment (due to the presence of ^99mTc stannous phytate). In clinical practice, dosimetry is often performed by first scaling the ¹⁶⁶Ho scout image such that the total of counts in the image (including tumor uptake and extrahepatic depositions) corresponds to the planned therapeutic activity. Therefore, due to this scaling procedure, the impact of scatter on healthy-liver dosimetry is dependent on the relative activity distribution amongst the compartments. The measured percentage in ¹⁶⁶Ho count-density difference in the healthy-liver serves as an upper bound to the error in healthy-liver dosimetry, i.e., when all ¹⁶⁶Ho activity resides within the healthy-liver compartment, the count-density offset due to poor scatter correction is negated as a result of the scaling process.

Limitations and future developments

The presented value for the optimal k-factor is specific for the energy windows used in this work and includes the weight factor to account for the difference in window widths between the 81-keV and 118-keV windows. When different energy window settings are applied, care has to be taken to properly adapt, or re-evaluate, the required k-factor.

Whereas the phantom was filled with uniform activity, a more heterogeneous ^99mTc activity distribution will be encountered in patients. This may cause a relatively low ^99mTc activity in the healthy tissue to be classified as non-healthy tissue. Thus, the possibility to automatically segment the healthy-liver based on a threshold approach using ^99mTc images needs to be validated in clinical practice by comparing against manual segmentations of contrast enhanced CT images. In particular, the impact of adopting this method rather than the current manual approach for segmentation and registration has to be investigated with respect to dosimetry, time required to perform these tasks and inter-observer influence.

Conclusion

Within the context of the ¹⁶⁶Ho-^99mTc dual-isotope protocol, the impact of ^99mTc on reconstructed ¹⁶⁶Ho count density in the healthy-liver could mostly be negated, provided an adequate k-factor for downscatter correction was applied during image reconstruction. The healthy-liver compartment of the phantom could accurately be segmented on the ^99mTc images using a thresholding method, irrespective of the amount of ^99mTc activity or in presence or not of ¹⁶⁶Ho, and the healthy-liver compartment was recovered to 100% of its nominal volume (Sørensen–Dice index > 0.9).

Because additional scatter due to the presence of ^99mTc can effectively be corrected, the dual-isotope protocol can safely be applied without compromising ¹⁶⁶Ho image quality. However, validation of the automatic segmentation method, and its effect on dosimetry, needs to be assessed in clinical practice.