The value of ^99mTc-MAA SPECT/CT for lung shunt estimation in ⁹⁰Y radioembolization: a phantom and patient study

Transarterial radioembolization (TARE) with ⁹⁰Y microspheres is an established treatment for unresectable hepatocellular carcinoma (HCC) and for liver metastases, with promising clinical results [1, 2]. In TARE, radioactive microspheres (glass or resin) are preferentially delivered and permanently implanted into hepatic tumors by exploiting the unique dual vascular anatomy of the liver. A major toxicity concern in TARE is radiation-induced pneumonitis and sclerosis due to hepatopulmonary shunting of ⁹⁰Y microspheres. Prior to treatment, lung shunting is evaluated by surrogate imaging with ^99mTc-macroaggregated albumin particles (^99mTc-MAA) that is assumed to mimic the distribution of the microspheres. For glass microspheres, the manufacturer’s recommendation (https://​www.​btg-im.​com/​BTG/​media/​TheraSphere-Documents/​PDF/​TheraSphere-Package-Insert_​USA_​Rev-14.​pdf) for individualization of the administered activity based on lung shunt uses an upper limit of 30 Gy for the absorbed dose to the lung relying on limited clinical observations [3, 4].

Currently, in standard clinical practice, the lung shunt fraction (LSF) is typically estimated by ^99mTc-MAA planar gamma camera imaging performed without accounting for attenuation or scatter effects. Because of differences in tissue densities of the lung and liver, the LSF will be overestimated when attenuation correction is not performed. Additionally, 2D imaging does not allow for accurate delineation of lung and liver regions and cannot be used to estimate the lung volume or mass for dosimetry. Although ^99mTc-MAA SPECT/CT is performed at many centers, for visual assessment of extra-hepatic deposition, it is not routinely used for the LSF calculation. With iterative reconstruction, image degrading physical factors such as attenuation and scatter can be accurately modeled in SPECT/CT, and tomographic 3D imaging facilitates accurate delineation of lung and liver volumes. There have been past studies that report significant overestimation in planar imaging-based LSFs compared with SPECT/CT-based estimates as well as relatively good agreement in patients undergoing Y-90 TARE [5‐7]. However, these studies did not include comparison with estimates from post-therapy imaging. Additionally, only one of these studies [7] included phantom measurements to access the accuracy and precision of the imaging methodology used. Because post-therapy imaging is not part of manufacturer recommendations for TARE and is not always performed, the agreement between ^99mTc-MAA-based lung shunt estimates and post-therapy imaging-based estimates has not been well studied. Additionally, post-therapy ⁹⁰Y imaging based on either bremsstrahlung SPECT/CT or ⁹⁰Y PET/CT is challenging, especially in low uptake regions such as the lung. The use of time-of-flight information in the case of PET [8] and model/Monte Carlo-based scatter correction in the case SPECT [9‐11] has led to improved Y-90 imaging, but these methods are not always available with clinic systems. Apart from the imaging methodology, another factor that can impact ^99mTc-MAA-based lung shunt estimation is the apparent in vivo breakdown of ^99mTc-MAA into smaller aggregates and to pertechnetate over time [12, 13].

Since February 2017, at our institution, in addition to planar ^99mTc-MAA imaging, SPECT/CT is performed with two bed positions used when the entire lung and liver are not encompassed within a single field-of-view (FOV). Thus, imaging data was available to perform SPECT/CT-based LSF calculation for comparison with the planar imaging-based calculation still used in the clinic. A workflow that uses automatic contouring tools was implemented to facilitate the process of SPECT/CT-based LSF estimation in the current study and with the goal of future clinical implementation. Experimental measurements with a liver/lung torso phantom were designed to mimic typical LSF values and imaging conditions in patients. Phantom results that establish accuracy/precision of planar and SPECT/CT-based LSF estimation used in our clinic are presented together with a retrospective comparison of patient LSF and lung absorbed dose results corresponding to the two modalities. The impact the time delay from ^99mTc-MAA injection to scanning has on LSF as well as on extra hepatic uptake in general is also evaluated. Additionally, some of the patients were enrolled in an ongoing research study where post-therapy ⁹⁰Y PET/CT imaging was performed. Thus, for a subset a comparison of pre-therapy vs. post-therapy imaging-based LSF values was also performed.

In this study, we investigated the potential value of ^99mTc-MAA SPECT/CT imaging compared with planar imaging for lung shunt calculation prior to TARE. The phantom study performed under clinically realistic conditions demonstrated that improved accuracy in LSF estimation (within 13% underestimation of true value) can be achieved using SPECT/CT with AC and SC compared with planar imaging (up to 44% overestimation). Compared with SPECT/CT with AC only, adding SC improved the LSF estimate for the case with the lowest LSF only. For the other cases, the underestimation in LSF with SC is potentially due to inaccuracies in the energy window-based SC. In a previous Monte Carlo simulation study [17], we demonstrated that the TEW method overestimates scatter in the lung region, thereby leading to an underestimation in the reconstructed counts. Other more sophisticated model/Monte Carlo-based methods have been shown to be superior to energy window-based SC [10, 18, 19], but are not always available with clinic systems. For the planar calculation, using the anterior view only for the liver counts and posterior view only for the lung counts, as in our patient studies, led to higher accuracy than using the geometric mean method. The geometric mean method results in higher LSFs because the liver counts go down substantially when the posterior view is included in the calculation while the lung counts are less sensitive to the inclusion of the anterior view. Both planar and SPECT/CT-based estimates had high precision except for the 0% lung shunt case where applying AC and SC to SPECT/CT resulted in high variability due to the very low counts in the lung region (Table 2). Clinically, this will have no impact because of the 0% lung shunt value. The phantom study did not assess respiratory motion effects, or the variability associated with manual segmentation of the liver and lung regions in the planar scans, which are relevant factors in clinical studies.

In patients, in addition to the overestimation in LSF due to lower attenuation in lung tissue relative to liver tissue, resolution effects and respiratory motion artificially increases the value due to liver counts spilling over to the inferior part of the lung. Hence, instead of using the entire lung volume, an exclusion area was used with SPECT/CT and PET/CT, but such a correction is difficult to apply in planar imaging due to the lack of anatomical information. In one case, the SPECT/CT-based LSF was as high as 12.6% without the exclusion region, but reduced to only 2.2% with the exclusion region. The patient results show the same trend as in the phantom study with the SPECT/CT-based LSF decreasing with AC and decreasing further with SC. As in the phantom study, in general, the impact of SC was less than the impact of AC, except at low LSFs. However, we believe SC should be performed when available because improved accuracy was shown in the phantom study at zero LSF and can also improve visibility/contrast of extra-hepatic deposition. The difference between LSFs corresponding to planar imaging and SPECT/CT was generally much higher in patients than in the phantom, which can be attributed to the fact that the phantom was not impacted by motion effects and operator variability in defining regions both of which can significantly impact the planar calculation in patients. The planar-based estimate can have considerable variability associated with the manual segmentation of patient liver and lung regions, which is operator-dependent. The operator variability can be all but eliminated with SPECT/CT-based automatic contouring tools, such as those used in the current workflow. Additionally, in patients, the planar LSFs can be affected by extra-hepatic uptake due to breakdown of ^99mTc-MAA over time because unlike with SPECT, planar imaging does not differentiate between activity originating in the source region and originating in overlying and underlying tissue. There can also be scattering off of the structures that are overlying or underlying the source region (for example, liver counts scattering off of breast tissue that is anterior to the lung will contribute to the planar lung counts, but not to the SPECT VOI counts), which can be mitigated by a simple energy window-based scatter correction. In our study, only photopeak window data was available for the clinic planar acquisitions; hence, scatter correction could not be performed. The relative impact of the various effects can be estimated by comparing the LSFs in Tables 2 and 3 with and without the different corrections and the exclusion zone. Based on these results, the use of attenuation correction and the exclusion zone had the biggest impact on SPECT LSF. We were not able to assess the impact of SPECT resolution recovery because all reconstructions included this (no option to turn resolution recovery off in Flash 3D).

Our patient and phantom results for planar vs. SPECT/CT-based LSF estimates are consistent with the recent report of Dittman et al. [7]. For the phantom study, they reported 40% overestimation in LSF with planar imaging and 5% accuracy with SPECT/CT. As in our study, for patients, they reported a much higher difference between planar and SPECT/CT-based LSF estimates than in the phantom study (for patients, their planar to SPECT/CT LSF ratio was on average 3.6, while ours was 4.3). They reported a mean LSF of 8.3 (range from 3.4 to 32.3%) for planar imaging and 2.9% (range from 0.8 to 15.7%) for quantitative SPECT/CT. The corresponding values from our study are 5.4% (range from 1.2 to 15.7%) and 1.5% (range from 0.4 to 6.0%). Their correlation between planar and SPECT/CT-based LSF estimates (R² = 0.83) was higher than the correlation we observed (R² = 0.46 in Fig. 4), potentially due to using a more systematic approach for planar ROI definition in their prospective study compared with our retrospective use of clinically defined ROIs. Additionally, their planar estimate was based on the geometric mean approach while we used the anterior liver and posterior lung. Their study did not include post-therapy LSF estimates for comparison with ours.

The overestimation in LSF translates to an overestimation in absorbed dose. In one subject, the SPECT/CT-based absorbed dose to the lung was 6.5 Gy (lung shunt 2.5%) while that based on planar imaging was as high as 29.4 Gy (lung shunt 13.6%), which is close to the recommended upper limit for treatment eligibility. The limitation of assuming a “standard” lung mass of 1 kg for all patients in the planar absorbed dose calculation was evident from the individual CT volume-based lung mass for this cohort (average 0.87 kg, range from 0.45 to 1.43) used for the SPECT/CT-based calculation. The maximum tolerable dose to the lung from Y-90 microspheres is not well established. Historically, significant toxicity was associated with a total lung dose of 25 Gy in external beam radiation therapy [20], but this is likely not applicable to TARE due to the non-uniform dose distribution of microspheres. The 30 Gy upper limit recommendation of the manufacturer is based on radiation pneumonitis in two clinical reports where lung absorbed dose was calculated from ^99mTc-MAA planar imaging-based LSFs [3, 4]. The relatively poor correlation observed in the present study between planar and SPECT/CT-based LSFs and absorbed doses (Figs. 4 and 5) makes it difficult to translate these past planar imaging-based findings to SPECT/CT-based limits. Obtaining accurate LSFs and lung tolerability dose estimates from SPECT/CT-based calculation could allow individual adjusted doses that improve therapeutic outcome while minimizing pulmonary fibrosis.

There was no statistically significant difference between mean LSF estimates from ⁹⁰Y imaging and estimates based on pre-therapy ^99mTc-MAA SPECT/CT with AC and SC. As evident in Fig. 7, the absolute difference between Y-90 PET and ^99mTc-MAA SPECT/CT-based estimates are small and will not be clinically important at these low LSFs (the largest difference is for patient 2, where the two LSFs are 2.6 and 1.1%, which translates to lung absorbed doses of 3.6 and 1.5 Gy, respectively). For planar imaging, the differences are larger (the largest difference is for patient 2, where the two LSFs are 2.6 and 12.3%, which translates to lung absorbed doses of 3.6 and 15.6 Gy, respectively). The observations of the current study need to be tested with a larger patient cohort including patients with higher LSFs, which can be difficult because high lung shunting is not typical, and the rare cases may be excluded from treatment. A past study with 14 patients reported that ^99mTc-MAA SPECT/CT imaging does not accurately predict lung absorbed doses after 166-Ho TARE [21]. A potential explanation for the disparity was the difference in ^99mTc-MAA and Ho-166 microsphere distributions due to differences in stability and size of the MAA particles and the microspheres, which is also a concern with Y-90 microspheres.

There was no correlation between ^99mTc-MAA-based LSFs and the time delay between injection and scanning. However, in the current study, most of the patients (29/40) were scanned within 90 min of ^99mTc-MAA administration (Table 4). A study of four patients who underwent repeat planar whole-body imaging up to 5 h after ^99mTc-MAA administration reported marked increase in lung shunt with time (mean lung shunt was 9.3, 14.7, and 22.1% for time intervals of < 1, 1–4, and > 4 h) [13]. In visual assessment of our patient cohort, extra hepatic uptake tended to be more likely in patients with a longer delay between MAA administration and scanning (Table 4). This can be attributed to the breakdown of ^99mTc-MAA over time followed by localization of products in the gastric and renal cortex, which warrant early imaging to improve image quality by minimizing this background activity.

A limitation of the phantom study was that it did not include respiratory motion, which prevented evaluation of the effect of breathing and validation of the use of the lung exclusion region. Motor controlled movable nuclear medicine phantoms have been designed to investigate motion effects [22] and can be considered in a future study. Other limitations of the study include the heterogeneous patient population and the small sample size, especially for post-therapy imaging, where only 20 studies were available as this was not part of the routine clinical procedure.

A phantom study with clinically realistic uptake patterns for the liver and lung demonstrated the improved accuracy of lung shunt estimates based on SPECT/CT with attenuation and scatter corrections over estimates from planar imaging. The patient studies demonstrated the same trend observed in the phantom study with substantially higher lung shunt estimates from planar imaging compared with those determined by SPECT/CT with corrections. There was no statistically significant difference between LSFs from post-therapy ⁹⁰Y PET/CT and pre-therapy ^99mTc-MAA SPECT/CT with attenuation and scatter corrections. There was no correlation between lung shunt estimates and the time delay between MAA injection and scanning, although on visual assessment, there was evidence of increased off-target uptake in extra-hepatic organs with time. In summary, the improved accuracy demonstrated by the phantom study, the agreement with ⁹⁰Y PET/CT and practicality of using auto-contouring for liver/lung definition suggests that ^99mTc-MAA SPECT/CT should be used for lung shunt estimation in TARE.