3D Monte Carlo bone marrow dosimetry for Lu-177-PSMA therapy with guidance of non-invasive 3D localization of active bone marrow via Tc-99m-anti-granulocyte antibody SPECT/CT

In radioligand therapy, dosimetry is recommended for appropriate treatment planning and aims for maximizing the absorbed dose to malignant structures, while minimizing the absorbed dose to risk organs. A variety of strategies for the assessment of tissue absorbed doses exist, which can be distinguished depending on how detailed the patient-specific information is considered or depending on whether a 3D-absorbed dose model is used or absorbed dose factors (S values) are applied [1‐5]. Monte Carlo techniques can be used to fully simulate all interactions of radioactive decay particles within the surrounding material in a step-by-step manner. Various Monte Carlo codes such as FLUKA or GEANT4 were extended to applications in nuclear medicine and are capable to consider the patient-specific 3D activity and anatomical characteristics via inclusion of the SPECT, PET, and CT data into the simulation [6‐12]. Thus, 3D-absorbed dose distributions with resolution and accuracy depending on the input image data can be provided. However, the absorbed dose to risk organs or tumors during radioligand therapy is usually calculated via organ-level S values, which were pre-calculated based on standardized anthropomorphic phantoms and which estimate the mean absorbed dose to the whole target region based on the mean time-integrated activity in a specified source region.

The active bone marrow (BM) is a main organ at risk during Lu-177-PSMA ligand therapy, especially as patients with advanced mCRPC often present with a high bone tumor burden and a potentially reduced hematological function [13‐15]. Typically, bone marrow dosimetry is performed by applying the aforementioned S values and accounts for the self-absorbed dose to the bone marrow from the blood, the cross-absorbed dose from the remainder of the body (ROB), and the cross-absorbed dose from major organs and tumors as specific source regions [15‐19]. However, during bone marrow dosimetry using organ-level S values, simplifying assumptions have to be made to estimate the absorbed dose from the overall tumor distribution, as no pre-calculated S values exist that consider all lesions in their size, shape, number, and location. Thus, Monte Carlo simulations may lead to improved bone marrow absorbed dose estimates, as they have the potential to fully account for the patient-specific 3D disease characteristics.

Another limitation of classical bone marrow dosimetry is that the actual localization of the active bone marrow is a priori unknown. Bone lesions might lead to a displacement of active bone marrow from the direct site of metastases and thus activity accumulation, which would drastically reduce the absorbed dose to the active bone marrow [20, 21]. However, bone marrow dosimetry using pre-calculated organ-level S values assumes a physiological bone marrow distribution [16]. Clinical imaging methods, such as magnetic resonance imaging or Tc-99m-anti-granulocyte antibody scintigraphy, can be used for non-invasive active bone marrow localization, within the spatial resolution of the corresponding imaging modality [22‐25]. Thus, such techniques might overcome the limitation of an a priori unknown target region for bone marrow dosimetry.

If not indicated otherwise, the term bone marrow always refers to the active bone marrow, which represents the radio-sensitive part of the overall bone marrow mixture.

The bone marrow is potentially the most critical organ and most limiting factor of the therapeutic window during Lu-177-PSMA therapy of mCRPC patients, as those patients are usually heavily pre-treated and often present with a considerable bone lesion load [14]. In a study of Rahbar et al. with 145 patients and an average administered activity of 5.9 GBq per cycle, hematoxicity showed the highest incidence for all grades as well as for grade 3–4 events [36]. Thus, bone marrow dosimetry is highly recommended in those patients. However, for bone marrow dosimetry to become predictive for hematoxicity, all relevant patient-specific parameters must be considered. These include the patient-specific activity accumulation over time, the anatomical characteristics, an appropriate localization of the bone marrow target region, and, not least, pre-therapies and the patient-specific disease and hematological baseline status.

With respect to the absolute values, patient-specific Monte-Carlo-based calculations resulted in higher bone marrow absorbed doses than the classical S value approach. For both models, MC2 and SMIRD, bone marrow absorbed dose estimates were well below the typically applied threshold of 2 Gy [19]. Although for MC2 the absorbed dose estimates were on median fourfold higher than those achieved with the reference model SMIRD, the deduction that multiple therapy cycles are applicable for these patients without the risk of severe marrow toxicities seems reasonable, at least with regard to the current dose limit. However, the assumption of full displacement of active bone marrow, as included in model MC2, might not be a priori justified in each patient and might lead to an underestimation of the bone marrow absorbed dose. For model MC1, assuming a physiological and unaltered active bone marrow distribution, median absorbed doses of 130 mGy/GBq were found with on median 17-fold and 3-fold higher absorbed dose estimates compared to SMIRD and MC2, respectively. According to dosimetry using model MC1, patients 4, 5, 7, and 8 would have received a bone marrow absorbed dose close to or in some cases even strongly exceeding the 2 Gy limit, with the consequent risk for severe marrow toxicities in these patients. However, none of the patients presented severe marrow damage, even after multiple therapy cycles. This observation questions the general applicability and significance of models like MC1, which do not account for bone marrow displacement by tumor lesions, especially in the presence of a heavy skeletal tumor burden.

Visual analysis of Tc-99m-anti-granulocyte antibody SPECT scans supports this thesis, as it indicates a displacement of active bone marrow from the direct location of bone lesions for both exemplarily investigated patients. In these patients, the re-analysis of bone marrow absorbed dose estimates using the individual Tc-99m-anti-granulocyte-based active bone marrow VOIs resulted in a clear absorbed dose reduction of approximately 40% compared to MC1, although the absorbed dose values were still higher than those from MC2. However, both the Ga-68-PSMA-11 PET and the Tc-99m-anti-granulocyte antibody SPECT have a finite resolution, which intrinsically results in a certain artificial overlap of both activity distributions and thus in an increased absorbed dose estimate compared to MC2. Furthermore, co-registration between both modalities is in general not perfect, particularly for challenging regions such as the ribs or the sternum, which might additionally cause a local overlap between tumor and bone marrow accumulation. Both finite spatial resolution and imperfect co-registration interfere with a potential incomplete active bone marrow displacement. Despite the additional complexity introduced by the finite resolution of the involved imaging systems and by the imperfect co-registration, the utilization of the additional image data from the Tc-99m-anti-granulocyte antibody scintigraphy may improve individualized bone marrow dosimetry by providing a realistic upper limit for the bone marrow absorbed dose. For patient 8, the usage of a Tc-99m-anti-granulocyte-based VOI resulted in a reduction of a total bone marrow absorbed dose from 1.1 Gy (MC1) to an expected maximum dose of 0.7 Gy. It may be hypothesized that the typically applied upper limit for the bone marrow absorbed dose of 2 Gy is too high for patients with advanced cancer disease, extensive skeletal tumor burden, and potentially decreased hematological function due to various pre-therapies. Despite the relatively small additional bone marrow absorbed dose (approx. 10 mSv for 400 MBq), Tc-99m-anti-granulocyte antibody scintigraphy might be justified in such patients. Studies based on a larger patient cohort are needed to identify, whether image-based active bone marrow localization in combination with Monte-Carlo-based absorbed dose calculation really improves the correlation between bone marrow dosimetry and hematoxicities, and may therefore be suitable to guide therapy planning in future workflows. Still, the comparison between models MC1 and MC2 demonstrates that the a priori unknown patient-specific active bone marrow distribution results in a large uncertainty of the bone marrow absorbed doses, even if Monte Carlo techniques are applied for absorbed dose modeling. Further, future studies should be performed to find an appropriate threshold for the bone marrow absorbed dose for Lu-177-PSMA therapy [14].

The resolution of non-invasive active bone marrow localization could be further improved by MRI bone marrow localization, which would also be beneficial if even small additional contributions to the bone marrow absorbed dose must be avoided [24, 25]. However, in clinical daily routine, the localization of active bone marrow in a large part of the patient body should be feasible with acceptable measurement and processing time, and it should be available for several patients per week. Further, a reduction of processing complexity and effort accompanied by a potential enhancement of the accuracy of the derived information is desirable. This could be achieved, for instance, by employing a standardized method for patient positioning over multiple scans at the same or at different imaging modalities via patient-adaptable storage matrasses, which facilitates image co-registration [37].

To derive a patient-specific activity template from the Ga-68-PSMA-11 PET instead from the Lu-177-SPECT reduces issues of spatial resolution during absorbed dose calculation. Further, for the patients considered in this study, only SPECT acquisitions of the abdomen were available, while for bone marrow dosimetry, the lesion distribution in the overall body is important. Using the Ga-68-PSMA-11 PET to model the 3D activity accumulation during therapy is an approximation and only applicable if the delay between PET acquisition and therapy is small (on average 2.5 weeks in this study) and if there is no change in the overall lesion load. To switch to a fully Lu-177-based activity template, further investigations are desired to improve the spatial resolution of the Lu-177 imaging and to enable a fast whole-body Lu-177-SPECT acquisition. Filling a fixed patient-specific activity VOI template with the respective segmented VOI activities from sequential Lu-177 imaging reduces co-registration errors, which could otherwise lead to an artificially increased bone marrow absorbed dose.

The analysis of blood parameters revealed a significant (p < 0.05) and strong negative correlation only for the change of platelet counts, irrespective of the exact bone marrow model used. The highest correlation was obtained with model MC2, which includes the assumption of full displacement of active bone marrow from the direct site of the bone lesions. To exploit the potential and impact of bone marrow dosimetry for therapy planning, a more comprehensive investigation of the correlation of the change of blood element counts with bone marrow absorbed dose estimates is desired. For this purpose, blood analysis should consider a higher number of patients and a prolonged time period. Both the baseline hematological status and its course after therapy are known to be affected by various parameters, such as pre-therapies or total lesion volume [17, 38‐40]. Thus, a patient stratification as for example proposed by Walrand et al. is mandatory, if the correlation between bone marrow absorbed doses and hematological response to therapy shall be analyzed [40]. So far, the correlation of bone marrow absorbed doses and blood parameters for different Lu-177-based radioligand therapies was assessed using S value-based methods. Svensson et al. observed moderate and significant correlations for the decrease of hemoglobin level, total white blood cells, and platelet counts for Lu-177 PRRT and for 46 investigated patients [17]. By contrast, Forrer et al. found no correlation between the decrease of platelet counts and bone marrow absorbed dose estimates for Lu-177 PRRT based on 15 patients and monitoring of hematological function until 6 weeks after treatment [18].

In clinical routine, the application of Monte Carlo simulations for dose calculations may be too time-consuming, especially if the simulation has to consider a large part of the patient body and small voxels (e.g., 0.05–0.001 ccm). However, Monte Carlo simulations can be made feasible with computing clusters. In this way, results with a high statistical validity could be obtained in 1 day, which is acceptable with respect to the time gap between successive cycles of radioligand therapy. An intermediate method for fast 3D dosimetry within minutes, which compromises the consideration of Monte Carlo techniques and computational effectiveness, is the application of Monte-Carlo-based dose kernels [41, 42]. Further investigations of this approach may be advisable to prospectively facilitate improved clinical dosimetry for monitoring and planning of radioligand therapies.

In this work, we introduced a weighting-based model to represent the different compartments of active and inactive bone marrow and hard bone. This model represents a reasonable simplification for the application to clinical routine imaging data, which have a spatial resolution above the characteristic size of the bone marrow microstructure. Hybrid Monte-Carlo-based models that alternate between macroscopic models of the overall patient anatomy and detailed microscopic models of the skeletal system are time-consuming but might further improve bone marrow absorbed dose estimates, at least for the understanding of important mechanisms to assess risk factors for marrow toxicities [34, 43].

Monte-Carlo-based bone marrow absorbed doses were found to be significantly increased compared to those derived from classical S value dosimetry. Particularly, a large spread between Monte-Carlo-based and S value bone marrow absorbed doses was observed, which implies a large uncertainty, especially for S value dosimetry due to the lack of an appropriate consideration of the patient-specific highly heterogeneous 3D lesion distribution. However, even for Monte-Carlo-based bone marrow dosimetry, the a priori unknown patient-specific active bone marrow distribution produces a large uncertainty of bone marrow absorbed doses. Assuming a co-localization between active bone marrow and all lesions (MC1) is hypothesized to lead to too exaggerated absorbed dose values (> 2 Gy per cycle for 27% of investigated cycles), as these values were not in concordance with the observation of severe hematological toxicities. Simultaneously, the a priori assumption of a full displacement of active bone marrow for each patient (MC2) might underestimate the patient-specific absorbed dose. Patient-specific image-based active marrow localization, as performed for a small subgroup of patients, yielded to intermediate bone marrow absorbed doses compared to MC1 and MC2, although issues of co-registration and finite image resolution might interfere with incomplete active bone marrow displacement. Future studies based on a larger patient cohort are recommended, to particularly determine whether patient-specific active bone marrow localization in combination with Monte-Carlo-based absorbed dose modeling can improve the prediction of hematoxicities and thus enables to exploit the full therapeutic window of Lu-177-PSMA therapy. Preliminary results showed a significant and strong correlation between platelet decrease and bone marrow absorbed doses, irrespective of the exact dosimetry model; however, highest correlation was observed for MC2.