Personalized dosimetry of ¹⁷⁷Lu-DOTATATE: a comparison of organ- and voxel-level approaches using open-access images

Abstract

¹⁷⁷Lu-DOTATATE (Lutathera^®) enables targeted radionuclide therapy of neuroendocrine tumors expressing somatostatin receptor type 2. Though patient-specific dosimetry estimates may be clinically important for predicting absorbed dose-effect relationships, there are multiple relevant dosimetry paradigms which are distinct in terms of clinical effort, numerical output and added-value. This work compares three different approaches for ¹⁷⁷Lu-DOTATATE dosimetry, including 1) an organ-level approach based on reference phantom MIRD S-values scaled to patient-specific organ masses (MIRDcalc), 2) an organ-level approach based on Monte Carlo simulation in a patient-specific mesh phantoms (PARaDIM), and 3) a 3D approach based on Monte Carlo simulation in patient-specific voxel phantoms.

Method.

Serial quantitative SPECT/CT images for two patients receiving ¹⁷⁷Lu-DOTATATE therapy were obtained from archive in the Deep Blue database. For each patient, the serial CT images were co-registered to the first time point CT using a deformable registration technique aided by virtual landmarks placed in the kidney pelves and the lesion foci. The co-registered SPECT images were integrated voxel-wise to generate time-integrated activity maps. Lesions, kidneys, liver, spleen, lungs, compact bone, spongiosa, and rest of body were segmented at the first imaging time point and overlaid on co-registered integrated activity maps. The resultant segmentation was used for three purposes: 1) to generate patient-specific phantoms, 2) to determine organ-level time-integrated activities, and 3) to generate dose volume histograms from 3D voxel-based calculations.

Results.

Mean absorbed doses were computed for lesions and 48 tissues with MIRDcalc software. Mean organ absorbed doses and dose volume histograms were obtained for lesions and 6 tissues with the voxel Monte Carlo approach. Lesion- and organ-level absorbed dose estimates agreed within ±26% for the lesions and ±13% for the critical organs, among the different methods tested. Overall good agreement was observed with the dosimetry estimates from the NETTER-1 trial.

Conclusions.

For personalized ¹⁷⁷Lu-DOTATATE dosimetry, a combined approach was determined to be valuable, which utilized two dose calculation methods supported by a single image processing workflow. In the absence of quantitative imaging limitations, the voxel Monte Carlo method likely provides valuable information to guide treatment by considering absorbed dose non-uniformity in lesions and organs at risk. The patient-scaled reference phantom method also provides valuable information, including absorbed dose estimates for non-segmented organs, and more accurate dose estimates for complex radiosensitive organs including the active marrow.

1. Introduction

¹⁷⁷Lu-DOTATATE (Lutathera^®) was recently approved by the US FDA for targeted radionuclide therapy (TRT)of somatostatin receptor type 2-positive (SSTR2+) gastroenteropancreatic neuroendocrine tumors. The recommended dosage is 7.4 GBq fixed fractions administered bi-monthly for four total cycles. While currently recommended dosage modifications are based on presentation of adverse reactions, there is mounting evidence that dosimetry-driven patient-specific optimization of administered activity may be beneficial to provide adequate lesion response while minimizing normal tissue toxicities [1-3].

The Society of Nuclear Medicine and Molecular Imaging (SNMMI) dosimetry task force recently identified that a lack of standardization of dosimetry methodology hinders such efforts, with different approaches (e.g. organ-level versus voxel-level) providing potentially variable results and output with varying degrees of information richness. To address this, image sets for patients receiving ¹⁷⁷Lu-DOTATATE therapy were made publicly available through the University of Michigan’s Deep Blue data sharing repository, to encourage community contribution of dosimetry computations for comparative assessment.

In this work, we computed dosimetry estimates for each Deep Blue patient using three unique dosimetry tools developed at Memorial Sloan Kettering Cancer Center. These include: (1) MIRDcalc [4], an organ-level dosimetry code based on the MIRD schema, (2) PARaDIM [5], a Monte Carlo-based application utilizing patient-specific mesh phantoms and organ-level inputs, and (3) an in-house Python program utilizing voxel phantoms and time-integrated activity maps within Monte Carlo simulations (herein referred to as ‘voxel MC’). We demonstrate an image processing workflow that supports all three methods synergistically for value-added output without Requiring additional manual effort. Finally, we compare each method, report the variability observed in calculations, and consider the utility of each method.

2. Method

2.1. Patient image data

Serial quantitative SPECT/CT images for two patients receiving ¹⁷⁷Lu-DOTATATE therapy were obtained from the Deep Blue database. The first patient being male, 64 years of age, and imaged following the second cycle of ¹⁷⁷Lu-DOTATATE treatment (7.21 GBq). Two lesions >2 cm³ were identified for this patient based on radiologist assessment. The second patient being female, 63 years of age, imaged after the first cycle of ¹⁷⁷Lu-DOTATATE treatment (7.31 GBq). For the second patient, four lesions were identified based on radiologist assessment, and the patient has no spleen. Both patients were imaged on an Intevo™ SPECT/CT scanner (Siemens^®) and images reconstructed with xSPECT Quant™.

2.2. Image segmentation, registration, and processing

3D Slicer [6] v4.11, an open-source image analysis software, was used for all image segmentation and registration. The methods we used were based on similar and generally accepted strategies found in literature [7-10]. A summary of our workflow is provided in figure 1; discussed in detail in the proceeding sections.

Figure 1.

Flowchart for derivation of voxel- and organ-level time-integrated activity (TIA) data for dose calculations with different software. Images are of the male patient.

2.2.1. Segmentation

Lesions, organs displaying elevated uptake (kidneys, liver, and spleen), potential dose-limiting tissues (bone marrow), and the rest-of-body were segmented using a combination of manual and semi-automatic methods in 3D Slicer to define source and target regions [11] for dose calculations. The lesions were segmented based on relative activity concentration thresholds (i.e. SPECT-based). A preliminary spherical volume-of-interest (VOI) was first drawn around each lesion providing ample margin for further refinement. The lesion foci were determined by computing the maximal-intensity voxel in each lesion VOI, and this value used to refine each segment by performing a local threshold; 50% of the maximum was used for most lesions, with the exception of two lesions in the female patient (posterior right lobe of the liver, and abdominal lymph node), for which a 70% of maximum threshold was used. For the liver, spleen, and cortex and medulla of the kidneys, the flood filling tool was utilized based on the CT volume, using an intensity tolerance of 25 HU and a neighborhood size of 5—values which we deemed to provide adequate contour accuracy and organ coverage for further refinement. Voxels where the algorithm propagated out of the organ were trimmed with the scissors tool, and holes were subsequently filled using the ‘closing’ smoothing tool. The pelvis of each kidney was delineated manually by expanding the cortex/medulla segment with the sphere paint tool. The lungs were defined by a simple threshold (approx. −1000 to −375 HU) followed by removal of non-lung regions with the scissors tool. The skeleton was defined by a simple threshold (~125 to maximum HU) followed by hole filling. The skeletal spongiosa regions were defined by eroding the skeletal segment by a 4 mm margin followed by manually removing cartilage regions with the scissors tool. Finally, the spongiosa segment was subtracted from the skeleton segment to define the compact bone and cartilage regions. The rest of body (defined only within the field-of-view) was defined using a simple threshold followed by hole-filling using the ‘closing’ or ‘add island’ tools. No partial volume corrections were performed.

2.2.2. Generation of patient-specific mesh phantoms

A tetrahedral mesh representation of each patient segmentation (i.e. a patient-specific mesh phantom) was generated using the Segment Mesher extension of 3D Slicer (figure 2). Each mesh conversion was performed using the Cleaver2 algorithm [12], which utilized a feature scaling parameter (controls mesh coarseness) of 2.0, a sampling rate (controls mesh detail) of 0.25, and an element size gradient parameter (constrains element size anisotropy) of 0.20. Elemental compositions and densities for each region were defined based on blood-inclusive values from ICRP Publication 110. Only regions derived during segmentation were generated in phantom. Phantom length/portion of patient modelled was defined by the SPECT/CT field of view.

Figure 2.

Patient-specific phantoms for the male patient involved in this study. (A) Segmentation used to define lesions (SPECT guidance; lower) and normal organs with elevated uptake (CT guidance; upper). (B) Patient-specific voxel phantom. (C) Patient-specific tetrahedral mesh phantom. Color legend: lesions—cyan, magenta; liver—light brown; kidneys—dark brown; spleen—violet; lungs—pink; compact bone—white; spongiosa—red.

2.2.3. Generation of patient-specific voxel phantoms

A voxel labelmap representation of each patient segmentation (i.e. a patient-specific voxel phantom) was generated using the Segmentations extension of 3D Slicer (figure 2). Elemental compositions and densities for each region were defined based on blood-inclusive values from ICRP 110.

2.2.4. Landmark placement

For each imaging time point, virtual landmarks were placed at the lesion foci and kidney pelves to support deformable registration (vide infra).

2.2.5. Registration

The CT scan acquired at the first time point was used as the ‘fixed’ volume to which the later time point CT scans were co-registered. The transforms generated at each registration step were applied to both the ‘moving’ CT and SPECT volumes. After manual alignment, rigid registration was performed, followed by B-spline deformable registration using the Plastimatch [13] extension of 3D Slicer. The singlestage deformable registration utilized a mean squared error cost function, a 4 × 4 × 2 image subsampling rate, 100 mm grid size, and landmark and second-derivative smoothness penalties of 100 and 0.1 were enforced, respectively. Of note, use of only rigid co-registration, and rigid + deformable co-registration without landmark information were both attempted, but yielded unsatisfactory lesion overlap.

2.2.6. Generation of time-integrated activity concentration maps and organ-level time-integrated activities

For each patient, the co-registered quantitative SPECT images were cropped and resampled to the same image dimensions and spacing. Using an in-house Python script, the SPECT images (in units of Bq/ml) were then voxel-wise integrated via the trapezoidal method up to the last measured time point, after which decay was assumed to occur via physical decay only. The time-integration period was taken as t = 0 to infinity. Activity concentration at t = 0 was assumed to be equal to that at the first measured time point. The mean time-integrated activity (TIA) concentration for each organ was computed from the resulting time-integrated activity concentration map [Bq·h/ml]. Whole organ time-integrated activity coefficients (TIACs, in units of hours) needed for use in the organ-level dosimetry codes were then computed as the product of the mean TIA concentration and the voxel-defined volumes for each region [ml], followed by division by the administered activity [Bq].

2.3. Internal dosimetry codes

2.3.1. MIRDcalc

MIRDcalc is an Excel^®-based organ-level dosimetry code which utilizes S-values[11] derived from specific-absorbed fractions promulgated in ICRP Publication 133 [14] for the ICRP 110-series computational reference phantoms. It approximates patient-specific dose calculations by scaling self S-values for reference phantom organs with patient-specific organ masses. For weakly-penetrating radiations (e.g. electrons):

$$S_{PS}(r_T\leftarrow r_S)=S_R(r_T\leftarrow r_S)\cdot \frac{m_R}{m_{PS}}{r}_T=r_S$$ (1)

and, for photons:

$$S_{PS}(r_T\leftarrow r_S)=S_R(r_T\leftarrow r_S)\cdot {\left(\frac{m_R}{m_{PS}}\right)}^{2∕3}{r}_T=r_S$$ (2)

where m_R is the mass of the reference phantom organ, m_PS is the mass of the patient organ, S_R(r_T ← r_S) is the S-value for the reference target organ, and S_PS(r_T ← r_S) is the scaled S-value for the patient organ. This approach does not consider differences in cross-irradiation arising from variation in organ mass, but due to the short beta particle range and relatively low gamma yield of ¹⁷⁷Lu, the approximation is considered accurate.

Whole-organ TIACs, computed using the method described in the preceding section, were used as input. The red marrow and rest of body regions in MIRDcalc include portions of the body outside of the SPECT image field of view. Therefore, the TIACs computed from the patient segmentations were linearly scaled up to the corresponding reference phantom region masses utilized in MIRDcalc.

MIRDcalc computes self absorbed doses for tissue spheres of arbitrary mass, based on log-log interpolated S-values. This functionality was used for lesion absorbed dose calculations; the lesions were assumed to be comprised of 100% soft tissue. These calculations ignore cross irradiation contributions from other tissues.

2.3.2. PARaDIM

PARaDIM is a dosimetry code based on the Particle and Heavy Ion Transport code System (PHITS [15]). It computes organ-level and 3D absorbed dose distributions from tetrahedral mesh region-level sources via direct Monte Carlo simulation. A uniform distribution of time-integrated activity within individual phantom organs is assumed. Whole-organ TIACs, in combination with the patient-specific mesh phantoms, were used as input.

2.3.3. In-house voxel dosimetry code (voxel MC)

PHITS Monte Carlo simulations were configured with an in-house Python program to compute voxel-level absorbed dose distributions from each patient TIA concentration map. The corresponding patient-specific voxel phantoms were used as the remaining input.

2.3.4. Monte Carlo simulation parameters and hardware

A total of 2 × 107 histories were used for each simulation computed using either PARaDIM or the voxel dosimetry code. Physical model settings used included PHITS-EGS5 method for treatment of multiple scattering, explicit treatment of fluorescent x-rays, consideration of Rayleigh and incoherent scattering, and consideration of electron-impact ionization. Sampling was utilized for determination of bremsstrah-lung polar angles, pair electron polar angles, and distribution of photoelectrons. Cutoff energies of 1.0 keV were utilized for electrons and photons. All emissions for ¹⁷⁷Lu specified in ICRP 107 [16] were included in the simulations (beta, monoenergetic electrons, gamma rays, and x-rays). Calculations we run on a HP Z8 workstation (3.6-GHz Intel Xeon 5122 processor, 32 GB RAM).

3. Results

Organ-level mean absorbed dose estimates obtained with voxel MC, PARaDIM, and MIRDcalc dosimetry software are presented in table 2. Absorbed doses for the kidneys, liver, spleen and rest of body agreed within 6% among the three methods evaluated (figure 3). For lesion mean absorbed doses, differences of up to 12% were observed for PARaDIM relative to the voxel MC method, and differences of up to 26% were observed for MIRDcalc spheres relative to the voxel MC method. Absorbed dose estimates for the critical organs (kidneys, liver, spleen, marrow) computed with all software were generally in line with those reported from the Neuroendocrine Tumors Therapy (NETTER-1) trial [17], with the exception of the red marrow dose computed for the female patient, which for all software was ~3–4 fold higher than the NETTER-1 average. Absorbed doses for the ancillary organs were computed only with MIRDcalc, as these non-segmented organs are pre-defined in the reference phantoms. The estimates computed with MIRDcalc were generally 1- to 2-fold higher than the NETTER-1 average; this was due in part to the limited field-of-view from which the rest of body TIAC was calculated.

Table 2.

Organ-level dosimetry estimates for two patients receiving ¹⁷⁷Lu-DOTATATE.

	Male (7.21 GBq administered)			Female (7.31 GBq administered)			NETTER-1a
Tissue	$\overline{D}$ [Gy] (PARaDIM)	$\overline{D}$ [Gy] (Voxel MC)	$\overline{D}$ [Gy] (MIRDcalc)	$\overline{D}$ [Gy] (PARaDIM)	$\overline{D}$ [Gy] (Voxel MC)	$\overline{D}$ [Gy] (MIRDcalc)	$\overline{D}$ [Gy]
Lesion 1	46.1	46.6	51.4	3.45	3.18	3.39	N/A
Lesion 2	47.6	42.6	53.7	23.2	20.6	25.0	N/A
Lesion 3	N/A	N/A	N/A	2.79	2.97	2.93	N/A
Lesion 4	N/A	N/A	N/A	5.45	5.66	5.94	N/A
R. Kidney	3.97	4.15	—	4.68	4.97	—	N/A
L. Kidney	4.39	4.63	—	3.44	3.65	—	N/A
Total kidney	4.17	4.39	4.65	4.27	4.50	4.66	4.9 ± 2.2
Liver	3.03	3.16	3.30	1.78	1.84	1.94	2.2 ± 1.7
Spleen	5.16	5.41	5.71	N/A	N/A	N/A	6.3 ± 6.0
Bone marrow	0.313	0.343	0.456	0.870	0.952	0.906	0.25 ± 0.20
Rest of body	0.404	0.411	0.476b	0.679	0.685	0.742b	c

^aNETTER-1 absorbed dose normalized to a single 7.4 GBq fraction, computed using stylized anthropomorphic phantoms.

^bMean for organs not listed in table 1; MIRDcalc generated doses for all target organs provided in Supplemental table S1 (available online at stacks.iop.org/BPEX/7/057002/mmedia).

^cNETTER-1 absorbed doses for all target organs provided in Supplemental table S1 N/A entries: the male patient has only 2 identified lesions. The female patient has no spleen.

Figure 3.

Organ-level dosimetry comparison.

Absorbed dose maps and cumulative dose volume histograms (DVH) obtained with the voxel MC method are presented in figure 4. The DVH for lesions in the male patient showed 40% of the volume of each lesion received at least 50 Gy of absorbed dose, with the intratumoral dose distribution for the Lesion 2 being less homogeneous. Three additional ‘hot spots’ were evident in the liver absorbed dose distribution of the male patient, which partly comprise the ‘tailed’ portion of the DVH (figure 4(A)); however, the fractional volume of liver tissue receiving >10 Gy was less than 6%. The female patient displayed significant intertumoral heterogeneity; Lesions 1 and 3 received comparable dose distributions while Lesion 2 received greatly elevated doses (over 70% of the volume of Lesion 2 received absorbed dose in excess of the maximal absorbed dose in the other lesions).

Figure 4.

Voxel-level dosimetry. (A) Upper panel: Axial, coronal, and sagittal slices, and maximum intensity projection, of the absorbed dose distribution for the male patient. Lower panel: Cumulative dose volume histograms for lesions and organs-at-risk. (B) Same as (A) but for the female patient.

4. Discussion

Dosimetry estimates for two patients administered ¹⁷⁷Lu-DOTATATE were computed with three different dosimetry software—MIRDcalc, PARaDIM, and an in-house voxel MC program—in good overall agreement.

4.1. Image segmentation, co-registration, and time-integration considerations

The same image segmentation, co-registration, and time-integration workflow was applied for each software, which enabled homogenization of input (i.e. consistency of organ volumes and integrated activity among software) and facilitated more direct comparisons of the output from each software. Therefore, we eliminated differences in organ/lesion volumes and TIACs as sources of input variation and ensured variation in organ-level absorbed doses arose mainly from the software algorithms or related assumptions. For example, differences in doses computed with PARaDIM versus voxel MC are expected to derive from the use of mesh phantoms and uniform organ-level TIA distribution in the former, whereas voxel phantoms and consideration of intra-organ TIA heterogeneity were considered in the latter method. Figure 5 presents a summary of similarities and differences among the methods.

Figure 5.

Comparison of calculation features.

Use of the versatile image processing workflow conveys several additional advantages, including reduction in personnel time, and simultaneous output from complementary dosimetry paradigms. However, certain limitations exist. Likely the largest sources of lesion absorbed dose uncertainty derive from inaccuracies in co-registration between serial imaging time points, and inaccuracies due to SPECT image resolution. For organ-level dose calculations, both factors can be mitigated by performing separate segmentations at each time point, which lessens uncertainties arising from patient motion/mispositioning, and enables partial volume corrections to be applied to work around suboptimal image resolution. Finally, we note that the classical organ-level dosimetry standard involves determining organ mean uptake values at each time point, followed by time-integration. Several time-integration methods are commonly used, including the trapezoidal method we have used, Riemann sums, non-linear fitting, or a combination approach for modeling different intervals of organ time-activity curves [1, 18]. Though each integration scheme can be applied in a voxel-wise manner to generate time-integrated activity maps, image noise and co-registration errors often translate to large uncertainties in the time-integrated activity for individual voxels.

4.2. Optimization: benefits of combining approaches to dosimetry calculations

The superiority or inferiority of voxel-level versus organ-level dosimetry continues to be widely debated [19]. A primary advantage of voxel dosimetry is that inhomogeneity in intra-lesion or intra-organ absorbed dose can be assessed and potentially utilized in prediction of therapeutic response of adverse tissue reactions. Its limitations primarily relate to the aforementioned image resolution (>10 mm for most SPECT systems) and voxel co-registration considerations; additionally, at voxel sizes practical for patient-specific phantoms (generally 1–10 mm), voxel geometry is suboptimal for defining small (e.g., lymph nodes, micrometastases), thin (e.g. epithelia of walled organs), or complex radiosensitive structures (e.g. red marrow). Further, in the case of ¹⁷⁷Lu, the typical voxel size is large in comparison to the beta particle range in tissue (~0.2 mm for the mean energy beta emission), which is insufficient to capture dose gradients that may be relevant to absorbed dose-effect relationships [19]. On the other hand, modern reference phantoms like those integrated in MIRDcalc are generally designed to support radiation protection as well as nuclear medicine, and incorporate detailed modeling of small-scale anatomy down to micrometer scales [20, 21]. For ¹⁷⁷Lu-DOTATATE, this method is expected to give more accurate dose estimates for the red marrow. However, reference phantoms represent an ‘average patient’ and do not directly account for the unique anatomy of specific patients. Considering voxel-based and reference phantom-based approaches together would overcome the limitations of each, as well as providing a measure of quality control.

Of note, the PARaDIM method is somewhat of a hybrid between voxel and organ-level approaches. In this method, a uniform distribution of time-integrated activity in whole organs is used, and organ-level dose deposition is scored in volume elements of a tetrahedral mesh phantom. Here, the phantom was derived from the patients’ tomographic images, and thus directly models patient anatomy and potentially complex lesion geometry. Additionally, a 3D voxel dose map can be output and used for generation of dose volume histograms. In the case of ¹⁷⁷Lu-DOTATATE, these additional features were seen to add little value; as the organ-level doses arise mainly from short range beta self-irradiation, the salient features of the dose volume histogram become dominated by statistical sampling and partial volume effects near the organ boundary surfaces.

4.3. Throughput

Time required for image segmentation, co-registration, and related processing was approximately 1 h, including setup. Monte Carlo calculations were run on standard PC workstation without hardware acceleration and required approximately 100 min to compute. Computation time could be reduced via parallel processing, but given the time required for segmentation, this likely would not constitute a bottleneck.

4.4. The big picture

Across the field many agree that there is likely a role for dosimetry-tailored treatment planning to personalize radiopharmaceutical therapies and improve outcomes. But the implementation of that practice remains unestablished, in part because of the complexity of many unique elements in the dosimetry calculation workflow and the unfavorable economics of increased imaging and treatment planning support. Dosimetry can be performed at a spectrum of complexities. While the most complex methods/tools offer the promise of the most accurate dose calculations, they also are the most difficult to implement due to expense and effort and are for that very reason difficult to standardize. The lack of standardization of dosimetry calculation methods has been noted as a major obstacle in our field’s efficacious use of dosimetry [22, 23]. Ultimately, moving dosimetry into the clinic is a multi-faceted challenge.

In this work we investigate the variability of dosimetry methods that can be reasonably performed and standardized across the academic community using (mostly) publicly accessible tools and 1st order dose calculations and assumptions. Larger sample sizes than those presented here are needed to establish true variability in populations. However, if and when those metrics are established, simple standardized methods may offer a practical path to widespread dosimetry; based on principles of practical implementation and community standardization.

In the case of Lutathera treatments, we have seen that the optimal number of treatment cycles based on maximum tolerable dose can vary between 2 and 10 [24]. Thus, even if we accept dosimetry has an uncertainty inherent in it, as well as uncertainty added from simplification efforts, it still may provide added value.

5. Conclusions

A paired MIRD schema/voxel Monte Carlo dosimetric approach, supported by a single image processing workflow, was determined to be valuable for personalized ¹⁷⁷Lu-DOTATATE dosimetry. Mean organ doses computed with three tested dosimetry software tools—MIRDcalc, PARaDIM, and a voxel Monte Carlo method—were found to be in good agreement.

Table 1.

Organ and lesion masses [g] for phantoms of different formats.

	Male patient			Female patient
Tissue	Tissue mass (mesh)a	Tissue mass (voxel)b	Tissue mass (MIRDcalc)c	Tissue mass (mesh)a	Tissue mass (voxel)b	Tissue mass (MIRDcalc)c
Lesion 1	49.4	52.1	52.1	10.0	11.8	11.8
Lesion 2	2.93	4.07	4.07	7.66	8.89	8.89
Lesion 3	N/A	N/A	N/A	60.5	65.4	65.4
Lesion 4	N/A	N/A	N/A	12.0	13.5	13.5
R. Kidney	240	246	246	203	210	210
L. Kidney	228	243	243	102	105	105
Liver	1950	1960	1960	1626	1630	1630
Spleen	236	244	244	N/A	N/A	187
Bone marrow	118	128	1390	130	161	1064
Rest of body	24100	21400	61800	15700	14300	52300

^aMass of region defined in patient-specific tetrahedral mesh phantom.

^bMass of region defined in patient-specific voxel phantom.

^cMass used in MIRDcalc program for S-value scaling.

Data availability statement

The data that support the findings of this study are available upon reasonable request from the authors.