Preclinical Voxel-Based Dosimetry in Theranostics: a Review

Abstract

Due to the increasing use of preclinical targeted radionuclide therapy (TRT) studies for the development of novel theranostic agents, several studies have been performed to accurately estimate absorbed doses to mice at the voxel level using reference mouse phantoms and Monte Carlo (MC) simulations. Accurate dosimetry is important in preclinical theranostics to interpret radiobiological dose-response relationships and to translate results for clinical use. Direct MC (DMC) simulation is believed to produce more realistic voxel-level dose distribution with high precision because tissue heterogeneities and nonuniform source distributions in patients or animals are considered. Although MC simulation is considered to be an accurate method for voxel-based absorbed dose calculations, it is time-consuming, computationally demanding, and often impractical in daily practice. In this review, we focus on the current status of voxel-based dosimetry methods applied in preclinical theranostics and discuss the need for accurate and fast voxel-based dosimetry methods for pretherapy absorbed dose calculations to optimize the dose computation time in preclinical TRT.

Introduction

Small animals, especially mice, have been increasingly used in preclinical research to develop novel theranostic agents and treatment modalities for human diseases, including cancer [1, 2]. Cancer treatment methods are typically based on the use of ionizing radiation, e.g., external beam radiotherapy (EBRT), brachytherapy, and radionuclide therapy. Recently, targeted radionuclide therapy (TRT), especially peptide receptor radionuclide therapy (PRRT), has gained increasing importance in the treatment of various cancers, including lymphoma, glioblastoma, neuroendocrine tumors, and prostate cancer with distant metastases [3–7]. TRT can simultaneously target primary tumor sites and distant metastatic disease that remains undetectable by diagnostic imaging [8]. Selective uptake and prolonged retention of the radiopharmaceutical within the tumor are required for successful PRRT [9]. Several new radiolabeled peptides are being developed to target specific receptors that are overexpressed in various cancers, such as glioma, ovarian cancer, neuroendocrine tumor, and cervical cancer [10, 11]. Such peptides are useful because normal human organs and tissues exhibit very limited expression of these receptors.

All cancer therapies that use ionizing radiation, including TRT, share a common goal: to deliver the highest possible absorbed dose to the tumor while sparing healthy tissues to achieve the highest therapeutic efficacy. To fulfill this objective, the careful selection of radionuclides combined with specific vectors is required to target the cancer cells. Many peptides conjugated with theranostic radionuclides, such as ¹⁷⁷Lu and ¹⁸⁸Re, have been considered valuable tools for novel and effective anti-cancer therapies [12]. Despite the excellent tumor targeting ability of these radiolabeled peptides, there is always a risk of renal and bone marrow toxicity due to the particulate radiation emitted from the theranostic radionuclides [13–15]. Therefore, internal dosimetry for absorbed dose calculation is essential to obtain absorbed dose-response relationships during clinical and preclinical TRT for the evaluation of therapeutic outcomes and toxicities [16–18]. However, accurately assessing the absorbed dose in tumors and normal organs or tissues during TRT treatment planning remains challenging [19–21].

Absorbed doses in small animals can be calculated using a Medical Internal Radiation Dose (MIRD) scheme if animal-specific S values (mean absorbed dose in a target organ per radioactivity decay in a source organ) are determined [22–24]. However, an organ-level MIRD method does not consider patient- or animal-specific anatomies and activity distributions because it assumes that there are homogeneous activity distributions in organs and a generalized geometry [25, 26]. Therefore, the inclusion of nonuniform activity distributions and tissue heterogeneity has been considered important for more accurate absorbed dose calculations at the voxel level [26–30]. This article reviews the current status of preclinical voxel-based dosimetry methods applied in theranostics and discusses the need for accurate and fast dosimetry methods to calculate pretherapy absorbed dose at the voxel level to optimize the dose computation time in preclinical TRT.

Radionuclides for Preclinical TRT

The selection of radionuclides is crucial in TRT treatment planning and should be performed individually, based on the characteristics of the tumor, adjacent tissues, and affinity for the targeting compound. The radionuclides that emit particulate radiation (α, β, and Auger) are of primary interest for TRT to cause non-reparable DNA damage by radiation-induced ionization [31, 32]. Because peptides are internalized, the physical and chemical characteristics of the radionuclides used in TRT should be fully explored [19]. Physical properties of radionuclides that are important for clinical and preclinical applications are mode of production, half-life, type of emission, particle energy, the maximum range of particles emitted, and linear energy transfer (LET).

Although many radioisotopes can be used as radiation sources, only a few have been developed and applied in preclinical and in vivo studies; these include ¹¹¹In, ⁶⁸Ga, ⁶⁴Cu, ⁹⁰Y, ¹⁸⁸Re, and ¹⁷⁷Lu. Due to their inherent theranostic natures, ¹⁷⁷Lu and ¹⁸⁸Re are used for both diagnostic and therapeutic purposes [33]. The shorter tissue penetration range of ¹⁷⁷Lu may reasonably exert a more favorable effect on small tumors compared with ⁹⁰Y and ¹⁸⁸Re [19]. Radionuclides that emit gamma emission are desirable in theranostics because monitoring of in vivo radiopharmaceutical distribution as well as the assessment of pre- and post-therapy dosimetry is possible. Radionuclides used in diagnostics emit a large fraction of distributed radiation energy, whereas radionuclides for therapy are designed to deposit the energy from radioactive decay locally [34]. A list of common therapeutic radionuclides with their physical properties and applications is presented in Table 1.

Table 1.

Common therapeutic radionuclides with their physical properties and applications (maximum energy and particle range in tissues are reported for beta emission)

Radionuclide	Half life (physical)	Useful emissions	Energy	Particle range in tissue (Rmax)	Application
188Re	16.9 h	β−	Emax, β−: 2.12 MeV	11 mm	Theranostics
		γ	Eγ: 155 (10%) keV
177Lu	6.73 days	β−	Emax, β−: 0.498 MeV	2 mm	Theranostics
		γ	Eγ: 208 (11%), 113 (6%) keV
166Ho	28.8 h	β−	Emax, β−: 1.85 MeV	8.7 mm	Theranostics
		γ	Eγ: 81 (7%) keV
153Sm	46.5 h	β−	Emax, β−: 0.805 MeV	0.33 mm	Theranostics
		γ	Eγ: 103 (28%), 70 (5%) keV
131I	8.02 days	β−	Emax, β−: 0.606 MeV	0.42 mm	Theranostics
		γ	Eγ: 364 (81%), 637 (7%) keV
111In	67.4 h	γ e− Auger	Eγ: 173 (83%), 247 (94%) keV E: 0.5–25 keV	10 μm 550 μm
		e− IC	E: 144–245 keV		Theranostics
90Y	64.1 h	β−	Emax, β−: 2.28 MeV	11.3 mm	Therapy only
89Sr	50.5 days	β−	Emax, β−: 0.583 MeV	0.7 mm	Therapy only
64Cu	12.8 h	β−	Emax, β−: 0.570 MeV	2.5 mm	Theranostics
		β+	Emax, β+: 0.653 MeV	0.7 mm

Imaging Modalities for Preclinical TRT and Dosimetry

During the preclinical development of radiopharmaceuticals, numerous small animals are injected with radioactive agents and sacrificed at different time points to obtain biodistribution data of the injected radiopharmaceutical for the assessment of toxicity and efficacy [16]. Recently, due to the advancement in high-resolution radionuclide imaging, a variety of small-animal single-photon emission computed tomography (SPECT) and positron emission tomography (PET) protocols were developed for preclinical imaging studies of radiolabeled peptides and antibodies to investigate disease development and therapy response [35–38]. PET imaging for small animals has been considered more accurate for quantitative studies due to its higher sensitivity and good spatial resolution [39, 40]. SPECT provides three-dimensional (3D) spatial information; however, physical quantities such as photon attenuation, scattering, and partial volume errors influence the absolute quantification of SPECT images [41].

Several researchers have proposed different techniques to solve these issues. Consequently, preclinical SPECT/CT cameras with multi-pinhole collimators have been developed to increase system sensitivity while maintaining good spatial resolution [42–45]. Finucane et al. [46] conducted phantom imaging using ¹¹¹In and ^99mTc sources and obtained quantitatively accurate information from a multi-pinhole SPECT/CT camera to assess radiotracer biodistribution in mouse models; these assays can replace conventional dissection studies. Gupta et al. [47] performed several phantom studies using ¹⁷⁷Lu and ^99mTc to evaluate the system performance of a multi-pinhole SPECT/CT and found that the parameters evaluated are suitable for obtaining quantitatively accurate SPECT images of mice for personalized dosimetry in preclinical ¹⁷⁷Lu PRRT. The CT-derived anatomic information obtained from integrated SPECT/CT improves SPECT quantification and organ delineation in TRT, which is an essential step in 3D voxel-based dosimetry [26, 48].

Organ-Level Internal Dosimetry

Internal dosimetry is the method of absorbed dose calculation in a volume of interest (i.e., whole organ, tumor, or voxel) from internally distributed radionuclides in radionuclide imaging and therapy. The absorbed dose of each organ depends on the physical properties of the radionuclide, the injected activity, and the kinetics of uptake and clearance of radioactivity within the tumor and normal tissue or cells [49].

The method for calculating the absorbed dose of radiation from internal sources was formulated by the MIRD committee of the Society of Nuclear Medicine in the 1960s [50]. MIRD is a dosimetry method that uses S values (mean absorbed dose in a target organ per radioactivity decay in a source organ), also known as dose factors released by the Radiation Dose Assessment Resource (RADAR) group, especially for human organ sizes and relationships [30, 50, 51]. According to the MIRD schema [30], the average absorbed dose D(r_T, T_D) is given by

$$D\left(r_T,T_D\right)=\sum _{r_s}{\int }_0^{T_D}\tilde{A}\left(r_s,t\right)S\left(r_T\leftarrow r_s\right)\mathit{dt},$$ 1

where r_T is the target organ, and $\tilde{A}$ (r_s, t) is the time-integrated activity in source organ r_s over the dose integration period,T_D.

The absorbed dose in small animals can be calculated using the MIRD schema if murine-specific S values are supplied [16, 22–24, 52]. However, the application of the MIRD method for the calculations of absorbed dose in mice is not direct because the size and relationship of mouse organs are significantly different than those of humans [27, 28, 53]. In addition, the anatomical difference between humans and mice becomes more important when the dosimetry is performed for beta-emitting radionuclides. The cross-beta absorbed dose should be considered when using MIRD formalism for mouse dosimetry because the size of organs in mice and the range of beta particles are the same order of magnitude. In addition, organs are very close to each other in mice [54, 55].

Based on the MIRD formalism and RADAR method of dose calculation, several dosimetry applications were developed. MIRDOSE [56] and Organ Level Internal Dose Assessment/Exponential Modeling (OLINDA/EXM) [57] are the most widely used fixed geometry dosimetry software packages. Organ-level dosimetry applications based on standard phantoms are inadequate for absorbed dose calculations in TRT because they do not incorporate patient-specific or tumor dosimetry. However, OLINDA/EXM is capable of modeling simple tumors in the form of unit density spheres of various sizes based on the absorbed fractions by assuming uniform activity distributions [58]. Because it is not always possible to model the size, shape, and location of every unique tumor with the reference phantoms used in OLINDA/EXM, this approach does not provide information about the 3D dose distribution in tumors [57–59].

Voxel-Level Dosimetry Using a MIRD Schema

Absorbed dose calculations can also be performed at the voxel level. Several researchers have addressed the limitations associated with organ-level dosimetry, and various steps have been taken to extend the MIRD schema to absorbed dose calculations at the voxel level by using voxel S values [29, 30]. MIRD pamphlet no. 17 [29] summarizes the methods of voxel-based dosimetry using voxel S values in the MIRD formalism that require the assessment of the 3D distribution of the radiopharmaceutical within the body. According to Bolch et al. [29], mean absorbed dose $\overline{D}$ to a given target voxel k from N surrounding source voxels h (including the target voxel itself, h = 0) can be calculated using the following equation:

$$\overline{D}\left({\text{voxel}}_k\right)=\sum _{h=0}^N{\tilde{\mathrm{A}}}_{{\text{voxel}}_h}.S\left({\text{voxel}}_k\leftarrow {\text{voxel}}_h\right),$$ 2

where ${\tilde{A}}_{{\text{voxel}}_h}$is the time-integrated activity in the source voxel, h and S(voxel_k ← voxel_h) are the voxel S values that are defined as the mean absorbed dose to a target voxel per radioactivity decay in a source voxel, both of which are contained in an infinite homogenous tissue medium [29].

Tabulations of voxel S values for several voxel dimensions calculated with MC code for a few radionuclides were reported in MIRD pamphlet no. 17. The MIRD formalism using voxel S values is possibly the most accepted and easy to implement method for voxel-based dosimetry that does not require volume integrations of the dose point kernel (DPK) over sources and targets [60]. However, the preclinical voxel S value dataset for different radionuclides at various voxel sizes is not yet available to perform voxel-based dosimetry in small animals using a MIRD schema. Moreover, a voxel S value approach does not consider tissue nonhomogeneity during absorbed dose calculations [29, 30].

Voxel-Based Small Animal Models

Over the past two decades, there have been significant developments in the generation of digital models of mice and rats for the calculation of organ-absorbed doses in preclinical TRT. Among these models, the majority of mouse phantoms are stylized, containing organs with simplified geometries [27, 28, 55]. With advancements in high-resolution tomographic imaging systems (CT and MRI), more anatomically realistic voxel-based models have emerged, either from tomographic images or from digital photographs of mice cryosections [61–64]. Significant differences in absorbed dose results have been reported when stylized and voxel-based models have been used in Monte Carlo simulations [63–65]. Hindorf et al. [55] used stylized mathematical models for mouse dosimetry and observed discrepancies in the calculated S values that led to differences in absorbed dose calculations, particularly for high-energy beta emitters, when the mass, shape, and relative locations of organs within the mouse body were considered. Their results showed that S values could differ as much as 80% from their true value if linear interpolation from S value tables is used to obtain an S value for the specific mass of an organ. They further found that the shape of an organ is not crucial for S value calculation. However, their results showed that the cross-absorbed S value is strongly dependent on geometry and emitted radiation. The new high-quality mesh-type adult reference computational phantoms developed by the International Commission on Radiological Protection (ICRP) include all important source and target tissues for the estimation of effective dose, thereby obviating the need for supplemental organ-specific stylized models [66].

In 2004, Segars et al. developed 4D digital mouse body (MOBY) and rat body (ROBY) phantoms that are based on non-uniform rational b-spline (NURBS) mathematical models to define organ boundaries [67–69]. Several researchers have adopted MOBY phantoms for MC simulations to calculate organ S values for absorbed dose estimation in TRT [17, 23, 24, 70–73]. However, a study performed by Mauxion et al. [23] demonstrated the limitations of using digital mouse models for MC-based absorbed dose calculations. They observed very different dose distributions in some organs when two similar mouse models generated from the same software were used. Kostou et al. [17] and Boutaleb et al. [52] concluded that no specific digital mouse model could be applied for individualized dosimetry in murine studies because small variations in mouse anatomy may significantly affect the absorbed dose results (Fig. 1).

Fig. 1.

Examples of mouse models. a Stylized (Hindorf et al. 2004). b MOBY (Segars et al. 2004). c Bitar et al. (2007)

Monte Carlo Radiation Transport Codes

Many MC radiation transport codes, such as MCNPx, EGSnrc, PENELOPE, and Geant4/Geant4 Application for Emission Tomography (GATE), were developed in the last few years to calculate absorbed dose at the voxel level. Following the development of more realistic preclinical voxel-based phantoms, several dosimetry studies have been performed using these MC codes to estimate small animal-specific absorbed fractions (SAFs), organ-level S values, and voxel S values for various diagnostic and therapeutic radionuclides [60–64, 70–73]. EGSnrc [74] and MCNPx [75] have been widely accepted and are considered reference MC codes for dosimetry simulations involving electron-photon transport at low energies down to 1 keV. Bitar et al. calculated S values for various radionuclides, including ⁹⁰Y and ¹⁸⁸Re, for a large number of source-target combinations using the MCNP MC Code in a 30-g mouse model [63]. Xie and Zaidi generated an S values database for various PET radionuclides using MCNPX MC in the MOBY phantom for the assessment of radiation dose to mice [70]. Lanconelli et al. provided a free dataset of voxel S values for therapeutic radionuclides, including ¹⁷⁷Lu, at various voxel sizes for TRT dosimetry using DOSXYZnrc code [60].

Recently, the GATE, an MC toolkit [76] based on Geant4 [77], is being used frequently for both clinical and preclinical dosimetry applications. By using Geant4, we can simulate the electromagnetic interactions of photons, electrons, hadrons, and ions with matter down to the electron volt energy scale [77, 78]. Various studies have been performed to validate the reliability and accuracy of GATE MC for preclinical dosimetry [17, 23, 72].

Voxel-Based Dosimetry

In TRT, normal tissue toxicity and therapeutic efficacy are of great concern, and there is much less tolerance for inaccuracies in absorbed dose calculations. Therefore, to consider non-uniform activity distributions in organs for accurately calculating absorbed dose, voxel-based dosimetry methods that use DPK [79, 80] or voxel S value (VSV) approaches [51] have been suggested. The DPK method of voxel-based dosimetry is limited when compared with the VSV approach because it requires CPU-intensive conversions of spherical coordinates to Cartesian coordinates over the target volumes. The use of the VSV approach is also limited because of the requirement for a VSV database of each radionuclide at various voxel sizes. Moreover, these methods do not incorporate heterogeneity in the organ tissues into the absorbed dose calculations.

Direct MC (DMC) simulation is believed to produce more realistic dose distributions at the voxel level with high precision because it can handle tissue heterogeneity as well as nonuniform source distributions [81–84]. With advancements in 3D imaging modalities and user-friendly MC codes, several groups have developed personalized dosimetry tools for clinical applications, such as 3D Internal Dosimetry (3D-ID) [85], the Royal Marsden Dosimetry Package (RMDP) [86], VoxelDose [87], 3D Radiobiological Dosimetry (3D-RD) [81], Dose Planning Method (DPM) [88], RAYDOSE [89], and VIDA [90]. Recently, Bednarz et al. [65] introduced an MC-based preclinical dosimetry platform, called the Radionuclide Assessment Platform for Internal Dosimetry (RAPID), which is capable of calculating murine-specific 3D dose distributions.

Although DMC simulation has been validated for clinical TRT dosimetry applications, only a small number of preclinical dosimetry studies using DMC simulation have been reported [17, 23, 72, 91–93]. Furthermore, these studies used standard mouse models, such as the MOBY phantom, for MC dosimetry simulations. In recent preclinical dosimetry simulations performed with MOBY phantoms, authors demonstrated that small variations in organ anatomy could significantly impact the dose calculations [17, 52]. Therefore, they suggested that personalized dosimetry might not produce accurate absorbed dose values when a specific murine model is used for preclinical TRT. Using 3D imaging data from individual mice in MC simulations might reduce errors in voxel-based dose calculations arising from the variation in organ anatomies and activity distributions. Gupta et al. [93] evaluated the feasibility of GATE MC simulations for preclinical voxel-based absorbed dose calculations using ¹⁸F-FDG PET/CT imaging data of individual normal mice. Because GATE MC simulation considers tissue heterogeneity and nonhomogeneous activity distributions within the mouse body, the voxel-based absorbed doses, as calculated from ¹⁸F-FDG PET data in their study, are considered to be more realistic and mouse-specific. From the validation study, it was concluded that the GATE MC simulation toolkit could be applied to murine-specific voxel-based dosimetry in preclinical PRRT to perform tumor absorbed dose calculations more accurately (Fig. 2).

Fig. 2.

Voxel-based dosimetry based on direct Monte Carlo (DMC) simulation. a CT image of a mouse used for GATE MC simulation. b Output of simulation or dose map. c Dose map overlaid on the CT image. Coronal (left) and sagittal (right) views of the images are shown respectively in the figure. Gupta et al. [93]

¹⁷⁷Lu-Based Preclinical PRRT Dosimetry

Peptide receptor TRT using ¹⁷⁷Lu has gained an established role in the preclinical assessment and implementation of clinically relevant diagnostics or therapeutics [94–100]. ¹⁷⁷Lu is a theranostic radionuclide having several advantages in performing imaging studies for the evaluation of radiotracer biodistribution as well as for cancer therapy [95]. The lower tissue penetration range and the efficient cross-fire effect of ¹⁷⁷Lu may kill small tumors more effectively relative to other theranostic radionuclides [19]. Because of their short range, β-particles emitted from ¹⁷⁷Lu traverse several cells (10–1000), causing the destruction of multiple cells in the vicinity of the cells that accumulate the radiotracer [94, 95]. This effect is known as the “cross-fire effect” and is crucial for ensuring that a sufficient dose is delivered to each cell and for improving the homogeneity of tumor dose during cancer therapy [96].

Quantitative ¹⁷⁷Lu-SPECT Imaging

The absorbed dose calculations for PRRT are based on residence times obtained from serial quantitative SPECT/CT images acquired at multiple time points. Hence, the quantitative accuracy during ¹⁷⁷Lu-SPECT imaging is essential for reliably assessing tumor uptake and tumor-to-normal tissue ratios necessary to perform personalized dosimetry [101–103]. Accurate image quantitation in small animal SPECT is always challenging due to the limitations in the instrumentation and imaging process; therefore, it is important to evaluate performance parameters of preclinical SPECT systems used for theranostic radionuclides during pre- and post-therapy dosimetry studies [103, 104]. Mezzenga et al. [103] evaluated the quantitative accuracy of ¹⁷⁷Lu SPECT imaging for both small and large objects using cylindrical homogeneous reference geometry. The authors investigated the relationship between 3D-OSEM algorithm, object size, and coefficient of variation to demonstrate that ¹⁷⁷Lu SPECT is suitable for activity quantification in small volumes.

Gupta et al. [47] performed several phantom studies to evaluate the quantitative accuracy and performance parameters (recovery coefficient, uniformity, spatial resolution, and sensitivity and calibration factor) of a multi-pinhole SPECT/CT system for ¹⁷⁷Lu imaging. They investigated the relationship between activity recovery and image uniformity at different iteration numbers during iterative image reconstructions (3D-OSEM). In addition, the performance parameters of the SPECT camera measured for ¹⁷⁷Lu were compared with those of ^99mTc. The quantitative accuracy achieved in their study plays an important role in activity quantification when performing ¹⁷⁷Lu-SPECT imaging in mice for personalized dosimetry.

Murine-Specific ¹⁷⁷Lu-Dosimetry

The activity administered during radionuclide therapy is generally determined conservatively to avoid potential toxicity; however, this approach may result in sub-therapeutic treatments, causing cancer recurrence [105]. In clinical trials, ¹⁷⁷Lu therapy delivered at a fixed level of activity shows promising results in terms of treatment outcome. Nevertheless, several studies showed that pretherapy patient-specific dosimetry plays an important role in treatment planning during TRT of various cancers to improve the probability of tumor control and reduce normal tissue toxicity [106–108]. This concept may also apply to preclinical TRT using ¹⁷⁷Lu-labeled radiopharmaceuticals, and voxel-based dosimetry is essential to achieve more accurate dosimetry results for the interpretation of the radiobiological response. However, only a few voxel-based dosimetry studies of preclinical ¹⁷⁷Lu TRT that are based on MC simulations of digital murine models have been reported [109–112].

Vilchis-Juarez et al. [49] performed voxel-based dosimetry using MC codes and compared the absorbed doses in kidneys and tumors to evaluate the therapeutic response of ¹⁷⁷Lu-AuNP-RGD in athymic nude mice bearing α(v)β(3)-integrin-positive C6 gliomas. Haller et al. [109] investigated the therapeutic anti-tumor effects and radiation-induced nephrotoxicity of ¹⁷⁷Lu-cm09 and ¹⁷⁷Lu-EC0800 in KB tumor-bearing mice by estimating equivalent absorbed doses using PENELOPE MC simulation code. Kuo et al. [110] synthesized ¹⁷⁷Lu-HTK01169, a close analog of 177Lu-PSMA-617, and performed biodistribution and dosimetry studies in mice bearing PSMA-expressing LNCaP tumor xenografts using SPECT/CT imaging and OLINDA software v.2.0. Timmermand et al. [111] performed whole-body and small-scale tumor dosimetry of ¹⁷⁷Lu-labeled hu11B6 in an LNCaP tumor-bearing mice model using mouse-specific S-factors calculated with the MCNP6 MC package. Gupta et al. [112] recently synthesized ¹⁷⁷Lu-IONPs-Folate for preclinical targeted imaging and therapy of folate receptor-positive cancer and calculated voxel-based organ absorbed doses using SPECT/CT images of normal and KB tumor-bearing mice using a GATE MC simulation. They compared kidney- and tumor-absorbed doses at the voxel level obtained with ¹⁷⁷Lu-IONPs-Folate with those obtained with ¹⁷⁷Lu-labeled folic acid without IONPs (¹⁷⁷Lu-Folate) and ¹⁷⁷Lu-IONPs only. They found that the voxel-based absorbed dose to kidneys from ¹⁷⁷Lu-IONPs-Folate was almost half relative to that obtained with ¹⁷⁷Lu-Folate only. The tumor dose at the voxel level received from ¹⁷⁷Lu-IONPs-Folate was the highest when compared with those of ¹⁷⁷Lu-Folate and ¹⁷⁷Lu-IONP. Their study suggested that voxel-based dosimetry using MC simulations is reliable and potentially more accurate for ¹⁷⁷Lu-IONPs-Folate therapy of folate receptor-positive cancers on a personalized basis (Figs. 3 and 4).

Fig. 3.

Maximum intensity projection images showing dose maps overlaid on the CT images of ¹⁷⁷Lu-Folate (a), ¹⁷⁷Lu-IONPs-Folate (b), and ¹⁷⁷Lu-IONPs (c) obtained with GATE MC simulations of SPECT/CT of KB tumor-bearing mice acquired at 6 h post-injection. Energy deposited in the tumor at the voxel level was the highest from ¹⁷⁷Lu-IONPs-Folate compared with those from ¹⁷⁷Lu-Folate and ¹⁷⁷Lu-IONPs. Gupta et al. [112]

Fig. 4.

Voxel-based absorbed doses in organs and tumor (mean ± SD) calculated by GATE MC from ¹⁷⁷Lu-Folate, ¹⁷⁷Lu-IONPs, and ¹⁷⁷Lu-IONPs-Folate in KB tumor-bearing mice. Gupta et al. [112]

Fast Voxel-Based Dosimetry for Preclinical TRT

Although MC simulation is currently the most accurate method for voxel-based absorbed dose calculations, it is time-consuming, computationally demanding, and often impractical to carry out in practice [113, 114]. Therefore, several groups have proposed fast, voxel-based dosimetry methods based on DPK, multiple VSV, and semi-Monte Carlo (sMC) techniques that have been shown to overcome the limitations associated with the DMC method in clinical dosimetry [115–119].

Hippeläinen et al. [117] proposed a fast sMC code for ¹⁷⁷Lu PRRT dose calculations, which was based on the assumption of local electron absorption and fast photon MC simulations. Since the simulations of electrons are the most demanding part of MC simulations, electron absorption assumptions accelerate the absorbed dose calculations dramatically. Their results suggested that there is no need of electron simulation for ¹⁷⁷Lu absorption calculations when SPECT imaging system is used. Therefore, when the sMC method is applied with its local electron absorption assumption, voxel-based dosimetry, especially for ¹⁷⁷Lu PRRT, can be successfully performed with four SPECT/CT scans in 4 to 5 min. However, this fast sMC dosimetry method is only applicable to low-energy electrons emitted from ¹⁷⁷Lu because this method has not yet been validated for radioisotopes emitting high-energy electrons, such as ⁹⁰Y or ¹³¹I.

Recently, Lee et al. [119] proposed a fast and new method for voxel-based absorbed dose calculations using multiple VSVs that considers both nonhomogeneous activities and heterogeneous media. They applied MC simulation to acquire multiple VSVs with various densities and utilized them to acquire dose maps using CT-based segmentation. The authors successfully implemented the proposed dosimetry approach on digital phantom and whole-body dynamic ⁶⁸Ga-PET images of patients. The results obtained were comparable to those with the DMC technique and had the advantage of a significantly reduced dose-computation time. Because their proposed dosimetry method was fast and more accurate and generated dose maps at the whole-body level, it can be applied for personalized dosimetry in TRT. Most recently, Lee et al. [120] proposed a voxel-based dosimetry approach using a deep convolutional neural network (CNN) to overcome the limitations of the DMC method. Their CNN-based personalized dosimetry approach produced results comparable to that of DMC simulation, but with significantly lower computation time. This novel method appears to be more advanced and fast compared to traditional dosimetry methods because new deep learning methods routinely outperform conventional signal and image analysis techniques in multiple medical engineering fields [121–126].

Extrapolation of Data from Mouse to Human

Because it is possible to translate preclinical dosimetry data to a clinical setting, preclinical research using small animals has been used with increasing frequency to develop novel theranostic agents and to establish a relationship between absorbed dose and biological effects during preclinical TRT [23, 65]. Because very few patient-specific dosimetry studies are available in clinical theranostics, the ability to extrapolate data from mouse to human for the assessment of absorbed dose in the critical organs from novel theranostic radiopharmaceuticals is essential.

Various authors reporting preclinical TRT studies have presented methods to extrapolate dosimetry values from preclinical to clinical settings for the evaluation of radiation-induced organ toxicity and treatment response during clinical trials of new radiopharmaceuticals [127–132]. However, several assumptions and technical complexities arise during the measurement of radiotracer biodistribution and the extrapolation of data from the animal model to humans [132]. In addition, predetermined absorbed dose values in humans that are derived from animal dosimetry data have not been validated. Stabin et al. presented a method for extrapolating biodistribution uptake data from mice to human for dosimetry calculations that was based on two options for interspecies scaling [57, 133].

During the extrapolation of data from mouse to human, several factors have been reported to cause discrepancies between mouse- and human-derived organ-absorbed doses. The variation in size and anatomy, interspecies differences in pharmacokinetics, and methodological differences in biodistribution measurements were considered to be the main factors responsible for such discrepancies. Repetto-Llamazares et al. [131] calculated the absorbed dose in mice from ¹⁷⁷Lu-tetraxetan-tetuloma b and ¹⁷⁷Lu-tetraxetan-rituximab and successfully extrapolated the dosimetry data to a human absorbed dose. Sakata et al. [132] observed that the extrapolation of data from mouse to human is roughly acceptable; however, the differences in the proportions of organ mass to total body mass cause inconsistencies in the predicted dosimetry for human subjects. Hence, dosimetry studies in humans to assess radiation risk and dose-response relationships are essential for clinical trials of newly developed theranostic radiopharmaceuticals.

Future Perspective

Due to the increasing use of preclinical TRT studies for the development of novel theranostic agents, several studies have been performed to accurately estimate absorbed doses to mice at the voxel level using reference mouse phantoms and MC simulations. The fast voxel-level dosimetry approaches developed for clinical TRT studies could also be validated for preclinical voxel-based dosimetry to reduce computation time when longitudinal SPECT/CT imaging of mice is performed. Accurate absorbed dose calculations at the voxel level are necessary to interpret the radiobiological response during personalized therapy of small animals. In EBRT, the methods for evaluating radiobiological effects are well established; however, a dedicated dosimetry method needs to be developed for TRT for radiobiological modeling.

Three-dimensional imaging-based dosimetry applications that have already been applied [85–90] to patient-specific dosimetry should also be upgraded and validated for voxel-based absorbed dose calculations in preclinical TRT. Bednarz et al. [65] recently proposed a preclinical voxel-based dosimetry platform, called RAPID, that is capable of calculating murine-specific 3D dose distributions from theranostic agents used for imaging and therapy. In addition to absorbed doses, dose-volume histograms (DVH) can also be calculated using RAPID to assess the absorbed dose heterogeneity in the tumors. Furthermore, RAPID is currently being upgraded to calculate the biologically effective dose (BED) at the voxel level, which will be useful for radiobiological interpretation and modeling of the dose distribution for response assessment in preclinical TRT. The use of real 3D imaging data of mouse instead of the standard mouse model for MC simulations could further reduce errors in mouse-specific absorbed dose calculations because this approach considers the individualized organ anatomy and activity distributions of each mouse during dosimetry simulations.

Conclusion

The estimation of absorbed doses from radionuclides has become essential for preclinical studies that use small animals because these estimations are essential to the development of novel theranostic agents. The development of more realistic voxel-based preclinical phantoms and user-friendly advanced MC codes has made more accurate absorbed dose calculations at the voxel level possible for preclinical theranostics. Although voxel-based dosimetry using MC simulation is currently the most accurate technique, it is cumbersome and time-consuming. Fast voxel-based absorbed dose calculations based on DPK, multiple VSV, and sMC techniques have been shown to overcome limitations associated with the DMC method in clinical dosimetry. The extension of such methods to preclinical TRT dosimetry would be very useful for optimizing the computation time required to perform absorbed dose calculations. In this review article, we discussed the present scenario of preclinical imaging and dosimetry at the voxel level that is performed using various voxel-based murine models and MC radiation transport codes. We also presented the limitations of various voxel-based dosimetry methods and gave an overview of the possibilities for fast voxel-based absorbed dose calculations in preclinical theranostics.

Funding Information

This work was supported by grants from National Research Foundation of Korea (NRF) funded by the Korean Ministry of Science, ICT and Future Planning (grant no. NRF-2016R1A2B3014645).

Compliance with Ethical Standards

Conflict of Interest

Arun Gupta, Min Sun Lee, Joong Hyun Kim, Dong Soo Lee, and Jae Sung Lee declare that they have no conflict of interest.

Ethical Statement

All procedures performed in studies involving animals were in accordance with the ethical standards of Seoul National University, College of Medicine, Seoul, South Korea.

Informed Consent

For this type of study, informed consent is not required.