WorksPhysics and Astronomy
Spatial scaling of heterogeneity transfer from boron concentration to absorbed dose in boron neutron capture therapy
Abstract
We investigate how the amplitude and spatial correlation length of boron-concentration fluctuations affect the boron-induced absorbed-dose distribution in boron neutron capture therapy (BNCT). The spatial energy-deposition response of alpha particles and lithium nuclei is obtained using Geant4 and convolved with three-dimensional correlated lognormal boron fields under a uniform neutron field. Approximately of the reaction-product energy is deposited within a radius of , with a characteristic deposition length of . Dose heterogeneity increases with both the fluctuation amplitude and correlation length , while the transfer factor exhibits an approximate scaling collapse as a function of . Analytical Gaussian and lognormal–Gaussian models provide reference predictions for this behavior. At short correlation lengths, the Geant4 response yields a higher transfer factor than a Gaussian response with the same characteristic length, highlighting the influence of the energy-deposition shape on dose smoothing. For , defined as the dose received by at least of the modeled volume, the relative difference between homogeneous and heterogeneous boron-dose models reaches at and a target ratio . We show that mean boron concentration alone is insufficient to characterize spatial nonuniformity of the boron-dose component.
Introduction
Boron neutron capture therapy (BNCT) is a radiotherapeutic approach that combines the delivery of compounds containing with neutron irradiation. Following neutron capture, produces alpha particles and nuclei that deposit their energy over distances of only a few micrometers. These short ranges allow energy deposition to occur close to the boron capture sites, making the spatial distribution of boron an important factor in determining the absorbed dose [1, 2, 3]. The development of accelerator-based neutron sources has expanded the clinical application of BNCT, particularly for certain head and neck cancers, while investigations involving brain tumors and other malignancies continue to advance [4, 5, 6, 7, 8, 9, 10, 11].
The effectiveness of BNCT depends not only on the characteristics of the neutron beam but also on the ability of boron delivery agents to reach the target tissue and achieve a suitable distribution of . Compounds such as boronophenylalanine (BPA) and sodium borocaptate (BSH) have long been investigated for this purpose, while the development of targeted molecules, polymers, and nanoparticle-based carriers aims to improve boron selectivity and retention [12, 13, 14, 15, 16, 17, 18, 19, 20, 21]. Nevertheless, boron concentration is not necessarily uniform, either across different regions of a tumor or at the cellular and subcellular levels. Measuring and mapping boron distributions with sufficient spatial detail remains challenging. The mean boron concentration alone may not adequately represent the spatial variations that govern energy deposition [22, 23, 24].
The importance of boron distribution at small spatial scales has been demonstrated through microdosimetric studies. Sato et al. developed a model that accounts for intracellular and intercellular variations in distribution, showing that microscopic dose distributions are important for interpreting the biological effects of BNCT [25]. Subsequent studies have further developed stochastic modeling approaches, particle transport calculations, and methods for estimating biological responses at the cellular scale [26, 27, 28, 29]. Recent microdosimetric investigations and measurements have also emphasized that energy deposition at micrometer scales cannot always be adequately characterized by the mean absorbed dose alone [30, 31, 32]. These findings highlight the relevance of boron heterogeneity to BNCT dosimetry, although the spatial distribution of absorbed dose should be distinguished from direct predictions of biological response.
Monte Carlo (MC) methods are widely used in BNCT treatment planning to calculate the contributions from different radiation components. Advances in computational tools and image-based modeling have improved the evaluation of absorbed-dose distributions [34, 35, 36] [33]. However, some dosimetric approaches still assume uniform boron concentrations or uptake ratios within specified regions. Teng et al. demonstrated that accounting for heterogeneous boron distributions can alter calculated tumor doses [37]. Subsequent studies using -BPA positron emission tomography (PET) have also reported differences in dosimetric indices when voxel-level boron variations are considered, while highlighting limitations associated with image resolution and segmentation [38, 39, 40]. At the same time, BNCT dose evaluation must account for the neutron field and radiation components other than those arising from boron neutron capture [41, 42].
These studies establish the importance of boron heterogeneity but leave a more fundamental physical question: how does the spatial scale of boron concentration fluctuations determine the degree of dose smoothing caused by alpha-particle and lithium-nucleus transport? Two boron distributions may have the same mean concentration and coefficient of variation while differing in their spatial correlation lengths, which characterize the typical sizes of boron-rich and boron-poor regions. Because the reaction products deposit energy over finite distances, small-scale boron fluctuations are expected to undergo stronger spatial smoothing than larger structures. This behavior cannot be determined from the mean concentration or the amplitude of boron heterogeneity alone without considering the spatial characteristics of energy deposition. A framework that independently controls the heterogeneity amplitude, boron correlation length, and energy-deposition response is therefore needed.
In this study, these effects are investigated using correlated lognormal boron concentration fields and a spatial energy-deposition response obtained from Geant4 simulations [43, 44]. The boron-dose component is calculated through three-dimensional convolution in a homogeneous medium under the assumption of a uniform neutron field. To quantify how variations in boron concentration are transferred to the absorbed-dose distribution, we define a heterogeneity transfer factor, , where is the coefficient of variation of boron concentration and is the corresponding coefficient of variation of absorbed dose. The dependence of on the ratio , where is the boron correlation length and is the characteristic energy-deposition length, is then compared with an ideal Gaussian prediction. By applying the Geant4 and Gaussian responses to identical boron concentration fields, the influence of the spatial shape of energy deposition can be examined directly. Finally, the dosimetric consequences are evaluated using , defined as the dose received by at least 95% of the modeled volume. This quantity is compared with the corresponding result obtained under the assumption of a homogeneous boron distribution with the same mean concentration. The present study focuses on the relationship between the spatial structure of boron concentration and the absorbed-dose component arising from boron neutron capture, rather than on complete clinical dose calculations or the biological effectiveness of BNCT. Within this framework, we aim to identify the conditions under which the homogeneous-boron approximation may underestimate dose nonuniformity and to clarify how spatial correlation and the shape of the energy-deposition response jointly influence the resulting dose distribution.
Model and Methods
Spatial model of boron distribution
This study employs a computational approach to investigate how spatial heterogeneity in boron concentration influences energy deposition in boron neutron capture therapy (BNCT). To isolate this effect, tissue is represented by a homogeneous medium with fixed density and composition, while the concentration is allowed to vary spatially. The neutron field is assumed to be uniform throughout the computational domain, so variations in absorbed dose arise from the spatial distribution of boron and the transport of reaction-product energy. The model is not intended to reproduce patient-specific anatomy but rather to examine how the amplitude and spatial scale of boron heterogeneity influence the resulting dose nonuniformity.
The boron concentration is represented by a three-dimensional scalar field, , with mean value . To generate heterogeneous distributions with strictly positive concentrations, a correlated lognormal random field is used:
Here, is a stationary Gaussian random field with zero mean and unit variance. Its spatial correlation function is given by
where denotes the correlation length of the underlying Gaussian field. The degree of boron heterogeneity is characterized by the coefficient of variation,
where is the standard deviation of boron concentration. The lognormal parameter is determined by , ensuring that the prescribed mean concentration and heterogeneity amplitude are independently controlled.
The spatial correlation of the boron concentration field differs from that of the underlying Gaussian field because of the lognormal transformation. In particular, the normalized autocorrelation function of can be expressed analytically as
We define the boron correlation length as the distance at which . This definition distinguishes the spatial scale of the boron concentration fluctuations from the correlation length used to generate the Gaussian field. For a prescribed heterogeneity amplitude and target correlation length , the corresponding Gaussian correlation length is therefore
This relation accounts for the change in spatial correlation introduced by the lognormal transformation and allows the prescribed heterogeneity amplitude and boron correlation length to be varied independently. In the limit of weak heterogeneity, approaches , whereas their difference becomes more pronounced as increases.
In the numerical analysis, the prescribed mean boron concentration is kept constant, while and are varied systematically. For each parameter combination, the Gaussian correlation length is determined from Eq. (5), and independent random realizations of the boron concentration field are generated. The actual correlation length of each realization is then evaluated from its spatial autocorrelation function to account for finite-domain fluctuations and numerical discretization. Two boron distributions can have the same mean concentration and heterogeneity amplitude but different spatial extents of boron-rich and boron-poor regions. By independently controlling these properties and averaging over multiple realizations, the effects of heterogeneity amplitude and correlation length on the absorbed-dose distribution can be examined separately.
Particle transport and spatial energy-deposition response
The transport of charged particles produced by neutron capture in is simulated using the MC toolkit Geant4 [43, 44]. The reaction considered is
which produces charged particles with ranges of only a few micrometers in tissue. In the dominant reaction branch, with a probability of approximately 93.7%, the alpha particle and lithium nucleus have kinetic energies of 1.47 and 0.84 MeV, respectively. The remaining branch, with a probability of approximately 6.3%, produces an alpha particle and a lithium nucleus with kinetic energies of approximately 1.78 and 1.01 MeV, respectively. Both branches are included according to their respective probabilities. Previous MC track-structure studies have also examined the influence of lithium-ion transport and interaction cross sections on BNCT microdosimetry [45]. The present study considers only energy deposition from alpha particles and lithium nuclei, excluding gamma-ray contributions and other neutron-induced interactions.
Particle transport is simulated in water as a simplified representation of biological tissue. Reaction products are generated at a single point at the center of the medium, with emission directions sampled isotropically. Each particle species and reaction branch is simulated separately, and the deposited energy is recorded as a function of distance from the reaction site using concentric spherical shells. The radial contributions are combined according to the corresponding branching probabilities to construct the normalized spatial energy-deposition response, . This response describes the average spatial distribution of energy deposited by the charged reaction products around a neutron-capture site.
Figure 1 illustrates the spatial energy-deposition response obtained from the Geant4 simulations. In Figure 1(a), energy deposition from nuclei is more concentrated near the reaction site, whereas alpha particles deposit energy over longer distances. This difference is also evident in the cumulative energy distributions shown in Figure 1(b), where approximately 90% of the total energy is deposited within a radius of . These results demonstrate that the reaction products deposit energy over micrometer-scale distances, suggesting that the spatial extent of boron concentration variations is important in determining the resulting dose nonuniformity.
Figure 1. Spatial distribution of energy deposited by alpha particles () and lithium nuclei () produced in the reaction, obtained from Geant4 MC simulations in water. (a) Normalized radial energy-deposition responses for the individual particle species and their combined contribution. (b) Cumulative fraction of deposited energy as a function of distance from the reaction site. The dashed horizontal line indicates the 90% energy fraction, while the dotted vertical line marks . The results show that energy deposition is localized, with alpha particles exhibiting longer ranges than lithium nuclei.
To relate boron concentration to the absorbed-dose distribution, the neutron field is assumed to be spatially uniform, and changes in boron concentration are assumed not to significantly perturb the neutron distribution. Under these conditions, the local neutron-capture rate is proportional to . Because the medium is homogeneous, the energy-deposition response depends only on the relative displacement between the reaction site and the energy-deposition location. The absorbed-dose distribution arising from the boron component can therefore be expressed as a convolution,
where is a normalization factor incorporating neutron fluence, capture probability, mean reaction-product energy, and conversion to absorbed dose. This formulation allows dose distributions for different boron configurations to be calculated through numerical convolution without repeating the particle transport simulations for each realization. Its accuracy could be further assessed by comparison with direct particle transport simulations for selected heterogeneous configurations.
Analytical theory and numerical analysis
The relationship between boron heterogeneity and absorbed dose is examined by separating each field into its mean and fluctuating components. Based on the convolution model in Eq. (7), the Fourier transforms of the dose and boron concentration fluctuations satisfy
This relation shows that particle transport acts as a spatial filter on boron concentration fluctuations. Variations occurring on spatial scales smaller than the energy-deposition range tend to be smoothed, whereas larger-scale variations are more readily preserved in the absorbed-dose distribution. Defining and as the power spectra of boron concentration and dose fluctuations, respectively, gives
This spectral relation connects the spatial structure of boron concentration with the resulting absorbed-dose fluctuations, with the energy-deposition response determining how strongly fluctuations at different spatial scales are attenuated.
To quantify the transfer of boron heterogeneity to the absorbed-dose distribution, we introduce the heterogeneity transfer factor
where is the standard deviation of absorbed dose. A small value of indicates that particle transport substantially smooths spatial variations in boron concentration, whereas approaching unity indicates that dose heterogeneity closely follows the underlying boron distribution. The transfer factor is defined for , with the homogeneous boron distribution serving as the reference condition. For a normalized energy-deposition response, and neglecting finite-domain effects, the transfer factor can be expressed as
(11) shows that heterogeneity smoothing depends on both the spatial spectrum of boron concentration and the shape of the energy-deposition response. The transfer of heterogeneity cannot generally be characterized by the fluctuation amplitude or a single correlation length without specifying the underlying spatial correlation structure.
To obtain a simple analytical prediction, we first consider a reference model with a Gaussian boron concentration covariance and a Gaussian energy-deposition response characterized by a length scale . In three dimensions, integration of (11) yields
This expression predicts that heterogeneity transfer is governed by the ratio , which compares the boron correlation length with the characteristic energy-deposition length. When , small-scale boron fluctuations undergo substantial smoothing, whereas for , energy deposition becomes relatively localized within the boron structures and approaches unity. (12) is exact for the assumed Gaussian covariance and response, but it does not fully describe the correlated lognormal boron fields used in the numerical calculations.
To account for the lognormal structure of the boron concentration field, we extend the analytical treatment while retaining the Gaussian energy-deposition response. Using the spatial correlation model defined in Sec. 2.1, the resulting heterogeneity transfer factor can be expressed as
where and is the correlation length of the underlying Gaussian field. As established in Sec. 2.1, is related to the boron correlation length through the lognormal transformation. Unlike the Gaussian reference prediction, this expression retains the dependence on heterogeneity amplitude through the spatial covariance of the lognormal field. In the limit of weak heterogeneity, approaches , and (13) reduces to (12). The two analytical expressions therefore provide complementary descriptions of heterogeneity transfer. The Gaussian reference model describes the dependence on the correlation-length ratio for an ideal Gaussian covariance, while the lognormal–Gaussian model additionally accounts for changes in the correlation structure associated with the lognormal transformation. Their derivations are presented in Appendix A.
In the numerical calculations, the characteristic energy-deposition length is determined from the second moment of the spatial response, . The Geant4 simulations, performed with a radial resolution of , yield . The actual boron correlation length, , is determined from the normalized autocorrelation function of the boron concentration field as the distance at which the correlation decreases to . This measured value is distinguished from the target correlation length specified during field generation to account for finite-domain fluctuations and numerical discretization. Absorbed-dose distributions are calculated through convolution using three-dimensional fast Fourier transforms with periodic boundary conditions. Spatial visualizations use a grid with a domain length of along each direction, whereas quantitative analyses use a grid with a domain length of . Different combinations of heterogeneity amplitude and correlation-length ratio are examined to characterize their effects on the resulting dose distribution.
To assess the influence of the spatial shape of energy deposition, additional calculations are performed using a Gaussian response with the same characteristic length as the Geant4 response. Both responses are applied to identical realizations of the lognormal boron concentration field, allowing their effects on the transfer factor to be compared directly. The analytical predictions in Eqs. (12) and (13) provide reference results for examining the influence of boron correlation structure and energy-deposition response. In particular, the lognormal–Gaussian prediction provides an analytical benchmark for calculations using the Gaussian response, while the comparison between the numerical Gaussian and Geant4 responses isolates the contribution of the energy-deposition kernel shape. For each parameter combination in the main quantitative analyses, 10 independent random realizations are generated. Results are reported as ensemble averages, with statistical uncertainties expressed as standard errors of the mean.
The dosimetric consequences of boron heterogeneity are further evaluated using , defined as the fifth percentile of the absorbed-dose distribution, corresponding to the dose received by at least 95% of the modeled volume. To enable comparisons across parameter combinations, the dose distribution is normalized by its mean, . Under this normalization, a homogeneous boron distribution produces a relative dose of unity throughout the domain. The difference in between homogeneous and heterogeneous boron models is therefore used to assess the effect of assuming a uniform boron concentration on the coverage of the boron-dose component. Unlike the transfer factor, which characterizes the overall amplitude of dose fluctuations, is sensitive to the lower tail of the dose distribution and can therefore reveal differences that are not fully captured by the coefficient of variation alone.
The numerical stability of the three-dimensional dose calculation is assessed separately from the convergence of the Geant4 energy-deposition response. Grid refinement is examined at a fixed physical domain size, whereas finite-domain sensitivity is evaluated by increasing the domain size while keeping the spatial grid spacing unchanged. The tests focus on and a target correlation-length ratio , corresponding to the largest relative difference in within the investigated parameter range. The reference
configuration of grid points and a domain length of is compared with a grid at the same domain length and with a grid in a domain. For each configuration, 20 independent random realizations are generated, and the transfer factor and relative difference in are evaluated to assess their sensitivity to the numerical settings. These comparisons provide a measure of numerical robustness under grid refinement and domain enlargement, while allowing statistical fluctuations between independent realizations to be taken into account. The numerical procedures and convergence results are presented in Appendix B, with the quantitative comparisons summarized in Table 1.
| () | () | (%) | ||
| 512 | 128 | 0.25 | 0.95691(77) | 87.08(71) |
| 640 | 128 | 0.20 | 0.95649(91) | 86.59(56) |
| 640 | 160 | 0.25 | 0.95762(68) | 85.95(56) |
Table 1. Numerical sensitivity of the heterogeneity transfer factor and relative difference for and target . Results are ensemble averages over 20 independent realizations. Parentheses indicate standard errors in the last reported digits.
Results and Discussion
Energy deposition and numerical stability
The spatial characteristics of energy deposition from the reaction products are presented in Figure 1. The radial distribution in panel (a) shows that energy deposition from nuclei is more concentrated near the reaction site, whereas alpha particles deposit energy over longer distances. The combined energy-deposition response generally decreases with radial distance, reflecting the finite ranges of these charged particles in the medium. The absorbed dose at a given location depends not only on the local boron concentration but also on energy deposited by reactions occurring in the surrounding region. The cumulative distribution in Figure 1(b) further shows that most of the reaction-product energy is deposited within a few micrometers of the capture site. At a radial resolution of , the radii containing 50%, 90%, and 99% of the deposited energy are 2.49, 6.00, and 7.91 , respectively. Only approximately 0.0028% of the recorded energy is deposited beyond 11 , indicating that the energy-deposition response is strongly localized around the neutron-capture site. This spatial localization provides the physical basis for examining how boron concentration fluctuations are smoothed over distances comparable to the ranges of the reaction products.
The numerical stability of the energy-deposition response was assessed by comparing two Geant4 simulations, each using particle histories per source but different radial bin widths and maximum transport step lengths. Reducing the radial bin width from 0.10 to 0.05 and the maximum step length from 0.05 to 0.025 resulted in relative changes of approximately 0.12% and 0.13% in and , respectively. The characteristic energy-deposition length, , calculated from the second moment of the spatial response, changed by only about 0.075%, with the finer simulation yielding . These small differences indicate that the principal spatial characteristics of the energy-deposition response are relatively insensitive to the numerical settings examined. Unlike , which specifies the radius containing a given fraction of deposited energy, characterizes the effective spatial width of the response through its second moment. This distinction is important because the smoothing of boron-induced dose heterogeneity depends on the full spatial distribution of deposited energy rather than on a single measure of particle range. The second-moment definition also provides a consistent basis for comparing the Geant4 response with a Gaussian response of the same characteristic length in § 3.3.
The numerical stability of the three-dimensional dose calculation was examined separately from that of the Geant4 energy-deposition response. In particular, the grid spacing affects the representation of short-range boron fluctuations, whereas the computational domain size can influence the statistical representation of larger spatial structures and the effects of periodic boundary conditions. To evaluate these effects, additional calculations were performed for and a target correlation-length ratio , corresponding to the largest relative difference in within the investigated parameter range. Grid-resolution sensitivity was examined by increasing the grid size from to at a fixed domain length of , while finite-domain sensitivity was evaluated by increasing the domain length from to at a fixed grid spacing of . Each numerical configuration was evaluated using 20 independent random realizations, and the resulting transfer factor and relative difference in were compared.
As summarized in Table 1 of Appendix § 10, grid refinement changed the ensemble-averaged transfer factor by approximately 0.043%, whereas increasing the domain size changed it by approximately 0.074%. The corresponding differences in were 0.49 and 1.13 percentage points, respectively. These results indicate that the transfer factor is relatively stable under the numerical changes examined, while the lower dose percentile exhibits greater variability. The differences in were comparable to the statistical uncertainties of the ensemble estimates, reflecting the sensitivity of the lower tail of the dose distribution to individual spatial realizations. Although these comparisons do not establish a strict bound on finite-domain errors, they show that the substantial reduction in persists under grid refinement and domain enlargement. Together with the stability of the Geant4 energy-deposition response, these findings support the numerical robustness of the principal results within the tested configurations.
Boron heterogeneity transfer to dose
The influence of spatial boron heterogeneity on absorbed-dose nonuniformity is first examined by comparing the boron concentration field, , with the corresponding absorbed-dose distribution, . Calculations are performed on a three-dimensional grid with a domain length of along each direction. The mean boron concentration and heterogeneity amplitude are held constant at , while the correlation length is varied to obtain target ratios of , 2.0, and 5.0. This allows the effect of the spatial extent of boron structures to be examined independently of their heterogeneity amplitude. Figure 2 shows two-dimensional slices of the boron concentration fields and their corresponding absorbed-dose distributions, calculated through convolution with the Geant4 energy-deposition response using Fourier transforms. Both concentration and dose are normalized by their respective mean values and displayed using a common color scale, allowing their spatial patterns to be compared directly.
Figure 2. Comparison of spatial boron concentration and absorbed-dose distributions for three different correlation lengths. Panels (a)–(c) show the boron concentration fields, while panels (d)–(f) show the corresponding absorbed-dose distributions calculated by convolution with the Geant4 energy-deposition response. All configurations have the same boron heterogeneity amplitude, , with target correlation-length ratios , , and , from left to right. Calculations are performed on a three-dimensional grid. Concentration and dose are normalized by their respective mean values and displayed using a common color scale.
At the shortest target correlation-length ratio, , the boron concentration field in Figure 2(a) contains numerous small boron-rich and boron-poor regions. After the reaction-product energy spreads into the surrounding medium, these spatial variations become considerably smoother, as shown in Figure 2(d). In particular, regions with high boron concentrations no longer produce equally sharp dose peaks, indicating substantial spatial smoothing when the characteristic size of boron fluctuations is comparable to or smaller than the energy-deposition length. At , larger boron-rich and boron-poor structures emerge, as illustrated in Figure 2(b). The corresponding dose distribution in panel (e) still follows the locations of boron-rich regions, although their boundaries become smoother. Spatial averaging therefore remains evident but is less pronounced than at the shorter correlation length.
At the largest target ratio, , the differences between the boron concentration and dose distributions become smaller. Figure 2(c) and (Figure 2f) show nearly identical spatial patterns, indicating that the characteristic size of the boron structures is substantially larger than the energy-deposition length. In this regime, most of the reaction-product energy is deposited within regions where boron concentration varies relatively little over the particle transport scale. These comparisons demonstrate that the correlation length strongly influences dose nonuniformity even when the mean boron concentration and heterogeneity amplitude remain unchanged. Short-range fluctuations undergo stronger smoothing, whereas larger spatial structures are more readily preserved in the resulting dose distribution. Thus, dose heterogeneity depends on both the amplitude and the spatial scale of boron concentration fluctuations.
To quantify this behavior, the coefficient of variation of absorbed dose, , is calculated for different heterogeneity amplitudes, , and correlation-length ratios, . Figure 3 shows as a function of for four target correlation-length ratios, , 1, 2, and 5. The calculations use a three-dimensional grid with a domain length of . Each data point represents the average of 10 independent random realizations, with error bars indicating the standard error of the mean. The dashed line provides a reference for the case in which boron heterogeneity is transferred to the absorbed-dose distribution without attenuation.
Figure 3. Coefficient of variation of absorbed dose, , as a function of boron concentration heterogeneity, , for four target correlation-length ratios, , 1, 2, and 5. Dose distributions are calculated by convolving three-dimensional boron concentration fields with the Geant4 energy-deposition response. Symbols represent averages over 10 independent random realizations, and error bars indicate the standard error of the mean. The dashed line represents the reference relation .
As shown in Figure 3, increases approximately linearly with for all correlation lengths considered, although the rate of increase depends strongly on the correlation length. At , dose heterogeneity is substantially smaller than boron heterogeneity, with at . For the same heterogeneity amplitude, increases to approximately 0.75 at , demonstrating that larger boron structures retain spatial dose variations more effectively. All curves remain below the reference line , indicating that particle transport reduces the amplitude of spatial fluctuations in the resulting dose distribution. This reduction is strongest at short correlation lengths and becomes progressively weaker as increases. Using the definition in Eq. (10), the corresponding transfer factors at are and for and , respectively. These results show that an increasing fraction of boron heterogeneity is preserved in the absorbed-dose distribution as the correlation length increases.
The approximately linear dependence of on indicates that the heterogeneity amplitude controls the overall magnitude of dose fluctuations, while the different slopes demonstrate that its influence depends on the spatial correlation length. This behavior is consistent with the spatial smoothing observed in Figure 2, where the energy-deposition response suppresses small-scale fluctuations more strongly than large-scale variations. To further characterize this relationship, the transfer factor is examined as a function of . This representation allows us to assess whether the transfer of heterogeneity follows an approximate scaling relation and to compare the numerical results with the analytical prediction of the ideal Gaussian model.
Scaling and homogeneous-boron limitations
To further characterize the spatial transfer of boron heterogeneity, the transfer factor is examined as a function of , which compares the boron correlation length with the characteristic energy-deposition length. Figure 4 presents the numerical results for three heterogeneity amplitudes, , , and , together with the analytical Gaussian prediction in Eq. (12). To isolate the influence of the spatial shape of energy deposition, two response functions are considered: the Geant4 response and a Gaussian response with the same characteristic length . Both are applied to identical realizations of the lognormal boron concentration fields, allowing their effects on the transfer factor to be compared directly. The ideal Gaussian prediction provides a simple reference for the dependence on , while the lognormal–Gaussian expression in Eq. (13) accounts for the spatial covariance of the boron fields used in the numerical calculations. These two analytical descriptions help distinguish the influence of boron correlation structure from that of the energy-deposition response.
Figure 4. Scaling behavior of the heterogeneity transfer factor, , as a function of the measured correlation-length ratio for three boron heterogeneity amplitudes, , 0.8, and 1.2. Filled symbols connected by solid lines represent the Geant4 response, while open symbols connected by dotted lines represent the Gaussian response with the same characteristic length. The dashed line shows the analytical prediction for the ideal Gaussian covariance and response. Both numerical calculations use identical lognormal boron fields. The inset shows the difference between the two numerical transfer factors, .
As shown in Figure 4, the transfer factor increases monotonically with . At short correlation lengths, boron concentration fluctuations are strongly smoothed by particle transport, resulting in relatively low values of . As the correlation length increases, the spatial extent of boron structures becomes larger than the energy-deposition scale, and the absorbed-dose distribution more closely follows the underlying boron concentration field. This behavior is consistent with the limiting prediction for and for . Moreover, the curves for the three heterogeneity amplitudes nearly overlap over the investigated range, indicating that the transfer factor is primarily controlled by , with a relatively weak dependence on . This behavior can also be understood from the analytical lognormal–Gaussian model in Eq. (13), which retains the dependence on but predicts only modest variations in for the parameters considered. The approximate scaling collapse therefore reflects the dominant role of the correlation-length ratio within the present model, rather than an exact independence from heterogeneity amplitude or a universal scaling law.
Despite their similar overall trends, the Geant4 and Gaussian responses produce different transfer factors at short correlation lengths. At , the Geant4 response yields , compared with approximately 0.20 for the numerical Gaussian response and 0.19 for the ideal analytical prediction. The lognormal–Gaussian expression provides a more appropriate analytical reference for the numerical Gaussian response because it retains the covariance of the underlying lognormal boron field. For example, at , this expression predicts for and approximately 0.21 for . These values indicate that the effect of lognormal correlation structure is relatively small in this regime compared with the difference between the two energy-deposition responses. The inset of Figure 4 shows that this difference reaches at the shortest correlation length and decreases rapidly as increases. Because both numerical responses are applied to identical boron fields, their difference can be attributed to the spatial shape of the energy-deposition kernel rather than to changes in the boron distribution. In spectral terms, short-range boron fluctuations contain stronger contributions from spatial frequencies at which the Geant4 and Gaussian responses attenuate fluctuations differently. Thus, matching the characteristic energy-deposition length alone does not guarantee equivalent smoothing, particularly when the spatial scale of boron heterogeneity is comparable to or smaller than the particle transport range.
The dosimetric consequences of boron heterogeneity are evaluated using , defined as the fifth percentile of the absorbed-dose distribution, corresponding to the dose received by at least 95% of the modeled volume. This quantity is compared between homogeneous and heterogeneous boron distributions with the same mean concentration. Their relative difference is defined as
where and denote the values obtained from the homogeneous and heterogeneous models, respectively. A positive indicates that the homogeneous-boron approximation overestimates relative to the heterogeneous model. Figure 5 shows for different heterogeneity amplitudes and target correlation-length ratios. The calculations use the Geant4 energy-deposition response on a three-dimensional grid with a domain length of 128 m, and each value represents an average over 10 independent random realizations. Because the dose distributions are normalized by their respective mean values, these comparisons isolate changes in the spatial distribution of the boron-dose component rather than changes in its average value.
Figure 5. Relative difference in between homogeneous and heterogeneous boron distributions, , as a function of heterogeneity amplitude and the target correlation-length ratio . Dose distributions are calculated by convolving three-dimensional boron concentration fields with the Geant4 energy-deposition response. The color scale represents the mean over 10 independent random realizations for each parameter combination. The largest relative difference is approximately 86.6% at and .
As shown in Figure 5, the relative difference in increases with both heterogeneity amplitude and correlation length. At a target ratio , increases from approximately 22.8% at to 64.6% at and 77.7% at . The correlation-length dependence is also pronounced: at , increases from approximately 30.4% at to 74.5% at . At short correlation lengths, particle transport smooths differences between boron-rich and boron-poor regions, reducing the spatial variation in absorbed dose. As the correlation length increases, however, boron-poor regions extend over larger distances relative to the energy-deposition range and can retain lower absorbed doses. This behavior reduces the lower percentiles of the dose distribution, even when the mean dose remains unchanged. The largest difference within the investigated parameter range occurs at and a target ratio , where , with the uncertainty expressed as the standard error in percentage points. Under these conditions, in the heterogeneous model is only approximately 13.4% of the corresponding homogeneous value.
The behavior at large correlation lengths can also be understood from the underlying lognormal distribution. When becomes much larger than the energy-deposition length, spatial smoothing becomes weak and the normalized dose distribution approaches the normalized boron concentration distribution. In this limit, the fifth percentile follows directly from the lognormal model in Eq. (1), giving
where is the inverse cumulative distribution function of the standard normal distribution. For , this expression predicts , which is close to the largest numerical value obtained in Figure 5. The agreement is consistent with the expected reduction of spatial smoothing at large correlation lengths, although it does not establish that the asymptotic limit has been fully reached within the simulated domain. This analytical limit also provides a useful reference for interpreting the magnitude of the changes in without introducing additional assumptions about the biological response to radiation.
The results in Figure 5 complement the scaling behavior in Figure 4. Although the transfer factor depends only weakly on over the investigated range, the relative difference in remains strongly dependent on heterogeneity amplitude. This distinction arises because characterizes the overall amplitude of dose fluctuations, whereas is determined by the lower tail of the dose distribution. Similar transfer factors do not necessarily imply equivalent dose coverage, particularly when the underlying concentration distributions differ in their fluctuation amplitude. More generally, these findings demonstrate that the mean boron concentration alone is insufficient to characterize absorbed-dose nonuniformity in a heterogeneous medium. Both the amplitude and spatial correlation length of boron fluctuations, together with the shape of the energy-deposition response, contribute to the resulting dose distribution. Because the present model considers only the boron-dose component in a homogeneous medium under a uniform neutron field, the calculated differences in should not be interpreted as differences in total clinical BNCT dose or biological effectiveness. Instead, they provide a physical basis for assessing the limitations of the homogeneous-boron approximation in spatial dosimetric analysis.
Conclusions
This study investigated the influence of boron heterogeneity amplitude and spatial correlation length on absorbed-dose nonuniformity in boron neutron capture therapy (BNCT). By combining Geant4 particle transport simulations, correlated lognormal boron concentration fields, and three-dimensional convolution, we examined how the spatial structure of boron concentration affects the resulting absorbed-dose distribution. The Geant4 simulations showed that approximately 90% of the reaction-product energy is deposited within a radius of 6.00 , with a characteristic energy-deposition length of . This length provides a characteristic scale for understanding how charged-particle transport smooths boron concentration variations at the micrometer scale.
The numerical results demonstrated that dose heterogeneity depends on both the amplitude and spatial correlation length of boron concentration fluctuations. Small-scale fluctuations undergo stronger smoothing, whereas larger boron structures are more readily preserved in the absorbed-dose distribution. The heterogeneity transfer factor, , exhibited an approximate scaling collapse as a function of , with relatively weak dependence on heterogeneity amplitude over the investigated parameter range. However, comparison with the Gaussian reference model revealed that the characteristic energy-deposition length alone does not fully determine the transfer of boron heterogeneity. At , the Geant4 response yielded , compared with approximately 0.20 for a Gaussian response with the same characteristic length. The difference decreased as the correlation length increased, highlighting the importance of the spatial shape of energy deposition, particularly at short correlation lengths.
The dosimetric consequences were further examined using , whose relative difference between homogeneous and heterogeneous boron models increased with both heterogeneity amplitude and correlation length. At and a target correlation-length ratio of , the relative difference reached , where the uncertainty represents the standard error in percentage points. This result demonstrates that assuming a uniform boron concentration can substantially overestimate when the underlying distribution is heterogeneous. Nevertheless, because the present model considers only the boron-dose component in a homogeneous medium under a uniform neutron field, the calculated differences cannot be directly interpreted as changes in total clinical BNCT dose. Our results establish a physical basis for assessing the limitations of the homogeneous-boron approximation and emphasize the combined roles of heterogeneity amplitude, spatial correlation length, and energy-deposition response shape in determining absorbed-dose nonuniformity.
Data Availability
The simulation data and source codes supporting the results of this study are available from the corresponding author upon reasonable request.
CRediT authorship contribution statement
R. Muslim: Conceptualization, Methodology, Writing – original draft, Software, Formal analysis, Validation, Visualization, Writing – review & editing. B. Pidhatika: Conceptualization, Formal analysis, Writing – original draft, Validation, Writing – review & editing. S. Sutanto: Formal analysis, Validation, Writing – review & editing. R. Fardela: Formal analysis, Validation, Writing – review & editing.
R. N. Wijaya: Formal analysis, Validation, Writing – review & editing. Writing – review & editing. All authors have read, reviewed, and approved the publication of this paper.
Declaration of Interests
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgments
R. Muslim acknowledges support from the YST Program of the Asia Pacific Center for Theoretical Physics (APCTP), funded by the Science and Technology Promotion Fund and Lottery Fund of the Korean Government, and by the Management Talent Program of the National Research and Innovation Agency of Indonesia (BRIN).
Gaussian and lognormal scaling theory
In this section, we derive the main analytical expressions presented in the main text. Starting from the convolution model in Eq. (7), the normalized energy-deposition response satisfies , which gives . The Fourier transforms of the fluctuating components are related by
The corresponding variances are
Using and , we obtain
which reproduces Eq. (11).
For the Gaussian reference model, the boron concentration covariance is
Its three-dimensional Fourier transform gives
The normalized Gaussian energy-deposition response is
where . Its Fourier transform is
Substituting Eq. (21) and Eq. (23) into Eq. (19) yields
where . Taking the positive square root gives
in agreement with Eq. (12).
We consider the correlated lognormal field defined in Eq. (1). Let , , and
Since and are jointly Gaussian with zero mean and unit variance,
where . It follows that
Subtracting the squared mean concentration gives
Expanding the exponential, we obtain
Using the Fourier transform identity
with , Eq. (30) gives
For the Gaussian response, . The contribution of each term to the normalized dose variance involves the integral
Substituting Eq. (32) and Eq. (23) into Eq. (17), and using Eq. (33), we obtain
At , the boron variance follows directly from Eq. (29),
Therefore,
which reproduces Eq. (13).
The Gaussian reference result is recovered in the weak-heterogeneity limit. For ,
Keeping the leading term in Eq. (36),
Thus,
Both models also satisfy for and for , under their respective continuum assumptions.
Numerical implementation
The numerical procedure consists of two stages: obtaining the spatial energy-deposition response from Geant4 simulations and calculating absorbed-dose distributions by convolving this response with heterogeneous boron concentration fields. This separation allows different boron configurations to be examined without repeating particle transport simulations and enables the effects of boron heterogeneity and energy deposition to be investigated independently.
Alpha particles and lithium nuclei are transported in water using particle histories for each particle species and reaction branch. Deposited energy is scored in concentric spherical shells with a radial bin width of and a maximum transport step length of . Contributions from the reaction branches are combined according to their branching probabilities to construct the total energy-deposition response . The energy deposited in each radial shell, , is normalized as , satisfying . These weights describe the relative spatial distribution of deposited energy independently of the total reaction-product energy.
For an isotropic response, the three-dimensional Fourier transform depends only on ,
The integral is evaluated discretely using the normalized radial energy weights. The condition ensures that convolution preserves the mean boron-dose component. The characteristic energy-deposition length is obtained from the second moment, , providing a common spatial scale for comparing the Geant4 and Gaussian responses.
Boron concentration fields are generated on a three-dimensional periodic grid. A correlated Gaussian field is constructed by filtering Gaussian random noise in Fourier space using the correlation spectrum corresponding to Eq. (2), consistent with spectral approaches for simulating stationary Gaussian random fields [46]. The field is normalized to zero mean and unit variance before applying the lognormal transformation in Eq. (1). For prescribed heterogeneity amplitude and target correlation length , the parameters and are determined from the lognormal model and the correlation-length mapping in § 2.1. The lognormal parameter is adjusted to reproduce the prescribed heterogeneity amplitude, and each boron field is normalized by its spatial mean. The actual correlation length is measured from the normalized spatial autocorrelation function at its crossing. Comparing the measured and target values allows finite-domain fluctuations and discretization effects to be assessed.
The absorbed-dose distribution is calculated by Fourier-space convolution,
where and denote the forward and inverse Fourier transforms. The convolution is evaluated using three-dimensional fast Fourier transforms implemented with the FFTW3 library [47], under periodic boundary conditions. Since dose distributions are normalized by their spatial means, the factor does not affect , the transfer factor , or the relative difference in .
The Geant4 and Gaussian responses are applied to identical boron realizations to isolate the influence of the energy-deposition kernel shape. The spatial visualizations in Figure 2 use a grid with a domain length of , corresponding to a grid spacing of . The quantitative calculations in Figure 3, Figure 4, and Figure 5 use a grid with a domain length of and a spacing of . For each parameter combination, results are averaged over independent random realizations. Statistical uncertainties are reported as standard errors of the mean,
where denotes the observable evaluated for each realization. The dose-coverage metric is defined as the fifth percentile of the dose distribution normalized by its mean. Since the normalized dose equals unity for homogeneous boron concentration, the relative difference between homogeneous and heterogeneous models is
where is the fifth percentile of the normalized heterogeneous dose. All comparisons therefore use the same mean boron concentration.
Numerical sensitivity was assessed through grid refinement at a fixed domain size and domain enlargement at a fixed grid spacing. The tests focused on and a target ratio , corresponding to the largest relative difference in within the investigated range. This long-correlation case also probes the sensitivity of low-dose regions to finite-domain effects. Each configuration was evaluated using 20 independent realizations, with the transfer factor and relative difference used to quantify numerical sensitivity.
As shown in Table 1, refining the grid from to at changed the ensemble-averaged by approximately 0.043% and by 0.49 percentage points. Increasing the domain length from 128 to 160 at a fixed spacing of 0.25 changed by 0.074% and by 1.13 percentage points. The corresponding measured correlation lengths were 11.874, 11.832, and 12.002 for the three configurations, respectively, compared with the target value of 12.289 . These deviations and the use of independent random realizations introduce ensemble variability into the numerical comparisons. Nevertheless, remained relatively stable, while the larger variations in were comparable to its statistical uncertainties. The substantial reduction in persisted under both grid refinement and domain enlargement, supporting the numerical robustness of the principal findings within the tested configurations, although a strict bound on finite-domain errors in cannot be established.
References
- [1]R. F. Barth, J. A. Coderre, M. G. H. Vicente, T. E. Blue, Boron neutron capture therapy of cancer: Current status and future prospects, Clinical Cancer Research 11 (11) (2005) 3987–4002. doi:10.1158/1078-0432.CCR-05-0035.DOI
- [2]R. F. Barth, M. G. H. Vicente, et al., Current status of boron neutron capture therapy of high grade gliomas and recurrent head and neck cancer, Radiation Oncology 7 (2012) 146. doi:10.1186/1748-717X-7-146.DOI
- [3]M. A. Dymova, S. Y. Taskaev, V. A. Richter, E. V. Kuligina, Boron neutron capture therapy: Current status and future perspectives, Cancer Communications 40 (2020) 406–421. doi:10.1002/cac2.12089.DOI
- [4]K. Hirose, A. Konno, J. Hiratsuka, et al., Boron neutron capture therapy using cyclotron-based epithermal neutron source and borofalan (10b) for recurrent or locally advanced head and neck cancer (jhn002): An open-label phase ii trial, Radiotherapy and Oncology 155 (2021) 182–187. doi:10.1016/j.radonc.2020.11.001.DOI
- [5]S. Shen, S. Wang, D. Zhou, et al., A clinician’s perspective on boron neutron capture therapy: promising advances, ongoing trials, and future outlook, International Journal of Radiation Biology 100 (8) (2024) 1126–1142. doi:10.1080/09553002.2024.2373746.DOI
- [6]K. Younous, M. El Kahfali, I. Bouadel, A. Biyi, R. Sebihi, Efficacy and safety of boron neutron capture therapy: a systematic review, International Journal of Radiation Biology 100 (12) (2024) 1611–1621. doi:10.1080/09553002.2024.2413583.DOI
- [7]T. Zhou, K. Igawa, T. Kasai, et al., The current status and novel advances of boron neutron capture therapy clinical trials, American Journal of Cancer Research 14 (2) (2024) 429–447. doi:10.62347/HBBE8688.pubmed.ncbi.nlm.nih.gov/38455422
- [8]R. F. Barth, G. Wu, M. d. G. H. Vicente, J. C. Grecula, N. Gupta, Boron neutron capture therapy of cancer: where do we stand now?, Cancer Communications 44 (8) (2024) 889–892. doi:10.1002/cac2.12581.DOI
- [9]L. Zhou, Y. Chen, M. Peng, et al., Clinical application of boron neutron capture therapy for cancer treatment: A systematic review, Precision Radiation Oncology 9 (4) (2025) 304–313. doi:10.1002/pro6.70031.DOI
- [10]T. Kutuk, E. Atak, M. Harrell, et al., Boron neutron capture therapy for high-grade cns tumors: Mechanisms, carriers, and clinical progress: A narrative review, International Journal of Molecular Sciences 27 (6) (2026) 2765. doi:10.3390/ijms27062765.DOI
- [11]D. Zheng, G. Han, O. D. M. Lemus, et al., Boron neutron capture therapy: A technology-driven renaissance, Cancers 18 (3) (2026) 498. doi:10.3390/cancers18030498.DOI
- [12]A. Monti Hughes, N. Hu, Optimizing boron neutron capture therapy (BNCT) to treat cancer: An updated review on the latest developments on boron compounds and strategies, Cancers 15 (16) (2023) 4091. doi:10.3390/cancers15164091.DOI
- [13]R. F. Barth, N. Gupta, S. Kawabata, Evaluation of sodium borocaptate (bsh) and boronophenylalanine (bpa) as boron delivery agents for neutron capture therapy (nct) of cancer: an update and a guide for the future clinical evaluation of new boron delivery agents for nct, Cancer Communications 44 (8) (2024) 893–909. doi:10.1002/cac2.12582.DOI
- [14]H. Xu, J. Liu, R. Li, et al., Novel promising boron agents for boron neutron capture therapy: Current status and outlook on the future, Coordination Chemistry Reviews 511 (2024) 215795. doi:10.1016/j.ccr.2024.215795.DOI
- [15]X. Li, P. He, Y. Wei, et al., Application and perspectives of nanomaterials in boron neutron capture therapy of tumors, Cancer Nanotechnology 16 (2025) 25. doi:10.1186/s12645-025-00324-3.DOI
- [16]R. M. Murilla, G. G. Edilo, M. L. M. Budlayan, E. S. Auxtero, Boron delivery agents in BNCT: A mini review of current developments and emerging trends, Nano TransMed 4 (2025) 100081. doi:10.1016/j.ntm.2025.100081.DOI
- [17]R. Yang, M. Shi, Y. Zhang, et al., Achievements, challenges, and perspectives in the design of boron delivery agents for neutron capture therapy, Coordination Chemistry Reviews 553 (2026) 217555. doi:10.1016/j.ccr.2025.217555.DOI
- [18]D. Tang, M. Shi, Z. Mao, et al., Boron-10 carriers and their applications in boron neutron capture therapy, Precision Radiation Oncology 10 (2) (2026) 205–241. doi:10.1002/pro6.70061.DOI
- [19]C. Selg, E. Hey-Hawkins, Boron neutron capture therapy at a crossroads: Translational gap and emerging delivery agents, Chemistry – A European Journal 32 (17) (2026) e03533. doi:10.1002/chem.202503533.DOI
- [20]S. Yadav, E. P. Onakpojeruo, C. Lansangan, R. Patil, Nanocarrier strategies for boron drug delivery in BNCT, Micromachines 17 (7) (2026) 846. doi:10.3390/mi17070846.DOI
- [21]H. Xie, C. Fan, F. Li, Nanoparticle boron carrier for boron neutron capture therapy: Research progress and perspectives in china, Nanomaterials 16 (14) (2026) 845. doi:10.3390/nano16140845.DOI
- [22]R. Zhou, Z. You, L. Liu, Boron neutron capture therapy in the context of tumor heterogeneity: progress, challenges, and future perspectives, Frontiers in Oncology 15 (2025) 1601013. doi:10.3389/fonc.2025.1601013.DOI
- [23]Q. Ma, P. Zhang, H. Zhang, et al., Core clinical procedures and efficacy-enhancing strategies in boron neutron capture therapy, Discover Oncology 17 (2026) 591. doi:10.1007/s12672-026-04774-y.DOI
- [24]C. Selg, W. A. G. Sauerwein, E. Hey-Hawkins, Quantification of 10b and 11b in biological samples in translational science and clinical applications for boron neutron capture therapy (BNCT), Medicinal Research Reviews 46 (5) (2026) 1362–1389. doi:10.1002/med.70060.DOI
- [25]T. Sato, S.-I. Masunaga, H. Kumada, N. Hamada, Microdosimetric modeling of biological effectiveness for boron neutron capture therapy considering intra- and intercellular heterogeneity in 10b distribution, Scientific Reports 8 (2018) 988. doi:10.1038/s41598-017-18871-0.DOI
- [26]T. Sato, S. Hashimoto, T. Inaniwa, K. Takada, H. Kumada, Implementation of simplified stochastic microdosimetric kinetic models into PHITS for application to radiation treatment planning, International Journal of Radiation Biology 97 (10) (2021) 1450–1460. doi:10.1080/09553002.2021.1956003.DOI
- [27]T. Furuta, T. Sato, Medical application of particle and heavy ion transport code system PHITS, Radiological Physics and Technology 14 (3) (2021) 215–225. doi:10.1007/s12194-021-00628-0.DOI
- [28]T. Shigehira, T. Watanabe, M. Suzuki, et al., Development of linear interpolation system for smk model parameters evaluated from cellular-scale simulation (LISMEC) and its application to BNCT dosimetry, Journal of Radiation Research 67 (2) (2026) 170–181. doi:10.1093/jrr/rrar075.DOI
- [29]R. Yamazaki, N. Hu, T. Takata, M. Nojiri, L. Zhao, H. Tanaka, Development of a microdosimetry-based method to derive cell survival rates for evaluating the biological effects of BNCT, Biomedical Physics & Engineering Express 12 (2) (2026) 025014. doi:10.1088/2057-1976/ae44a2.DOI
- [30]A. Selva, A. Bianchi, L. Bellan, E. Fagotti, A. Pisent, V. Conte, Microdosimetry for BNCT: First measurements at different site sizes, Radiation Measurements 178 (2024) 107298. doi:10.1016/j.radmeas.2024.107298.DOI
- [31]G. A. Santa Cruz, Microdosimetry in BNCT, Precision Radiation Oncology (2026). doi:10.1002/pro6.70063.DOI
- [32]B. Marcaccio, A. M. Portu, G. A. Santa Cruz, et al., Unraveling the role of boron microdistribution in BNCT dosimetry of glioblastoma multiforme: combined theoretical and experimental insights, Physics in Medicine & Biology 71 (8) (2026) 085004. doi:10.1088/1361-6560/ W.-B. Zhong, J. Chen, Y.-C. Teng, Y.-H. Liu, Introduction to the monte carlo dose engine COMPASS for BNCT, Scientific Reports 13 (2023) 11965. doi:10.1038/s41598-023-38648-y.
- [34]D. Pistone, S. Bortolussi, S. Fatemi, et al., A gate monte carlo study on ICRP110 phantoms for BNCT dosimetry evaluation, Applied Radiation and Isotopes 220 (2025) 111724. doi:10.1016/j.apradiso.2025.111724.DOI
- [35]Y. Zhu, Z. Lin, G. Wu, et al., Development of a scientific version tps workflow for BNCT dosimetry based on DICOM and monte carlo engine, Physical and Engineering Sciences in Medicine 49 (3) (2026) 1321–1331. doi:10.1007/s13246-026-01743-1.DOI
- [36]R. Kakino, N. Hu, A. Sasaki, et al., Integrating measurement and simulation for evaluation of patient-specific whole-body dosimetry in clinical BNCT, Medical Physics 53 (5) (2026) e70474. doi:10.1002/mp.70474.DOI
- [37]Y.-C. Teng, J. Chen, W.-B. Zhong, Y.-H. Liu, Correcting for the heterogeneous boron distribution in a tumor for BNCT dose calculation, Scientific Reports 13 (2023) 15741. doi:10.1038/s41598-023-42284-x.DOI
- [38]Y. Kobayashi, S. Nakamura, M. Takemori, et al., Comparison of dose distribution with and without reflecting heterogeneous boron distribution using 18F-BPA positron emission tomography in boron neutron capture therapy, Applied Radiation and Isotopes 219 (2025) 111720. doi:10.1016/j.apradiso.2025.111720.DOI
- [39]Y.-C. Teng, J. Chen, W.-B. Zhong, Q. Lin, Y.-H. Liu, Hyposegmentation of 18f-bpa pet-based boron distribution: a pragmatic approach for efficient BNCT treatment planning, EJNMMI Research 16 (2026) 18. doi:10.1186/s13550-025-01341-4.DOI
- [40]Y.-C. Teng, Y. Cao, X. Zhu, et al., Pet-guided modeling of mucosal boron heterogeneity improves dose-toxicity prediction in BNCT, International Journal of Radiation Oncology Biology Physics (2026). doi:10.1016/j.ijrobp.2026.05.056.DOI
- [41]R. Narita, Y. Sakurai, Estimation of internal-exposure contribution in radiation dose exposure for boron neutron capture therapy, Radiation Protection Dosimetry 200 (7) (2024) 623–628. doi:10.1093/rpd/ncae073.DOI
- [42]N. Hu, T. Nakamura, R. Kataura, et al., Implementation of a real-time neutron monitor system for use in routine quality assurance of an accelerator-based neutron system for clinical boron neutron capture therapy, Journal of Applied Clinical Medical Physics 26 (8) (2025) e70190. doi:10.1002/acm2.70190.DOI
- [43]S. Agostinelli, J. Allison, K. Amako, et al., Geant4—a simulation toolkit, Nuclear Instruments and Methods in Physics Research Section A 506 (2003) 250–303. doi:10.1016/S0168-9002(03)01368-8.DOI
- [44]J. Allison, K. Amako, J. Apostolakis, et al., Recent developments in Geant4, Nuclear Instruments and Methods in Physics Research Section A 835 (2016) 186–225. doi:10.1016/j.nima.2016.06.125.DOI
- [45]Y. Han, C. Geng, J. N.-D. Kondo, M. Li, J. Ramos-Méndez, S. Altieri, Y. Liu, X. Tang, Microdosimetric analysis for boron neutron capture therapy via monte carlo track structure simulation with modified lithium cross-sections, Radiation Physics and Chemistry 209 (2023) 110956. doi:10.1016/j.radphyschem.2023.110956.DOI
- [46]A. T. A. Wood, G. Chan, Simulation of stationary gaussian processes in [0, 1]ᵈ, Journal of Computational and Graphical Statistics 3 (4) (1994) 409–432. doi:10.1080/10618600.1994.10474655.
- [47]M. Frigo, S. G. Johnson, The design and implementation of FFTW3, Proceedings of the IEEE 93 (2) (2005) 216–231. doi:10.1109/JPROC.2004.840301.DOI