reté Nucléaire (IRSN), Fontenay-aux-Roses, France
Abstract
Working Group (WG) 6 'Computational Dosimetry' of the European Radiation Dosimetry Group promotes good practice in the application of computational methods for radiation dosimetry in radiation protection and the medical use of ionising radiation. Its cross-sectional activities within the association cover a large range of current topics in radiation dosimetry, including more fundamental studies of radiation effects in complex systems. In addition, WG 6 also performs scientific research and development as well as knowledge transfer activities, such as training courses. Monte Carlo techniques, including the use of anthropomorphic and other numerical phantoms based on voxelised geometrical models, play a strong part in the activities pursued in WG 6. However, other aspects and techniques, such as neutron spectra unfolding, have an important role as well. A number of intercomparison exercises have been carried out in the past to provide information on the accuracy with which computational methods are applied and whether best practice is being followed. Within the exercises that are still ongoing, the focus has changed towards assessing the uncertainty that can be achieved with these computational methods. Furthermore, the future strategy of WG 6 also includes an extension of the scope toward experimental benchmark activities and evaluation of cross-sections and algorithms, with the vision of establishing a gold standard for Monte Carlo methods used in medical and radiobiological applications.
Export citation and abstract BibTeX RIS
1. Introduction
The European Radiation Dosimetry Group (EURADOS) e.V. is an association of more than 70 European institutions and has about 600 individual scientists as associate members. The mission of EURADOS is to promote the scientific understanding and technical development of dosimetry in the fields of radiation protection, radiobiology and medical use of ionising radiation (e.g. radiation therapy and diagnostic radiology) by stimulating collaboration between European research institutions. Currently, EURADOS has eight working groups (WGs), which organise scientific meetings, training activities, intercomparisons and benchmark exercises for promoting quality assurance (Rühm et al 2015, 2018, 2020, Alves et al 2019).
WG 6 'Computational Dosimetry' has a cross-sectional role and promotes good practice in the application of computational methods for radiation dosimetry in radiation protection and the medical use of ionising radiation. As computational methods are widely used in radiation protection and other areas of radiation dosimetry, for example in the design of experiments and in the interpretation of results, WG 6 strongly engages in collaborations with the other WGs that are focused on various subject areas of practical dosimetric application (harmonisation of individual monitoring, environmental dosimetry, internal dosimetry, dosimetry in radiotherapy, retrospective dosimetry, high-energy radiation fields and dosimetry in medical imaging). These collaborations cover a large range of current topics in radiation dosimetry, including more fundamental studies of radiation effects in complex systems. In addition, WG 6 also performs scientific research and development as well as knowledge transfer activities, such as training courses.
Monte Carlo techniques, including the use of anthropomorphic and other numerical phantoms based on voxelised geometrical models, play an important part in the activities pursued in WG 6. However, other aspects and techniques, such as neutron spectra unfolding, have an important role as well. As a hybrid activity between knowledge transfer and research, a number of intercomparison exercises have been carried out in the past where participants were invited to solve predefined computational problems with some freedom to choose their methodology and approach (Tanner et al 2004, Gualdrini et al 2005, Price et al 2006, Siebert et al 2006, Gómez Ros et al 2007a, 2007b, 2008a, de Carlan et al 2008, Gualdrini et al 2008, Lopez et al 2010a, Broggio et al 2012, Barros et al 2014, Vrba et al 2014, 2015, Caccia et al 2017, Gómez-Ros et al 2018, Villagrasa et al 2019, Li et al 2020a); the participants' results could then be compared, in an effort to investigate common issue and difficulties, provide practical feedback and learn lessons. Participation in the exercises is open to everybody, and participants have been solicited by announcements on the EURADOS website (www.eurados.org) and in the regularly issued EURADOS newsletter as well as through conference papers and publications (Tanner et al 2004, Gualdrini et al 2005).
The main purpose of such exercises was to provide information on the accuracy with which computational methods are applied and whether best practice is being followed. However, within some of the exercises that are still ongoing, i.e. neutron spectra unfolding (Gómez-Ros et al 2018) and investigations of the cross-sections used as input parameters in track structure codes (Villagrasa et al 2019), the focus has changed towards assessing the uncertainties that are associated with these computational methods. Furthermore, the future strategy of WG 6 also includes an extension of the scope toward experimental benchmark activities and evaluations of the cross-sections and algorithms underlying the computations, with the vision of establishing a gold standard for Monte Carlo methods used in medical and radiobiological applications.
EURADOS WG 6 currently has 29 full members and 24 corresponding members (individuals who contribute to the WG work without full membership of the WG). Traditionally WG 6 has been subdivided into task groups (TGs) that cover a range of activities, but there is considerable interaction and working across the different TGs, along with other WGs.
2. Tasks within WG 6
2.1. Neutron spectra unfolding
Task 1 on neutron spectra unfolding deals with the computational tools related to neutron spectrometry, as a prerequisite for the dosimetric assessment of the neutron component of radiation fields. Neutron dosimetry is a particular challenge in radiation protection, as there is a pronounced energy dependence of the biological effects of neutrons and the neutron energies may be spread over more than ten orders of magnitude (ICRP 2007). As neutrons are indirectly ionising particles, they are generally detected via the recoil protons produced in nuclear reactions. To cover the large relevant energy range, sets of so-called Bonner spheres (BSs) are used, which feature a central sphere filled with 3He that is surrounded by spherical shells of polyethylene of different thickness and different inner and outer coatings of suitable metals. The neutron energy spectrum changes as a result of neutron interactions in the BS. The materials and sizes of the BSs are chosen such that through this moderation process different parts of the incident neutron energy spectrum are matched to the resonance of the production of tritium (3H) by neutron capture of 3He.
A neutron field measurement thus consists of a set of count rates from the different BSs, which then needs to be transformed into information on the neutron energy spectrum via spectrum unfolding techniques. A number of codes are used for this purpose, and one of the main activities of the TG has been to analyse the performance of those different tools (Barros et al 2014). Currently, the task is focused on conducting an intercomparison exercise on reconstructing neutron spectra from knowledge of the system response function and the values of counts obtained with a set of BSs. In addition, the TG is engaging in a joint activity with EURADOS WG 11 'High Energy Radiation Fields' to prepare a training course on neutron spectra unfolding, which is planned to take place in 2021.
Within the BS spectra unfolding exercise organised by TG 1, participants were invited to unfold BS data using a unfolding tool of their choice (Gómez-Ros et al 2018), having been provided with limited information about the nature of the neutron field and the calculated response matrices for idealised BSs. The BS data consisted of the count rates and measurement uncertainties of a set of 13 BSs, which were calculated by the organising team using Monte Carlo simulations of the experimental setups.
The following four scenarios for BS measurements of neutron fields were considered: (a) at two points near a medical linear accelerator (LINAC) in a treatment room with a maze; (b) in a simulated workplace field; (c) inside an irradiation room with a radionuclide source in an iron sphere; (d) environmental measurement at 100 m from a nuclear plant ('sky shine scenario').
For each case a reference solution was determined before the start of the exercise by members of the TG who were not involved in the simulation of the BS counts given to the participants, so as to ensure that the problems were solvable. At present, analysis of the solutions delivered by the 20 participants from 15 countries worldwide using 14 different unfolding techniques is close to completion and a paper on the results is being drafted.
2.2. Micro- and nanodosimetry
Task 2 on micro- and nanodosimetry has its major activities in investigating the uncertainty associated with track structure Monte Carlo simulations and fundamental issues in track structure calculations, such as conceptional questions to do with scoring the energy deposition pattern in charged particle tracks and the relevance of quantum effects in track simulations. Furthermore, the TG also supports the efforts of the international Geant4-DNA collaboration to include cross-sections for materials other than liquid water in their database. Major activities at present concern a code intercomparison exercise for estimating uncertainties in micro- and nanodosimetric simulations (Villagrasa et al 2019) and a code intercomparison exercise held jointly with WG 7 'Internal Dosimetry' on the dose enhancement around gold nanoparticles (Rabus et al 2020, Li et al 2020a, 2020b).
Within the two-step 'uncertainty' exercise, the goal of the first step is to use the discrepancies between results of different codes to estimate an uncertainty budget for the cross-sections for low-energy electron scattering in liquid water, which are the most likely cause of the discrepancies. The exercise was defined by a given simulation setup (i.e. a specified source and 'detector' geometry, electron source energy spectrum, etc) for which the microdosimetric specific-energy spectrum or the nanodosimetric cluster size distributions were to be calculated (Villagrasa et al 2019). The microdosimetric mean quantities calculated by the participants in this exercise generally agreed well (within a few percent), suggesting that all codes employed in this part of the exercise can be applied with radiobiological models that relate microdosimetric quantities with biological effects. In the second step of the exercise it was planned to implement a unique cross-sectional data set in different track structure codes, together with the estimated uncertainties. The task was then to transport those uncertainties through the simulation to obtain the uncertainty component of the simulation results that originates from the uncertainty of the cross-sections. Given that the nanodosimetric radiation quantities obtained in the first part of the exercise exhibited such large discrepancies that they would require the uncertainty to be assumed to be as high as 100%, this second part of the exercise is currently being redefined.
In the joint code intercomparison exercise with WG 7, participants were asked to simulate the emitted electron spectrum and the microscopic dose deposition around gold nanoparticles for x-ray irradiation in a simple, idealised geometry that consisted of a pencil beam of comparable diameter to the nanoparticle (Li et al 2020a). This artificial situation was meant to emphasise the differences in interaction cross-sections that could be expected, which is particularly important because the radiation interaction cross-sections of low-energy electrons in gold have not yet achieved the same level of evaluation as the cross-sections for interactions in water. In the first analysis, the ISO Guide to the expression of uncertainty in measurements was employed (ISO 1995) to interpret the reported results as independent measurements and to derive an uncertainty estimate for the dose enhancement by gold nanoparticles, where the probability distribution function had to be assumed to be log-normal to account for the large spread of the results. Further analysis revealed that the large differences between the reported results could be traced back to conceptional misunderstanding and improper implementations of the exercise (Rabus et al 2019, Li et al 2020b).
2.3. Individual monitoring
Task 3 is concerned with Monte Carlo simulations for individual monitoring in radiation protection. As computational methods play an important role in designing and evaluating the dosemeters and instruments used to assess the operational quantities, the TG is engaging in collaboration with WG 2 'Individual Monitoring' to provide a training course on the use of Monte Carlo techniques for this purpose. In addition, the TG has also been contributing to a cross-WG activity on producing a report that intends to aid and guide the radiation dosimetry community following the imminent change to the operational dose quantities proposed by the International Commission on Radiation Units and Measurements (ICRU) (ICRU 2020). One further recent major research activity conducted by this TG is related to retrospective dosimetry, performed in collaboration with WG 10 'Retrospective Dosimetry'. The aim of this work is to develop the means for relating the dose measured by a retrospective dosemeter (e.g. the glass and electrical components of a mobile phone) to the detriment to the individual, where the relationship is a function of location, orientation, exposure geometry and energy. This encompasses: the exploration of optimum dose quantities for use in scenarios encountered in nuclear emergency and accident situations (Eakins and Ainsbury 2018a, 2018b); the generation of a database of conversion coefficients from absorbed dose-related signals measured from mobile phone components to organ doses, for the most relevant exposure sources [e.g. (at present) 137Cs, 60Co, 131I and 192Ir radionuclides, as well as 100 keV x-rays] and for different irradiation geometries; and the development of conversion algorithms that incorporate the database and can easily be used in emergency scenarios. In addition, a new research activity is under development by this TG that relates to skin dosimetry, and in particular the generation of an improved skin model. This work aims to address the problem that the voxel size in the currently used numerical ICRP phantoms for the standard man and woman (ICRP 2009) are too big to capture details of the skin that become relevant for short-ranged charged particles: the skin model of ICRP publication 118 considers dermis and epidermis as unstructured layers of different thickness depending on the location on the body (ICRP 2012), which may not be sufficient in particular for stochastic effects or very inhomogeneous exposures, such as from 'hot' (radioactive) particles.
2.4. In vivo dosimetry
Task 4, on in vivo dosimetry, has been terminated recently after a long record of joint activities in collaboration with WG 7 'Internal Dosimetry'. Those activities encompassed a series of intercomparison exercises on dose assessment from in vivo measurements of radionuclide contamination in the lung, the knee or the skull (Gómez Ros et al 2007a, 2007b, Moraleda et al 2007, Gómez-Ros et al 2008a, Lopez et al 2010a, Broggio et al 2012, Vrba et al 2014, 2015, López et al 2019). In the most recent such exercise, on the determination of the activity of the nuclear fall-out product 241Am in the human skull, intercomparisons of Monte Carlo simulations were performed for three levels of complexity of the simulation problem (Vrba et al 2014, 2015). In the lowest complexity level, the response function of a given detector measuring a defined numerical skull phantom was to be simulated for a completely defined measurement geometry and given material data (Vrba et al 2014). The intermediate complexity problem encompassed simulation of a real detector where the relevant material parameters and dimensions were to be retrieved by the participants for a detector of their choice. The highest complexity level related to a real measurement geometry, including the detectors and their placement with respect to the skull phantom (Vrba et al 2015). All three intercomparison parts were addressed by 15 or 16 participants, mostly using the MCNPX code (Pelowitz 2005), but there were also results produced using other codes, namely EGS4 (Nelson et al 1985), Geant4 (Agostinelli et al 2003) and VNC (Hunt et al 2003).
2.5. LINAC modelling
Task 5 involves conducting an intercomparison exercise on the design and dosimetry assessment of a LINAC facility (Caccia et al 2017, 2020, Rühm et al 2020). In the first part of the exercise, participants were asked to determine the electron energy distribution and beam size impinging on the head of a LINAC operated at the French designated institute for ionising radiation metrology (LNHB), using information provided on the LINAC head geometry and material composition as well as results from measurements of the lateral and depth–dose profiles in a water phantom that were performed at LNHB. In the second part of the exercise, simulations were to be performed starting from the photon source given by the LINAC head and the electron beam to obtain lateral and depth–dose profiles within simplified patient models, realised by a water phantom with inhomogeneities mimicking bone and lung tissue, i.e. material denser or less dense than water. Here, four cases were considered: (a) lung in water; (b) bone in water; (c) bone and lung in water; and (d) two lungs in water. The densities of the lung and bone 'tissue' and the simulation geometry were provided as input for the simulations. Six participants provided solutions to the exercise, using four different Monte Carlo codes, namely Geant4 (Agostinelli et al 2003), EGSnrc (Kawrakow 2000), Tripoli 4 (Brun et al 2015) and MCNPX (Pelowitz 2005). Using a gamma index criterion as applied in clinical practice, the simulation results were again benchmarked with measurements performed at LNHB in real phantoms. The main conclusion from the exercise was that with respect to the first task to be solved, different electron beam parameters could equally well reproduce the measurements in a simple homogeneous water phantom, leading to significant discrepancies for the cases with heterogeneities (Caccia et al 2017, 2020). The simulation set-up used in the exercise has been adopted by the international Geant4 collaboration (Agostinelli et al 2003, Allison et al 2006) as an advanced example for self-training of new users of that code system.
2.6. Voxel phantoms
Task 6 is concerned with the implementation of voxel phantoms in Monte Carlo simulations, particularly for assessing radiation protection quantities (organ doses, effective dose). Significant contributions have also been made in the past to intercomparisons of in vivo dosimetry, in collaboration with Task 4 and WG 7 (Gómez Ros et al 2007a, 2007b, 2008b, Lopez et al 2010b). Another important activity has been conducting training courses on the implementation of voxel phantoms in Monte Carlo simulations (see section 3). At present, the TG is focused on the final analysis stage of a group of intercomparison exercises that tested participants' capabilities to properly implement the ICRP reference computational phantoms (ICRP 2009) in Monte Carlo simulations and calculate organ doses (and from these, equivalent doses and the radiation protection quantity effective dose). The exposure scenarios considered were designed such that cases of occupational, accidental and medical exposures were covered, as well as several radiation types (photons, neutrons, electrons).
The occupational exposure scenarios were irradiation by a 60Co photon point source or a monoenergetic 10 keV neutron point source, both with the source in front of the human body at 125 cm from the feet and 100 cm from the chest. The former scenario was addressed by 17 participants, while for the latter only eight solutions were received.
The accidental exposure scenarios were: ground contamination with an 241Am radionuclide distributed evenly within a disc of 2 m radius with the phantom standing in the centre, which attracted 13 participants; and exposure in a room where the air is contaminated with the β-emitting radionuclide 16N, for which six participants delivered solutions. In the latter case, part of the exposure would be from inhaled 16N radionuclides in the lung.
For medical scenarios, an idealised typical x-ray examination of the chest in PA geometry (irradiation from the back) and of the abdomen in AP geometry (irradiation from the front) was chosen. Here, the eight participants submitting solutions had to determine the location of the point source relative to the phantom co-ordinate system for the given target organ to be screened, and determine the organ absorbed dose conversion coefficients with respect to the dosimetric quantities air kerma (free in air) at the entrance skin surface of the body and kerma–area product. The latter two quantities would be measured in x-ray screening practices in the frame of quality assurance of the x-ray source.
A sixth exposure scenario was also proposed that considered the internal exposures occurring in occupational, accidental and medical situations. Here the task was to evaluate quantities relevant to dose assessment for artificial cases in which monoenergetic photons or electrons were emitted within specified source organs, and for cases in which the radiation was emitted by specific radionuclides. Twelve participants submitted solutions for this challenge.
For each of the six parts of the intercomparison, two members of the organising group independently performed calculations before the exercise was announced, to ensure that the problems were solvable and that master solutions could be established. The participants were advised to use the reference computational phantoms as described in ICRP Publication 110, and recommended either to apply the method proposed in ICRP Publication 116 (ICRP 2010) for determining the doses to red bone marrow and bone surface or else describe in detail any deviating bone dosimetry method if they chose to use an alternative.
2.7. High energies
Task 7 on high energies collaborates closely with WG 11 'High Energy Radiation Fields' and WG 9 'Radiation Dosimetry in Radiotherapy'. In the past, the modelling of radiation protection devices was an important activity (Rollet et al 2009, Wiegel et al 2009, Silari et al 2009). More recently, BS responses for high-energy neutrons have been modelled and benchmarked with measurements (Rühm et al 2014). The discrepancies due to the different models employed in the various high-energy codes to generate neutron-scattering cross-section data for the energy domain from MeV to GeV remain an ongoing challenge. Currently, a literature review is being conducted to compile a summary of the capabilities of the codes in use. The definitions of intercomparison exercises that will aim to assess the implementation of these models in the codes are also in preparation.
2.8. WG 6 strategy (formerly uncertainty and sensitivity assessment)
Originally established to address the overarching topic of uncertainties in computational dosimetry following the comprehensive Quality Assurance of Computational Tools for Dosimetry (QUADOS) exercise that had been conducted (Tanner et al 2004, Gualdrini et al 2005, Price et al 2006, Siebert et al 2006), task 8 is currently concerned with the steering and strategy of the WG. A major activity has been providing input to the Strategic Research Agenda of EURADOS (Rühm et al 2015, 2018).
2.9. Nuclear medicine
Task 9 engages with WG 12 'Dosimetry in Medical Imaging' and WG 7 'Internal Dosimetry', in a collaboration with the European Association of Nuclear Medicine (EANM), on dosimetric issues related to diagnostic and therapeutic nuclear medicine. These topics include the doses received by carers/comforters of patients receiving nuclear medicine treatment, or doses to members of the public from persons who have undergone a nuclear diagnostic procedure. Currently the research is focused on the development of a computational approach for determining the external dose rates received by third-party individuals from nuclear medicine patients, for a set of specific geometries. Major challenges encountered in this context are related to the biokinetics of the radionuclides in the human body.
2.10. Radiotherapy and radiation diagnostics
Task 10 on radiotherapy and radiation diagnostics was established in 2020 in response to an increased demand from WG 9 'Radiation Dosimetry in Radiotherapy' and WG 12 'Dosimetry in Medical Imaging' for support of their research projects by Monte Carlo simulations. Examples include: the assessment of surrounding healthy tissue doses in brachytherapy, where sources of encapsulated radionuclides emitting short-range photons or electron radiation are employed; and dosimetry issues with proton therapy, where unwanted irradiation of tissue outside the region targeted for treatment is a major concern, and may arise due to neutrons produced by nuclear collisions in the irradiated tissue, as well as in components of the irradiation facility used for shaping or moderating the proton beam. In both cases, data analyses of experimental measurements often require auxiliary information from radiation transport simulations. A further issue of high practical concern is the dose received by the foetus if the mother undergoes radiotherapy, for example before the pregnancy has been realised. Another major activity is a collaboration with WG 7 'Internal Dosimetry' and WG 12 'Dosimetry in Medical Imaging' on dosimetry issues of emerging radiopharmaceuticals (e.g. 68Ga, 177Lu).
3. Education and training activities
A common insight from the intercomparison exercises in computational dosimetry referred to in the previous section is that a major source of uncertainty of the simulation results can be attributed to the code users (Siebert et al 2006, Price et al 2006). In contrast to analogous comparisons in metrology, for example, where the technical protocol is agreed in advance by participants who typically already have a track record of experience in the measurements to be performed, the intercomparisons in computational dosimetry have a different character. For instance, while the exercises are prepared by experts in the field and methodology, many of the participants are less experienced code users, often early stage researchers. Moreover, such users may be easily misled by the apparent ease of use that many general purpose Monte Carlo codes offer nowadays, or be tempted to believe that, for example, small standard deviations on their modelled results are a sign of low uncertainties overall (rather than just being a metric of statistical precision), or that dividing results by the number of primary particles simulated would be an appropriate normalisation. Apart from annoying aspects that complicate data analysis, such as results not being reported using the provided templates and/or with binning structures different from those requested, the intercomparisons also often reveal naivety and conceptual misunderstandings of the simulation problem itself, as well as a lack of awareness that simulation results need to be checked for self-consistency and plausibility, such as having the obvious right order of magnitude or correct sign.
In most of the intercomparison exercises mentioned in the previous section, the task was solved by two members of the respective TG to ensure that the problem was solvable and to obtain a master solution as the reference. In the evaluation of these intercomparison exercises, there has therefore usually been a feedback loop where participants were informed that their solution deviated more than expected from the master solution, and they were requested to provide further information on their simulations and to check their results. In quite a few cases, the participants then delivered revised results for which, in general, the deviation from the master solution would decrease, while still sometimes being significant (Vrba et al 2014, 2015, Li et al 2020b).
In some areas of computational dosimetry, and particularly in micro- and nanodosimetry, the rapid development of codes is an additional complication, as different codes may use different cross-section data or models for their implementation that can lead to significantly different results (Lazarakis et al 2012, Villagrasa et al 2019). In general, however, the outcomes of these exercises are snapshots of the state of the art in the field, not necessarily reflecting the capabilities of the codes but rather the performance of code users. The group of participants may not even be a representative sample, as interest and willingness to take part in such voluntary exercises already perhaps demonstrates a level of awareness that may not be common within the community of code users.
Therefore, like the other WGs within EURADOS, WG 6 also contributes to EURADOS' mission of promoting harmonisation and good practice in radiation dosimetry by organising and contributing to education and training events (Alves et al 2019), such as dedicated training courses, winter schools and conference events on topics related to computational dosimetry. A list of such training and dissemination events over the past decade is given in table 1.
Table 1. Training courses and workshops delivered by EURADOS WG 6 over the past decade.
| Activity | Place and dates |
|---|---|
| Organisation of Training School on the development of voxel human geometries for Monte Carlo codes | Fonteny aux roses, France, 11–13 October 2011 |
| Organisation of 6th Winter School 'Status and Future Perspectives of Computational Micro- and Nanodosimetry' | Barcelona, Spain, 6 February 2013 |
| Co-organisation of International Workshop on 'Challenges in Micro- and Nanodosimetry for Ion Beam Cancer Therapy (MiND-IBCT)' | Wiener Neustadt, Austria, 7–9 May 2014 |
| Organisation of Training School on the development of voxel human geometries for Monte Carlo codes | Neuherberg, Germany, 13–15 May 2014 |
| Contribution to 10th EURADOS Winter School 'Internal dosimetry for radiation protection and medicine' | Karlsruhe, Germany, 2 March 2017 |
| Contributions to 11th EURADOS Winter School 'Application of physical and computational phantoms in dose assessment' | Lisbon, Portugal, 8 February 2018 |
| Organisation of EURADOS Training Course on the application of Monte Carlo methods for dosimetry of ionizing radiation | Karlsruhe, Germany, 12–16 March 2018 |
| Organisation of Satellite workshop on micro- and nanodosimetry to the 3rd International Conference on Dosimetry and Applications | Lisbon, Portugal, 26 May 2019 |
| Contributions to the 7th International Workshop on Computational Phantoms | Neuherberg, Germany, 22–24 July 2019 |
With the progress in computational power that has taken place over the last few decades, the further development of Monte Carlo techniques and the emerging use of machine learning in dosimetry, an increasing number of different codes are now being used. Whilst this opens up new opportunities, they are, however, not for free, as quality assurance of the simulation results requires thorough training and experience along with efforts to establish a Monte Carlo 'gold standard'. It is important that users become fully aware of the need for uncertainty assessments of their simulation results, and that it is necessary to study the influence of variance of input data (e.g. geometry, cross-sections, physics, etc) on them, although this may involve considerable effort for standard Monte Carlo codes. The organisation of intercomparison exercises and training courses therefore will remain an important task of WG 6 in the future. Furthermore, high-level intercomparisons will also be needed to ensure that important issues addressed by computational methods are treated using more than one code: experimental benchmarks are often missing in emerging areas, such as in high-energy fields or track structure simulations, so the codes may only be validated against each other. Overall, comparisons of results obtained using different approaches (e.g. condensed history versus full track structure, Monte Carlo versus analytical, or machine learning versus Monte Carlo) will be needed to highlight differences and limitations, depending on the specific configuration of the simulation.
4. Outlook
Computational methods play an essential role in the area of radiation dosimetry. Some of the dosimetric concepts in radiation protection, such as the so-called protection quantities that are defined in anthropomorphic computational phantoms that cannot be constructed physically, can only be determined by computational methods. With ICRU currently revising the operational dose quantities to be also defined using anthropomorphic phantoms, computational dosimetry will play a key role in adjusting the radiation protection system to these changes, for example by evaluating the new conversion coefficients.
The rapid development of codes, and their flexibility and capability to provide solutions with detailed geometric descriptions, has made Monte Carlo simulation the tool of choice in radiation protection dosimetry as well as in medical applications. In radiation protection, the possibility of monitoring the dose received by an individual in real time and as a personalised assessment might also soon be within reach. With the move towards personalised medicine, there is also the increasing importance of patient-specific dosimetry, for which computational dosimetry is an essential tool, where taking into account individual variability makes the computational task more complex.
The recent developments of numerical phantoms (mesh or NURBS phantoms) that no longer rely on a voxelised geometry, and also potentially permit adaptation of phantoms to an individual's personal anatomy, in principle open the door towards individualised computational dosimetry that might even take into account motion. Practical realisation of these is currently hampered by incompatibility with the approaches used for geometric description in most Monte Carlo codes, or by the computational effort involved. These obstacles can only in part be overcome by increased computational power or the development of new variance reduction techniques. This opens the way for methodologies based on machine learning, such as convolutional neural networks, that would facilitate the use of personalised phantoms with realistic anatomies. They may also (within certain limits) replace current Monte Carlo simulations or be used for their acceleration to achieve truly real-time dose assessments.
Validation of such novel approaches will require intercomparisons with established Monte Carlo techniques, and assessments of the associated uncertainties. Developing a validation scheme for machine learning-based approaches by Monte Carlo techniques, and benchmarking of flexible numerical phantoms, are likely to be two main extensions of the scope of WG 6 activities in the future. This will have to be accompanied by endeavours on the ongoing challenge of uncertainty assessment in computational dosimetry. Addressing this challenge involves critical assessment of the basic cross-section data that are used in the codes, particularly for novel medical applications involving new types of radionuclides or nanomaterials (Schuemann et al 2020), micro- and nanodosimetry, as well as novel workplace radiation environments. It will also involve developing means for sensitivity analysis of the simulation results on variation of the code input quantities, as well as the establishment of appropriate reference experiments as benchmarks.
Acknowledgments
The following persons are acknowledged as current active members of EURADOS WG 6: Aliaksandr Bancer, Manuel Bardiès, Janine Becker, Roberto Bedogni, Abdel-Hai Benali, Marie-Claude Bordage, David Broggio, Kevin Capello, Jean-Christophe David, Liset de la Fuente Rosales, Marko Fulop, Elisabeth Lindbo Hansen, Bernd Heide, Jan Jansen, Alexis Jinaphanh, Pavel Kundrat, Yi-Kang Lee, Weibo Li, Wladimir Mares, Vladimir Markovic, Montserrat Moraleda, Heidi Nettelbeck, Dragoslav Nikezic, Alessio Parisi, Yann Perrot, Michaël Petit, Marcin Pietrzak, Marcel Reginatto, Uwe Reichelt, Neil Roberts, Vaclav Stepan, Katarzyna Tymińska, Olivier VanHoey, Pedro Vaz, Roberto Versaci, Francois Vianna-Legros, Fernanda Villegas Navarro.