Additive manufacturing in radiation oncology: a review of clinical practice, emerging trends and research opportunities

The additive manufacturing (AM) process plays an important role in enabling cross-disciplinary research in engineering and personalised medicine. Commercially available clinical tools currently utilised in radiotherapy are typically based on traditional manufacturing processes, often leading to non-conformal geometries, time-consuming manufacturing process and high costs. An emerging application explores the design and development of patient-specific clinical tools using AM to optimise treatment outcomes among cancer patients receiving radiation therapy. In this review, we: • highlight the key advantages of AM in radiotherapy where rapid prototyping allows for patient-specific manufacture • explore common clinical workflows involving radiotherapy tools such as bolus, compensators, anthropomorphic phantoms, immobilisers, and brachytherapy moulds; and • investigate how current AM processes are exploited by researchers to achieve patient tissue-like imaging and dose attenuations. Finally, significant AM research opportunities in this space are highlighted for their future advancements in radiotherapy for diagnostic and clinical research applications.


Introduction
The additive manufacturing (AM) process has produced countless research advancements not only in the engineering research and development space but also in medical applications. Well known applications for AM in the medical space involves the fabrication of patient-specific surgical models [1], bone implants [2][3][4][5][6], dental moulds and implants [7][8][9][10], scaffolds for tissue engineering applications [11][12][13][14][15][16][17], and the fabrication of inexpensive and personalised radiotherapy tools utilised by clinicians in radiation therapy (RT) to improve patient treatment outcomes. During RT, it is common to use beam modifiers during radiotherapy to accommodate irregular patient surfaces, for instance, wedge filters [18], which are used to modify the isodose distribution to a certain amount to achieve dose homogeneity, as well as electron beam cut-outs [19], which are utilised to shape beams according to a specified cut-out geometry. Other forms of beam modifiers include boluses [20] and compensators [21], which are personalised devices and can achieve better isodose distributions for more irregular surfaces. Another commonly used clinical tool are immobilisers [22], which are used by radiotherapists as fixation devices for patients during treatment to minimise body movements and avoid the irradiation of healthy tissues. More complex radiotherapy clinical tools include brachytherapy moulds for localised cancer treatment using radioactive seeds, in particular for prostate cancer procedures [23] and the use of in-vitro dosimetry models such as anthropomorphic radiotherapy phantoms to validate the overall deliverability of treatment plans involving imaging and dosimetry [24,25].
Without proper clinical interventions by medical physicists, radiotherapists and oncologists, these clinical devices fall short as they do not exactly replicate patients in terms of body dimensions and tissue inhomogeneities. For instance, commercially available radiotherapy phantoms to validate patient treatment plans were observed to produce a lack of dose information due to the unaccounted presence of patientspecific lesions. Furthermore, these phantoms are fabricated through moulding and casting, which are known to be very timely and costly due to the use of specialised materials [26,27]. Available AM processes such as stereolithography (SLA), fused deposition modelling (FDM), polymer material jetting (PJT/MJT), selective deposition lamination (SDL), and laser object manufacturing (LOM) can be utilised to highlight such clinical phantoms with patient-specific dimensions at relatively low costs and rapid manufacturing lead-times.
In this paper, we discuss in detail the current clinical applications of AM in radiotherapy, focusing on commonly used radiotherapy tools such as bolus, anthropomorphic phantoms, immobilisers, compensators, and brachytherapy moulds. Key findings and future insights are discussed throughout this study involving past and current methodologies and how AM is integrated in the current day to day clinical practice in radiotherapy.

Radiation therapy overview
RT plays a significant role in the management and treatment of cancer. Significant research is directed towards the non-invasive use of external beam radiotherapy (EBRT) and their advanced and localised cancer treatment capabilities including stereotactic body ablative radiotherapy (SABR) and proton therapy. For invasive procedures, internal radiotherapy (IRT) utilises invasive surgical procedures and seed implementation (brachytherapy) where radioactive seeds are directly implanted adjacent to cancerous cells.
The common EBRT pathway is illustrated in figure 1 and consists of the following steps: (1) patient immobilization and imaging with modalities ranging from computed tomography (CT), magnetic resonance imaging (MRI), positron emission tomography (PET), single-photon emission computed tomography (SPECT), ultrasound or through simple x-ray radiographs; (2), (3) followed by tumour segmentation where attending clinicians identify tumour volume parameters (i.e. shape and position) and margins considering critical structures prior to image validation procedures; (4) computerised treatment planning systems (TPS) are then utilised by clinicians and physicists to virtually simulate the radiotherapy process where dose beam directions and intensities are optimised for enhanced patient treatment. The optimisation can either be done by the operator who choses beam number, shape, directions and dose contribution (forward planning) or done using a computer algorithm, a process called inverse treatment planning. Once treatment plans are finalised and approved, the actual patient irradiation process is implemented where (5) the intended full dose is divided into small doses (fractions) and delivered to the patient in specified intervals (i.e. usually daily) to allow healthy tissues to recover between treatments.

Treatment planning process
The treatment planning process utilises computer algorithms, using TPS, to determine the best treatment parameters for the disease of the individual. These parameters include target volume(s), dose-limiting structures, dose prescription, dose fractionation, dose distribution, positioning of the patient, treatment machine settings, and adjuvant therapies. The system also produces reference images and other data which aid set-up and position verification of the patient in each treatment fraction. The final output of this process is then followed precisely by attending clinicians over several weeks [28].

Radiotherapy quality assurance
Given the high and potentially dangerous doses of radiation given to patients, quality assurance (QA) is an essential part of radiotherapy practice [29,30]. In several clinical trials, the importance of high-quality delivery has been demonstrated [31] with technical QA typically assessing both dose delivered and the geometric precision of the delivery. QA processes of radiotherapy delivery can be commonly divided into two parts, the validation of a process in a generic phantom or the verification of dose delivered in a delivery sequence for an individual patient. The problem with both is that they typically employ a standard representation of a human patient, as illustrated in figure 5.
For the end-to-end testing of a new procedure, anthropomorphic phantoms are commonly used. They mimic the radiation properties of a standard patient. While paediatric size phantoms are available, the largest size of phantom mimics a small normal-weight healthy male. None of the commercial phantoms represents an obese patient or includes pathology such as tumours, emphysematous lungs or metallic hip replacements. The same applies to patient-specific QA, where the radiotherapy plan developed for an individual patient is delivered onto a detector array that usually is a Perspex cylinder of human dimensions (see figure 2). If the measurement that can easily be taken in the phantom compares well with the treatment plan, it is assumed that the dose distribution calculated for the patient using the same algorithms is also correct.
AM allows us to address both of these shortcomings. It can create realistic phantoms (or modify existing ones) that can mimic real patients with co-morbidities, different shape lesions and a variety of different body shapes. Similarly, AI can create phantoms that can represent the complexities of real humans adequately providing for more meaningful QA for individual patients.

Considerations for AM design
AM provides designers with potential to extend their capabilities into highly accurate patient-specific models [1]. There are technological and material specific constraints that apply when designing for AM. To help identify which of these constraints apply to your product design, the appropriate AM technology needs to be selected for the intended use. Computed aided drawing (CAD) is an essential component of the design process in any AM product. The complexity and quality of the product design is dependent on the skill of the CAD designer. The constraints are defined by the build angle tolerance of the printer and the orientation which impacts the mechanical properties of the product. Layer thickness varies with the different AM technologies. Print orientation will affect the duration of the build and in some cases the quality of the product. All these constraints will influence the final product and its manufacturability.

Advances in AM for radiation oncology
A patent from the Leland Stanford Junior University briefly discussed the familiarity of their inverse planning method for radiosurgery with the very first three-dimensional AM techniques, SLA, where ultra-violet beams solidify liquid resins in such a way that a particular shape is obtained [32]. AM in Radiotherapy was soon conceptualised by Ju et al [33], illustrating its feasibility in manufacturing low-cost proton range compensators which were found to have comparable dosimetric and density results with conventional compensators manufactured through computerised milling machines (CMM). Since then, interests in AM have extended towards the use of personalised bolus devices [34,35], which are now part of the day-to-day clinical work for some radiotherapy centres. Active research is also currently being implemented for the AM of patient-specific radiotherapy phantoms [36], immobilisation devices [37], and brachytherapy moulds [38].

Bolus
Clinically used bolus devices are moulded thermoplastics sheets with tissue-equivalent properties to modulate the dose received by the patient's skin. Compared to standard wedge filters which are placed close to the radiation source (treatment head), boluses are placed directly on the patient's skin. Common tissue-equivalent materials for bolus manufacture include superflab, aquaplast, and superflex (see figures 4(a) and (c) (top)) and are categorised into pliable or rigid forms where: pliable boluses conform with patient skin surface at given thicknesses, and rigid boluses are usually implemented for smaller regions not requiring compliance with patient's skin. However, inherent limitations of standard bolus devices include insufficient contact with patient's skin (i.e. nose, ear, and scalp), leading to air gaps and a reduction in both the maximum and surface dose [39], further leading to inadequate treatment of superficial lesions.
Recent developments in AM-bolus devices include the use of the FDM printing process [40][41][42], which is the extrusion process of thermoplastics layer-by-layer [43]. This process allows the fabrication of geometries conforming to irregular surfaces (i.e. nose, chest, etc) with controlled thickness values to provide better dose coverage compared to traditional flat bolus sheets (see figures 3(a) and (b)).
Common filament materials for AM-bolus include acrylonitrile butadiene styrene (ABS) [34,44] and polylactic-acid (PLA) [45,46]. Due to non-toxicity and biocompatibility with comparable treatment planning system (TPS) dose profiles, PLA is currently preferred by clinicians for clinical bolus use [47]. The use of flexible material such as thermoplastic polyurethane (TPU) is also being investigated for their improved compliance and reduced air gaps for patient comfortability [48,49]. A recent release of flexible (soft) PLA may be able to highlight both patient skin conformity and tissue-dose equivalence [50].

Compensators
Compensator devices are fabricated using the moulding and casting process followed by milling, utilising solid water [51], lead [52], tungsten [53], or woods metal (cerrobend) [54] materials for megavoltage x-ray radiotherapy. Similar to wedge filters, these devices are attached close to the primary beam to attenuate beam dose with insignificant effects towards dose scattering within the patient [55]. Significant factors to be considered in manufacturing these devices include the field size, beam energy, and depth of the point of interest as they determine the required compensator thickness for treatment. These factors require personalised moulds, which are often expensive due to the fabrication process and materials involved. With the AM process, the fabrication of compensator devices not only reduces costs and manufacturing time but also enable the development of more complex geometries for beam attenuation [56,57] (see figure 4(d)). The early literature of 3D printed compensators includes compensator blocks [58] and proton-range compensators [33,59] (see figure 4(e)).

Anthropomorphic phantoms
Commercially available anthropomorphic phantoms are utilised for quality assurance purposes (see figures 5(a)-(d)) [60][61][62][63], for instance, the commissioning of new treatment procedures/machines or the validation of patient treatment plans using dose measurement tools such as thermoluminiscent dosimeters (TLD), dosimetry films, and ionisation chambers. Unfortunately, these phantoms do not fully represent patient-specific anatomies and pathological features, leading to unaccounted dose errors. More recently, review studies regarding AM of radiotherapy phantoms include the AM of imaging and dosimetry phantoms [64] and the varying imaging modalities used for AM-phantoms such as CT, MRI, PET, SPECT, and ultrasound [65]. Both studies have shown the feasibility of utilising inexpensive printing machines in particular for the FDM process to fabricate printable patientspecific geometries from patient imaging data.
Similar to the AM-bolus devices, common printing filaments used for AM-phantoms include standard materials from PLA and ABS, to specialty filaments from TPU, high-impact polystyrene (HIPS) and polymethyl methacrylate (PMMA). The main objective of AM-phantoms is to mimic patientspecific human tissue in terms of (a) imaging attenuations, for instance Hounsfield units (HUs) in CT and (b) dosimetric attenuations considering Compton effects in radiotherapy where electron density and atomic composition of the associated printing materials are both key factors for tissue mimicry within the radiotherapy setting (see figures 5(e)-(h)) [66][67][68][69].
Workflow for AM-phantoms initiates from (1) utilising patient CT, (2) the segmentation process of region of interests (i.e. lung tissue, bone, soft tissues), (3) 3D-model generation of segmented regions, post-processing, and printing parameter modifications and finally, (4) the FDM printing process of sliced models, assembly of printed models and the final implementation for imaging or dosimetry work (see figure 6). Other printing technologies utilised for AM-phantom fabrication also include PJT/MJT which allows multi-material and deformable material printing [66,70,71] and paper-based 3D printing utilising LOM and SDL processes [72,73]. The following selected methodologies further highlight innovative AM research associated with the fabrication of anthropomorphic phantom models: • Nowak et al [74] developed an open-source software to accommodate the rapid conversion of contours from DICOM CT data to printable models and which can be extended towards other applications such as bolus devices.
• Leary et al [75] developed a method to convert CT DICOM data to fabricate bi-modal head phantom slice using polymer material jetting technology. • Leary et al [76] demonstrated the feasibility of modulating phantom HUs using one printing material by utilising engineered square voids and its effects on CT imaging, known as the partial volume effect (PVE).   [63], and patient-specific phantoms manufactured through 3D-printing: (e) patient-specific thorax phantom with 3D-printed bone and lung regions (silicone thorax body) and CT comparison of lesions between patient and phantom data, (f) an image of a standard head phantom slice and its 3D-printed version using FDM printing, (g) a 3D-printed thyroid phantom, and (h) a head phantom slice illustrating internal nasal-oral cavities and Larynx. (e) Reproduced with permission from [66].  [80] have demonstrated the feasibility of a 3D printed lung phantom with printed lung lesions for stereotactic lung study using a motion platform device to simulate realistic lung movements (QUASAR, Modus Medical Devices, Canada). • Jahnke et al [72,73] have demonstrated the feasibility of fabricating realistic head phantom models through (1) a paper-based 3D printing method and (2) a customised LOM printer where it utilises a toner from the laser printer to act as an adhesive to stack all printed sheets. This application shows promising results however, as this application is still new, CT tissue-equivalence (HU) and dosimetry are yet to be quantifiable further research must be conducted regarding reproducibility.

Immobilisers
When previously discussed constraints (section 2.3) are combined into a clinical environment, another level of complexity is realised. The main clinical implications relate to the level of patient comfort, the rigidity of the product to provide the correct level of immobilisation and the suitability of the product design for individual patients [20,22] and the response of material to irradiation in terms of degradation and attenuation. Immobilising tools in radiotherapy are utilised by clinicians to accurately position and rigidly immobilise patients prior to treatment. Common immobilisation devices include beaded bags [81,82], polyurethane foam castings [83][84][85][86][87][88][89][90][91][92], orthopaedic plastics [93][94][95][96], head masks/holders with bite blocks [97] or without [98] using thermoplastics [99] (see table 1). Current limitations of standard immobilising techniques were observed to cause mental and physical distress towards patients ranging from anxiety to physical discomfort and claustrophobia [100][101][102][103].
Utilising AM, fabrication of immobilising devices highlights personalised and customisable geometries achievable through the AM modelling and printing process [31,37,104,105]. The fabrication process of these devices follows a similar pathway for all 3D printed radiotherapy devices initialising from CT data, to model processing and generating a printable 3D model and finally, printing [106] (see figure 7). The other pathway to develop printable models utilises surface laser scanners for generating reference models for constructing immobilising shells [107,108].
Furthermore, AM-immobilisers in the literature utilise a range of printing techniques including FDM, SLA, SLS, and PJT/MJT with extensive work regarding materials focusing on mechanical and imaging properties [108,109], with both PLA and ABS a common printing material. Other than the general head and neck 3D printed immobilisation devices [106,110] (see figures 7 and 8, respectively), segments suited for immobilisation also include abdominal area [111], the mouth [112] and breasts [113].

Brachytherapy moulds
Brachytherapy (a form of radiation therapy using an internal radiation source) can provide highly conformal radiation dose distribution to a targeted lesion. Brachytherapy mould is often used for nonmelanoma skin lesions that usually involve irregular surface, requiring a customised mould to fit along the surface (see figures 4(b) and (c)). Available data confirm that brachytherapy is an efficient and well-tolerated treatment that offers excellent cosmesis and low toxicity for skin cancer patients [114,115]. Conventional methods to construct a customised mould include using specialised polymers, acrylic resin, wax (similar to dental wax) or a thermoplastic material or similar [116,117]. Commercial flap applicators, which can be prefabricated with the desired geometry, are also being clinically used. Such commercial solutions include the Freiburg flap (Elekta), the H A M (Mick Radio-Nuclear Instruments), and the Catheter Flap set (Varian Medical Systems). These applicators can be cut to any size to fit the target area, however, this may result in an air-gap when not perfectly fitting, in particular for irregular and large skin lesions.
AM can play an important role here to offer better wellfitting personalised AM-moulds than the aforementioned conventional approaches. An increasing number of studies have demonstrated the application of additive manufacturing in skin brachytherapy domain, and its feasibility studies show promising results [118,119]. There is still more work to be done to ensure skin moulds can accurately model dosimetric calculation for radiotherapy. Current literature in AM-radiotherapy (2019-2020) are summarised in table 2. Capabilities of AM technologies regarding HU-equivalence are also summarised in table 3.

Opportunities and future direction of AM in radiation oncology
Radiotherapy is leaning towards bespoke models to accommodate the advancing diagnostic and treatment technologies     * not specified, ** FDM printer not specified, *** MJT material not specified, **** SLA material not specified. such as the use of high precision stereotactic radiosurgery and proton therapy in allowing the full potential of personalised radiotherapy. Recent work by the Association of University Radiologists Radiology Research Alliance (AUR-RRA) task force on 3-dimensional printing has published two separate papers investigating the logistics of clinical translation using the 3D printing technology in radiotherapy [142] and a more comprehensive insight towards the clinical applications of 3D printing in radiotherapy [143].
The following sections focus on significant research opportunities at the interface between radiotherapy and AM.

Generative design for medical additive manufacturing
Additive manufacture enables the design of high-complexity bespoke structures that are eminently compatible with the requirements of medical radiotherapy applications, including the bolus devices, compensators, phantoms, immobilisers and brachytherapy moulds reviewed in this work. Despite the opportunities for enhanced medical outcomes enabled by these technologies, the ability of a human designer to manually accommodate the complexity inherent to such design is limited. Consequently, AM-medical applications are potentially subject to high cost due to manual design and error checking effort, as well as regulatory challenges associated with confirming the suitability of this manually generated design.
4.1.1. Introduction to generative design systems. In response to these challenges, generative design (GD) systems are emerging within the AM community to enable the low-cost design of high-complexity systems, such as patient-specific medical systems. In broad terms, GD systems refer to 'the rules for generating form, rather than the forms themselves' [144]. In Creative Evolutionary Systems, GD tools enable design decision making and reporting to occur with the speed and repeatability of the available computational resources, rather than requiring manual decision making, thereby enabling design outcomes that are practically unachievable by a human designer.
A schematic representation of a GD system is provided in figure 9, including phases of problem, where the technical and oncological design requirements are formally specified; embodiment design, consisting of the generation and evaluation of high-level solutions to the problem definition; and detail design, where the specific solution is formally defined to enable manufacture [145]. In GD design, a generative design system algorithmically replaces one or more of these identified design phases.
In order to provide a useful design for additive manufacturing (DFAM) outcomes, the generative design system must replace one or both of the embodiment or detail design phases. For medical oncology applications, it is likely that the AM and oncological subject matter experts would agree on a preferred embodiment based on their subjectmatter expertise, but that the detail design phase be applied with a GD system such that patient-specific design outcomes can be algorithmically generated. GD outcomes are particularly important for high-value AM product that is customised to the specific requirements of a particular oncological scenario. This methodology has been applied in the design and manufacture of patient-specific radiotherapy phantoms based on observations associated with the partial volume effect (PVE); which allows local values of the phantom HU to be tuned by geometrically manipulating the local phantom geometry [76]. This method thereby allows the rapid manufacture of patient-specific phantoms based entirely on algorithmic manipulation of (see figure 10) the patient CT data, resulting in a locally defined density field, that is then converted to an AM data input file for manufacture. This methodology enables low-cost deployment of high-value oncological products according to the patient-specific requirements, as well as providing certification documentation as required for medical manufacture.

Water-tight and multi-material 3D printing
Material extrusion printers (FDM) are gaining greater attention due to new developments in printable materials and novel methodologies with reduced manufacturing costs and time with comparable printing resolutions to high-end printers such as PJT/MJT and SLA. For example, fabricating watertight devices requiring high-resolution printing as close to 25micron layer height, may require the use of the SLA technology, which uses a UV-curing process to polymerise liquid resins for microchannel fabrication [146] or the fabrication of a head phantom with fillable compartments using tissue-like liquids for magnetic resonance imaging applications [147] (see figures 11(a) and (b)). However, low-cost FDM printers may also be utilised for such applications as users can modify a huge range of printing parameters to achieve water-tight parts with comparable results to polymer jetting and SLA systems with high-resolution printing [148,149]. Relatively low-cost FDM printers with high-resolution and multi-material printing capabilities are tabulated in table 4.
A very recent research paper published by Skylar-Scott et al [150], have demonstrated a novel fabrication technique of multi-material FDM printers (multi-material multi-nozzle 3D printing, MM3D) which enables rapid material extrusion using nozzle numbers as high as 128 (16×8 nozzles) with capabilities in printing deformable materials (see figure 11(c)). This significant advancement in the FDM technology not only raises the possibility of ultra-rapid 3D printing but also has importance in terms of heterogeneous tissue printing where varying tissue-like print filaments can be printed simultaneously for the patient-specific fabrication of radiotherapy phantoms and tissue-engineering applications. Unfortunately, new technologies as such will still need some validation in terms of material compatibilities, part manufacturability, and their significance in the clinical workflow.
Moving on to commercially available high-end multimaterial printer such as the Stratasys J750 (Stratasys, US).
The J750 printer uses the PJT/MJT process which utilises the UV-curing process to polymerise two or more photopolymer resins extruded in tiny droplets. Available materials for this printer include Agilus30, ABS, RGD720, Tango, Vero, VeroClear and VeroFlex materials. Varying material ratio combinations between different materials can produce multicoloured parts with certain degrees of material flexibility (shore hardness).
Recent PJT/MJT applications to radiotherapy involve a mobile thorax phantom for 4DCT [71] and a mouse phantom for dosimetry [151]. Another printing methodology enabled by the J750 printer apart from multi-material printing involves the voxel printing technology which ignores the use of 3D model inputs for printing (i.e. STL, STP, 3MF, etc) and directly use RGB or BMP image stacks as inputs. This process, however, requires image pre-processing from image resizing and masking to image dithering [152]. The voxel method not only is a faster printing alternative as it bypasses the 3D model generation process but also provides better representation and material diffusion of heterogeneous structures from medical DICOM images for surgical guide model fabrication [153] (see figure 11(d)). With new proprietary tissue-like materials released by Stratasys [154] including: GelMatrix™ support resin for easy removal in printed narrow blood vessels, TissueMatrix™ resin for heart tissue fabrication and BoneMatrix™ resin for bone, voxel printing may be able to extend towards the fabrication of heterogeneous patient-specific phantoms directly from patient DICOM images. However, printed phantoms using this technology will still need various imaging and dosimetric studies to validate their manufacturability and tissue-like radiotherapy equivalence. Figure 10. GD system applied for algorithmic following schematic in figure 9 to generate a patient-specific radiotherapy imaging phantom using material jetting technology (MJT). Reproduced with permission from [76]. Copyright © 2020 by ASME.

Modular anthropomorphic AM-phantoms
The modularity of commercially available anthropomorphic phantoms is limited due to the manufacturing process of moulding and casting. For instance, horizontally sliced anthropomorphic phantoms such as the CIRS™ ATOM phantoms and the Alderson RANDO phantoms (ART) [25] only allow the insertion of TLDs and ionisation chambers with limited spaces for dosimetry films. Modularity in the phantom design highlights significant opportunities in 3D printing phantom slices with customised TLD and ionisation chamber insertion holes as well as placements for dosimetry films. Most AM-phantoms have demonstrated the feasibility of fabricating heterogeneous phantoms utilising different printing materials and technologies, with thorax phantoms consisting of patient-specific lesions [66,71].
The reproducibility and modularity of AM-phantoms are key requirements for clinical use. However, the development  of modular phantoms allowing reproducible insertions of patient-specific segments (i.e. lesions, lung, soft tissues, bone), for varying patient treatment cases, is yet to be developed. A possibility to resolve this limitation is to utilise current GD systems in parallel with a range of AM technologies to design and fabricate inhomogeneous building blocks such as lesions, lung and soft tissues, and bone, much like LEGO ® , to assemble a realistic phantom depending on a specific case for treatment planning.

4D Phantoms
Precision radiotherapy relies on accurate conformal delivery of radiation, and one of the major challenges is the fact that organs move, squash and stretch in the abdomen regularly during breathing and due to organ filling and tumour response to treatment. The precision of modern radiotherapy techniques is limited by such anatomy changes, resulting in different spatial components of tumours and surrounding organs 'seeing' different radiation doses than that planned. Limitations as such will introduce discrepancies between planned and delivered radiation doses. In this context, dynamically moving phantoms can add invaluable inputs on radiotherapy quality assurance throughout the process, imaging, planning, image verification and treatment [155]. Additive manufacturing can play a critical role here to construct more reliable, realistic, and cost-effective phantoms than limited commercial solutions, such as QUASAR™ (Modus Medical) CIRS dynamics (Varian Medical System) that simulate simplified anatomy motion.
An exciting application for AM 4D phantoms is to add a deformable feature in dynamic phantoms to better simulate deformably moving anatomy using AM technology. Until now, there is no solution in the commercial as well as research domains to emulate real patient-like anatomy change. AM has great potential to construct deformable structures that can emulate anatomy deformation. Such a phantom can be an invaluable tool to validate the mathematical modelling of anatomical deformation, where there is a lack of experimental data [71,156].

Radiomic phantom models
Radiomics is based on the premise that important features of medical images are not necessarily perceivable by a human observer [157,158]. These features can occur in several dimensions and pertain to shape of structures, texture, noise patterns and the relationship between neighbouring voxels of an image. Importantly radiomics features have shown to be related to tumour biology [159] and possibly clinical outcomes [160].
Radiomics features are usually extracted using computer analysis. They can be grouped and assessed using machine learning to link selected subsets of features to clinically relevant outcomes [161]. However, there is considerable variation in the quality of radiomics analysis between different institutions and validation of outcomes by independent observers is often difficult [159,162].
It is important to develop test objects that mimic radiomics features in a predictable and reproducible fashion. Additive manufacture with its ability to modify object features in a systematic fashion appears to be a perfect tool to develop such objects [163]. One challenge is the required fine detail of resolution that is required to represent texture features of biological tissues. However, many groups are currently working on solutions to these issues, and it can be expected that the wide availability of such quality assurance and standardisation tools for radiomics will help to make the field more robust.

Open-source 3D model database for AM-radiotherapy devices
To further increase and promote the significance of AM in Radiotherapy, the growing clinical data on these 3D printed devices require the need for proper documentation and the development of open-source forums and databases including: generated 3D models from clinical data, their associated printing parameters, their imaging and dosimetry outputs. There may be an opportunity for developing such a global initiative for clinicians, especially for medical physicists and engineers, to validate printing methodologies prior to clinical use. The National Institutes of Health (NIH, US Department of Health and Human Services) have developed a database, known as NIH 3D Print Exchange consisting of 3D models related to bioscience and medicine where users can share their 3D models extracted from clinical patient CT data [164]. However, models used in this database are not intended for clinical use and are focused only in the visualisation and guidance for surgical procedures.
Another great initiative from the US, the 3D Printing Special Interest Group (SIG) formed by the Radiological Society of North America (RSNA), have established a 3D printing registry containing valuable data associated with clinical 3D printing and assistance with clinical decision making regarding the use of the AM. Current efforts have attracted major 3D printing companies such as Formlabs, HP, Materialise, and Stratasys to provide substantial support for further database development [165].

Summary and outlook
3D-printing is revolutionising the way clinicians approach the diagnostic and treatment stage of patients in radiotherapy. At present, the AM-radiotherapy field is in an exciting phase of exploration and innovation, evident by the increasing amount of research towards quality assurance of 3D printable materials, their limitations, and more importantly, their contributions to enhance the delivery of clinical patient treatments.
The following summary can be gained from this review: • Over the years, the use of 3D printing in radiotherapy has sparked debates regarding implementation of AM fabricated phantoms for clinical use. The perspective that '3D printing technology will eventually eliminate the need for purchasing commercial phantoms for clinical medical physics QA procedure' [166] is becoming more achievable as AM-radiotherapy phantoms advance towards faster, high-resolution and multi-material printing workflows with controllable deformability of materials. This also applies to other applications such as bolus, immobilizers, compensators, and brachytherapy moulds. • Generative design is an emerging methodology for medical product design that enables commercial AM outcomes for scenarios that would be otherwise economically intractable, in particular, the manufacture of highvalue patient-specific product at low cost and accommodating the requirements for certification documentation. The implementation of such GD systems provides a significant opportunity for enhanced patient outcomes; however, such systems should be implemented with reference to standard methods of medical system design such that verification and validation of the intended GD system are robustly implemented [167]. • Research directions for AM-radiotherapy applications must lean towards the development of new tissueequivalent materials and structures through the FDM process to maintain low costs and print controllability.
Other techniques include paper-based printing (LOM and SDL) to further simulate realistic CT imaging of tissues, although are still expensive and are yet to be validated for dosimetry work are used for training and optimisation for CT-guided procedures. Another alternative for improved printing of tissue-like components include bioprinting. However, similar to LOM and SDL, the associated printing and material costs are still very high compared to standard FDM printers. In addition, limitations observed from bioprinting in terms of parameterisation and vascularisation are yet to be solved and remain to be the biggest challenges in the bioprinting research field [168]. The FDM printing process remains to be the most commonly used printing method for AM-radiotherapy due to lower printing costs with relatively high-resolution and multi-material printing capabilities. • For professionals (i.e. clinicians, physicists, and medical engineers) intending to use 3D printing for radiotherapy clinical use, extensive imaging and dosimetric assessments must be performed on printing material for their tissue-equivalence, considering inter-and intra-variabilities of outcomes. It is also recommended that proper documentation of clinically used 3D printed radiotherapy devices be implemented as it will play a significant role in providing guidance in future clinical AM-radiotherapy work, in particular for clinical regulations.