Introduction
Typical methods for constructing radiotherapy treatment shells include high-temperature thermoplastics (e.g., cellulose acetate butyrate) moulded onto a plaster positive of the patient’s face and low-temperature soft-drape thermoplastic devices that require direct moulding to the patient’s face. Reference Devereux, Grundy and Littman1, Reference Gilbeau, Octave-Prignot, Loncol, Ranard and Gregoire2–Reference Schulte, Fargo, Meinass, Slater and Slater4 Both of these methods involve procedures that can be distressing for patients and the resulting shells do not always fit perfectly. Sharp et al. state that the mask making process, coupled with the ongoing stress of cancer diagnosis can lead to an acute anxiety or state of panic.Reference Sharp, Lewin, Johansson, Payne, Gerhardsson and Rutqvist5 In this paper the feasibility of using three-dimensional (3D) printers to directly construct either a ‘positive’ face or a treatment (negative) shell from computed tomography (CT) or MRI data is investigated. Techniques to construct 3D geometric models which can be interpreted by 3D printers are developed and the suitability of a range of 3D printer materials is determined by testing their dosimetric properties in comparison to existing immobilisation shell materials. Creating masks directly from the CT or MRI data will alleviate the need for moulding of the mask to the patients’ face. Helen Bulbeck of the BrainsTrust recently conducted a survey on the impact of having a radiotherapy shell made and 50% of the respondents found the experience horrific. Additionally the Macmillan website describes the plaster process as potentially uncomfortable.Reference McCloskey and Moore6
Previously, the authors have used techniques to utilise CT and MRI datasets for medicalReference Tam, Laycock, Bell and Chojnowski7 and cultural heritageReference Laycock, Bell, Mortimore, Greco, Corps and Finkle8 applications. A recent review articleReference Esses, Berman, Bloom and Sosna9 highlighted some of the possible clinical applications of using physical 3D models derived from CT data created by rapid prototyping. There has been work on the use of 3D surface scanning techniques in combination with 3D printing to generate radiotherapy treatment shells,Reference Sanghera, Amis and McGurk10, Reference McKernan, Bydder, Deans, Nixon and Joseph11 however, the authors are not aware of other published research using Digital Communications in Medicine (DICOM) data derived from either CT or MRI scans to generate such shells. Since CT and MRI technology is readily available it seems logical to investigate the potential of using this type of data to aid in the construction of anatomically accurate radiotherapy treatment shells. In light of preliminary findingsReference Hulse, Isherwood, Scrase, Laycock and Bell12 and an earlier studyReference Devereux, Grundy and Littman1 methods are presented for working with both CT and MRI datasets. The suitability of a variety of 3D printer materials for use during radiotherapy treatment is also investigated and discussed.
Materials and methods
The head and neck of a whole body CT scan was reconstructed at a slice thickness of 3·75 mm and the DICOM dataset was then processed using Tomomask software (www.tomomask.com) with the following stages:
(1) Masking – removal of any extraneous items/features from the images which would cause problems during the automatic segmentation process (e.g., support materials or image distortions caused by metal fillings/teeth, etc).
(2) Segmentation – identification of the flesh/air boundary by straightforward thresholding of the pixel intensity values.
(3) Binarisation – conversion of the images to black and white (white areas will subsequently be printed) where the interior of the head and neck (from the flesh boundary) is solid white and the exterior, ‘air’, is black. This step was performed using an automatic flood fill operation.
(4) Conversion to a hollowed out ‘positive’ head (to save on printer material) – a reduced size version of the head and neck was created by 3D erosion using a spherical structuring element of 6 mm radius. A ‘positive’ head was then created by setting the pixels of the eroded head and neck to black.
(5) Creation of a ‘negative’ shell – first, a positive is created around which the negative shell can be wrapped. The head and neck are dilated slightly (1 px) to give a small tolerance for fitting. As the shell must be pulled over the head, there can be no parts of the shell which wrap behind the head. This is achieved by applying a 1×∞ pixel linear dilation directed towards the plane at the rear of the head (i.e., each white pixel is converted into a line of white pixels running towards the image lower edge). A negative shell was then created by 3D dilation of the white pixels (using a spherical structuring element) before setting the original white pixels to black.
(6) Conversion to a Surface-Tessellation-Language (STL) file format before 3D printing – the white areas of the images, forming the positive and negative shells (created in steps 4 and 5 above), were then converted to STL files using the Tomomask ‘Export to STL’ feature based upon a marching cubes algorithm.
The two STL files produced (i.e., both the ‘positive head and neck’ and the ‘negative’ shell) were printed using a Z-Corps 650 printer (3D Systems, http://www.3dsystems.com/3d-printers/professional/projet-660pro). The full size positive head and neck was used to test the production of a soft-drape shell. A 50% of full size ‘positive’ head and neck was also printed to test the production of a traditional high temperature beam direction shell (BDS).
A head and neck MRI was also performed on a volunteer (co-author: M. D. Tam), using a 1·5T GE magnet (Figure 1a). A 3D volume scan was acquired axially using a 3D fast spoiled gradient echo sequence, TR 12·2 ms, TE 5·0 ms, TI 450 ms, with a FOV of 44×44 cm, and a matrix size of 256×224 with 1 NEX (number of excitations). A total of 182 images with a slice thickness of 2·0 mm and 0·55 mm slice gap were obtained with a scan time of <12 minutes. The DICOM dataset from the MRI scan was then processed using Osirix software (www.osirix-viewer.com) using threshold segmentation, where the user controls the threshold levels for the segmentation, to convert the image into a binary dataset of black and white pixels. These were then further processed using Tomomask as above to produce a hollowed-out ‘positive’ of the head and neck (Figure 1b) and a ‘negative’ STL shell to fit the ‘positive’ head and neck of the volunteer (Figure 1c). The STL files were then printed at 50% size for a 3D print of the ‘positive’ hollowed-out head and neck model (Figure 1d) and at 50% size for a 3D print of the ‘negative’ head shell, both of which were printed out using the Z-Corps 450 3D printer. In addition, using just the face part of the ‘negative’ shell’s STL file, a ‘negative’ shell was printed at full scale by Objet Printer Solutions UK Ltd using an Eden250 system with Objet’s ‘VeroWhitePlus’ material (Figure 1e).
Constructing the masks from magnetic resonance imaging (MRI) data. (a) Image of a slice of the MRI data which has been segmented using Osirix, (b) three-dimensional (3D) computer model of the head derived from the segmented MRI dataset, (c) Edited ‘negative’ shell (holes were subsequently formed in the shell to demonstrate the possibility of further reducing material cost), (d) 3D print of the model of the head and (e) 3D print of an example ‘negative’ shell.
Dosimetric properties of 3D printing materials
In our experiment we measured the change in dose produced by three materials suitable for 3D printing. These were compared with a 2·4 mm thick ‘unstretched’ thermoplastic ‘soft drape’ radiotherapy shell produced by Orfit (http://www.orfit.com/en/radiation-oncology-patient-immobilization-and-positioning/), which represents a standard material.
Measurements were made in a water equivalent material irradiated with a 10×10 cm2 beam of energy 6 MV. The dose was measured at two points:
(i) 5 cm deep in water using a 0·6 cm3 Farmer chamber
(ii) 1 mm deep in water using a Roos chamber
In both cases the radiation source to chamber distance was 100 cm. The dose was measured with and without the test materials placed on top of the water equivalent material. For each setup, three repeat readings were acquired with a standard deviation of <0·1%. For point (i), the change in dose produced by the material (∆D5 cm) assesses its effect on the dose to the tumour. For point (ii), the change in the dose (∆D1 mm) represents its effect on the dose to the skin.
Results
The dosimetric properties of the 3D printing materials are summarised in Figure 2. Figure 2a shows that all the materials produce a small decrease (up to 1%) in the dose to the tumour due to attenuation of the beam. The 2 mm materials produce lower attenuation than the Orfit mask. Figure 2b shows that all of the materials produce an increase in skin dose between 54 and 80%. This can be minimised by using the thinner materials. The skin dose for the 2 mm materials is similar or lower than the Orfit mask.
(a) The change in dose at 5 cm depth, ∆D5 cm for the Orfit thermoplastic mask in comparison to three different three-dimensional (3D) printer materials. (b) The change in dose at 1 mm depth for the Orfit thermoplastic mask in comparison to three different 3D printer materials.
Discussion
The initial stage in the process for constructing the mask is to first perform a CT scan of the patient in the intended treatment position without an immobilisation shell. This represents a change in the standard practice although can be achieved with reference to standard set up instructions as would be used in conventional mask production, provided that the patient is comfortable and stable during the CT scan. Once the CT or MRI scan has been obtained the head must be segmented from the background. There is a clear edge between the dataset and surrounding air which aids the segmentation step but further evaluation to ensure this segmentation is accurate for skin is required. Furthermore, motion during the MRI sequence can blur the edges. This was seen particularly at the orbits, which can lead to artefacts causing errors up to 1 mm that can be detrimental in some cases.
When constructing radiotherapy shells on a 3D printer the material used is fundamental as an ideal material should not affect the radiation dose delivered to the patient. Ideally the 3D printer material chosen should perform at least as well as the materials used in conventional methods. We found that 2 mm samples of Visijet clear (Projet Material, http://www.3dsystems.com/materials/visijet-sl-clear) and EOS PA 3200 (EOS Material, http://www.eos.info/material-p) produced changes in beam attenuation (<0·6%) and increases in skin dose (up to 55%) that were comparable to an unstretched Orfit head mask. During the experiments solid sections of the printing materials were used. In practice holes can be created within the masks to help to reduce skin dose. This can be further minimised by using as thin a material as possible, while still maintaining the rigidity of the shell. Compared to standard techniques it is not expected that the dosimetric changes introduced by these new materials will have a significant clinical impact on the treatment.
Initial pre-clinical trials of such shells will need to decide on the type of 3D printer and the material used. These decisions will depend on such factors as the ‘build size’ of the 3D printer, the size of the shell required, patient acceptability and cost. It is likely that ‘negative’ shells will be produced directly as printing ‘positive’ heads first seems unnecessary and it is significantly more expensive. Current soft drape masks cost approximately £30–£65 depending on the size while printing just one shell using a commercial printer currently costs £80 (although mould room time is avoided). However, the printing costs are significantly reduced if multiple masks or other objects such as bolus can be printed within the same build chamber.Reference Zou, Fisher and Swann13 In the last 5 years we have seen the 3D printing and rapid prototyping industries really push the boundaries in technology and materials. Mid-range devices have become available at significantly cheaper prices bringing down the costs of printing and the authors believe that as the technology develops and the applications in medicine become more numerous that it will become viable for a large hospital or perhaps a cluster of smaller hospitals to have access to their own high end 3D printer for medical applications. Some 3D printing materials are already certified for use within surgical applications. In terms of time print jobs could be initiated at the end of the working day enabling the completed models to be extracted with limited post-processing the next morning.
It is of paramount importance that any radiotherapy shell employed using 3D printing technology must be at least as efficient (and hopefully even more efficient) than a standard soft drape or BDS shell produced in the standard way in terms of immobilising the patient. Studies have suggested that both soft drape and BDS radiotherapy shells produced in a standard way do allow a certain amount of movement and hence the recommendation to employ some method of checking patient position before commencing each treatment.Reference Devereux, Grundy and Littman1, Reference Gilbeau, Octave-Prignot, Loncol, Ranard and Gregoire2–Reference Schulte, Fargo, Meinass, Slater and Slater4, Reference Sanghera, Amis and McGurk10, Reference Kim, Park, Ki and Ye14–Reference Kang, Lovelock, Yorke, Kriminski, Lee and Amols16 Rotondo et al.Reference Rotondo, Sultanem, Lavoie, Skelly and Raymond15 present an approach to calculate errors between immobilisation techniques by determining displacement errors between co-registered datasets of patients undergoing different immobilisation techniques. Recently Kang et al.Reference Kang, Lovelock, Yorke, Kriminski, Lee and Amols16 have investigated accuracy in positioning and immobilisation for head and neck cancer treatment and determine translational intrafractional errors by six DoF registration of <2 mm, although they did find that there was substantial motion during treatment in some cases. Future work to that discussed in this paper will be to conduct immobilisation studies of the proposed 3D printed radiotherapy masks along similar lines to those used by Sanghera et al.Reference Sanghera, Amis and McGurk10 Provided these are satisfactory an application will be made for ethical approval to carry out a prospective clinical comparison with radiotherapy shells produced in the standard way with patient-related but also treatment-related endpoints.
Conclusions
Techniques have been shown which create 3D printed immobilisation shells produced from patient data using CT or MRI. The majority of the possible candidate 3D printing materials tested result in very similar attenuation of a therapeutic radiotherapy beam as the Orfit soft-drape masks currently in use in many UK radiotherapy centres. The thickness of the shell should obviously be kept to a minimum to prevent large increases in skin dose. In terms of imaging, it is clear that standard image acquisitions can be used. The prototype CT used was created using thicker slices than typically used for radiotherapy planning. The MRI sequence was created with 2 mm slices covering from the vertex of the skull to the upper shoulders which took <12 minutes. Image acquisition would in practice need to be in the intended treatment position according to the site to be irradiated.
In summary, 3D printed shells can be produced from CT and MRI datasets which will potentially avoid the moulding of shells directly to the patients’ face. New shells could also be produced using the volumetric datasets acquired as part of an image-guided and adaptive radiotherapy programme. Associated cost is likely to be driven down by the emerging technology that is rapid prototyping.
Acknowledgments
The authors wish to thank Mr Chris Heal from EMCO UK, Mr Rob Thompson from Objet Solutions UK, Mr Phil Kilburn 3TRPO Ltd and Mr Garth Stephenson from ESOS UK for helpful advice and providing 12×12 cm square sheets of their respective 3D print materials. The technical help of Mr D. Andaver, Mr J Johnson and Mr G Cox are gratefully acknowledged.
Financial Support
The authors wish to thank Big C research fund for partially supporting this work (Grant Number: 12-07R).
Conflicts of Interest
None.
