Eliminating computed tomography imaging artifacts through 3D printed radiotherapy head supports

Abstract Purpose The geometry of an immobilization device such as a headrest can cause undesired computed tomography (CT) artifacts that may affect both volume definition and dosimetry in radiotherapy of the brain. The purpose of this work was to reduce CT artifacts caused by a standard hard plastic hollow radiotherapy headrest. This was to be achieved through design and prototyping of a custom‐made head support. Methods A series of CT scans were acquired of both a water phantom and an anthropomorphic head phantom which were resting on custom‐made three‐dimensional (3D) printed supports. All custom‐made supports were made of polylactic acid (PLA) plastic filament and printed by fused deposition modeling (FDM) 3D printing technology. Initial designs were studied with a water phantom using a simplified support with straight and curved shapes both at the edges and as infill patterns. Imaging of a 3D printed clinical prototype was then compared to our standard headrest using an anthropomorphic head phantom. Results The presence of dark streaks inside both phantoms was seen on the CT images for headrests involving supports with straight shapes at the edges or as infill patterns. Such artifacts were ascribed to the exponential edge‐gradient effect (EEGE). No such artifact was observed when the support was designed with a combination of curved edges and infill patterns. Conclusion When developing immobilization accessories for use in CT scanners, more attention could be paid to artifact attenuating design elements. This work illustrates the usefulness of 3D printing in prototyping radiotherapy accessories and solving concrete clinical problems.


| INTRODUCTION
While planning radiotherapy for brain cancer treatment in our clinic, computed tomography (CT) images exhibit dark streak artifacts that can interfere with detection and delineation of structures of interest.
Inspection of these images reveals that most streaks appear to radiate from the edge of the plastic headrest used during the CT scan.
The signature of this artifact is similar to a specific case of partial volume artifact also referred as the exponential edge-gradient effect (EEGE). 1 The latter is partly explained by a slight underestimation of the linear attenuation coefficient when the boundary between two media is parallel to the X-ray beam of the CT scanner. If this border extends as a straight line, then the underestimation will be summed and back-projected, resulting in dark streaks in the CT images. 2 The main objective of this work was to develop a new headrest design to prevent EEGE artifacts. The first part of the study was carried out on a water phantom to reproduce the clinical conditions causing the artifact. A set of simplified phantom supports were threedimensional (3D) printed using different shapes in order to influence the presence of artifacts. This allowed us to test the theoretical concepts underlying the headrest design and assess the impact of the infill pattern required for 3D printing. The second part of the study was carried out with an anthropomorphic head phantom to compare the new headrest design with the standard clinical one.

| MATERIALS AND METHODS
Two designs of a simplified support were made starting from a cube of dimension (40 × 40 × 40) mm 3 . The first support attempted to reproduce the properties of a clinical headrest model using straight lateral sides. A V-shaped opening is added at the top to support a cylindrical phantom. The second support reproduces the dimensions of the previous model except for its four lateral S-shaped sides. A
Both the simplified supports and the redesigned headrest were printed using a fused deposition modeling (FDM) Ultimaker S5 printer (Ultimaker, Utrecht, Netherlands) and polylactic acid (PLA) filament.
The Cura software (Ultimaker) was used to convert 3D object models from stereolithography (.stl) files to specific instructions for the 3D printer. The printing parameters were set with a nozzle diameter of 0.4 mm and a layer thickness of 0.15 mm. Each simplified support model is printed with an infill density of 20% using either a "triangles" or "gyroid" pattern, as shown in Fig. 1(b). The redesigned headrest model is printed with an infill density of 10% using the gyroid pattern.
A list of the printing parameters is provided in Table 1.

| RESULTS
The CT images obtained with the simplified supports and the water phantom are presented in Fig. 4. Figures 4(a)

| DISCUSSION
The origin of the artifact is studied on the water phantom with a set of simplified supports to determine which design features have to be included in our custom-made headrest. Computed tomography images on the water phantom shown in Fig. 4 allows an EEGE-type artifact to be attributed to the presence of straight structures in the geometry of the support, both on the lateral sides and infill pattern.
One way to avoid this type of artifact is to replace the side walls of the support with curved shapes. Three-dimensional printing also brings a new set of constraints, notably with the use of an infill. The choice of an infill pattern with rounded shapes, such as the gyroid, is indicated to avoid introducing new EEGE-type artifacts. Figure 5 reflects the clinical application of these principles by demonstrating that a headrest with rounded shapes yields artifact-free images. The difference maps highlight the presence or absence of such artifacts, but also reflect small unrelated registration errors at the edges of bone structures.

| CONCLUSION
A 3D printed headrest has been designed avoiding straight walls and infill patterns, which were found to be causing undesired streak arti- scanners, more attention could be paid to artifact attenuating design elements. Future works will include developing a web-based platform to share our 3D designs to the radiation oncology community.

ACKNOWLEDGMENTS
The authors thank Sylvain Bélanger for CAD support, Karim Boudam and Martin Lebeau for clinical support, and Daniel Markel for proofreading.

CONFLI CTS OF INTEREST
No conflicts of interest. Curved edges F I G . 5. Computed tomography images of the head phantom on two headrests models. The top row shows the data obtained with the clinical headrest and the bottom row with the custom-made one. The window center is 50 HU and the window width is 220 HU. Computed tomography volumes are registered on an acquisition without headrest, also used to generate the difference maps with a greyscale bar expressed in HU. The arrows point to the artifacts introduced by the presence of the headrest, visible on the axial and coronal views.