Quality assurance for a six degrees‐of‐freedom table using a 3D printed phantom

Abstract Purpose To establish a streamlined end‐to‐end test of a 6 degrees‐of‐freedom (6DoF) robotic table using a 3D printed phantom for periodic quality assurance. Methods A 3D printed phantom was fabricated with translational and rotational offsets and an imbedded central ball‐bearing (BB). The phantom underwent each step of the radiation therapy process: CT simulation in a straight orientation, plan generation using the treatment planning software, setup to offset marks at the linac, registration and corrected 6DoF table adjustments via hidden target test, delivery of a Winston‐Lutz test to the BB, and verification of table positioning via field and laser lights. The registration values, maximum total displacement of the combined Winston‐Lutz fields, and a pass or fail criterion of the laser and field lights were recorded. The quality assurance process for each of the three linacs were performed for the first 30 days. Results Within a 95% confidence interval, the overall uncertainty values for both translation and rotation were below 1.0 mm and 0.5° for each linac respectively. When combining the registration values and other uncertainties for all three linacs, the average deviations were within 2.0 mm and 1.0° of the designed translation and rotation offsets of the 3D print respectively. For all three linacs, the maximum total deviation for the Winston‐Lutz test did not exceed 1.0 mm. Laser and light field verification was within tolerance every day for all three linacs given the latest guidance documentation for table repositioning. Conclusion The 3D printer is capable of accurately fabricating a quality assurance phantom for 6DoF positioning verification. The end‐to‐end workflow allows for a more efficient test of the 6DoF mechanics while including other important tests needed for routine quality assurance.


| INTRODUCTION
The robotic patient positioning table is a vital component in external beam radiation therapy treatments. These tables can mechanically drive a patient to a desired treatment position with the aid of external skin marks or fiducials. Furthermore, most modern linear accelerators (linacs) are equipped with two-or three-dimensional image guidance in order to correct for and minimize interfractional setup uncertainties. Image-guided radiation therapy (IGRT) has improved accuracy for daily treatments and have allowed clinicians to decrease planning target volumes in order to spare normal tissues. 1 However, with linac-based treatments demanding greater accuracy, quality assurance (QA) tolerances for image-guidance and table positioning must be in congruence with more precise treatments. The American Association in Physicists in Medicine (AAPM) task group report 142 (TG-142) regarding quality assurance on medical accelerators specifies these tolerances with the type of treatment being delivered, especially with more complex modalities like intensity modulated radiation therapy (IMRT), stereotactic radiotherapy (SRT) and stereotactic body radiation therapy (SBRT). 2 Conventional treatment tables are designed with four degreesof-freedom (4DoF) in order to adjust positioning in the patient's vertical, lateral, and longitudinal directions, as well as an additional yaw rotation. A relatively new technology that is being outfitted on linacs are robotic six degrees-of-freedom (6DoF) tables that allow for mechanical adjustments of pitch and roll rotations in addition to the standard 4DoF adjustments. Early work in 6DoF corrections was investigated using BrainLAB's ExacTrac 6D system (BrainLAB AG, Feldkirchen, DE). BrainLAB utilizes a stereoscopic x-ray system with 2D-to-3D registration system to reference digitally reconstructed radiographs (DRR). 6DoF registration showed a superior submillimeter localization accuracy against 3DoF registration using a head phantom. 3 Rotational corrections with the Robotic Tilt Module on the Exactrac system showed an overall accuracy of 0.31 AE 0.77 mm with a quadrature summation of positional accuracy and isocentricity uncertainty. 4  Other prostate studies observed rotational adjustments greater than 2°and commented on the importance to correct for larger deviations. 10,11 Two studies categorized 6DoF accuracy in terms generalized disease sites. Guckenberger et al compared 6DoF setup accuracies in terms of nonfixated immobilization (body) and fixated immobilization (cranial or head and neck). 12 Schmidhalter et al classified 6DoF accuracies by cranial and extracranial treatments. 13 In both studies, it was observed that extracranial, or body-type, treatments required a larger translational and rotational correction.
While clinical implementation of 6DoF tables yield more accurate image-guided results, routine QA for 6DoF tables has been limited.
Schmidhalter et al has demonstrated reproducible 6DoF table performance using a combination of graph paper, inclinometers, and imaging methods. 14 These tests have demonstrated a process in which routine QA for 6DoF tables can be established. While 6DoF commissioning and QA have been characterized by other studies, it is to the best of our knowledge that a streamlined procedure has yet to be developed.
One such technology that is capable of establishing an efficient workflow is 3D printing. This "relatively new, rapidly expanding" technology has advanced personalized medicine by developing customized prosthetics, models, and medical devices. 15 Tack et al provided a systematic literature review regarding 3D printing publications in medicine. 16 They observed a rapid rise in publications after January 2011, with a majority of the publications originating from the surgical domain in medicine. This can be exemplified by such work on 3D printed frames for laser interstitial thermotherapy, 17 creating 3D anatomical models from magnetic resonance imaging, 18 and accurately printing organs with heterogeneous tissues. 19 Moreover, the emergence of 3D printing technology in various arenas in medicine has garnered similar interest in radiation oncology. A significant amount of effort into 3D printing has been utilized for patient-specific devices, which include: bolus for electron treatments, [20][21][22] compensators for photon treatments, 23,24 and patient immobilization. 25 3D printing in brachytherapy applicators has also been investigated. Dosimetric evaluations of an FDAapproved material 26 and various material infill densities 27 has shown promising results. 3D printed phantoms have also been successfully developed for various QA demands. Ehler et al fabricated an anthropomorphic phantom to test the feasibility of rapid prototyping for patient-specific QA. 28  printed replication of a commercial PET/CT phantom to the commercial phantom itself. 30 In terms of linac-based QA, little effort has been explored in creating 3D printed models to meet specific QA purposes.
The aim of this work is to determine the feasibility of generating a 3D printed QA phantom in order to test the accuracy and reproducibility of a 6DoF table alignment. The 6DoF phantom will be constructed such that the tests can be performed in a streamlined fashion. The workflow will be aimed to follow typical end-to-end testing, where it will be CT simulated, planned for isocentric localization, setup at the treatment linac, and imaged following stereotactic IGRT protocols. After establishing baseline alignment shifts and isocentricity values, the phantom can be implemented into the daily QA routine.  The 3D model was created using a computer-aided design (CAD) freeware program. The final version of the design was saved as a stereolithography (STL) file. Once generated, the STL file was imported into a proprietary 3D printing preparation software and converted into a language compatible for the 3D printer called a GCODE file. A GCODE file is a numerically controlled programming language that gives multivariate instructions regarding printer extruder location, speed, temperature, and rate, along with other printerspecific settings.

2.C | Establishment of end-to-end test
In order to establish a streamlined test for 6DoF registration and mechanical motion, an end-to-end test was devised. Figure 1 shows one example of the 3D printed phantoms. The CAD model was built with known angular and translational offsets for alignment. The faces of the print were designed with 2.0°angular offset for the yaw, pitch, and roll rotations. An additional 3D printed leveler was also fabricated so that the skewed phantom would be leveled when used together in every rotation for reference imaging. Two sets of 1.0 mm wide lines were designed on the faces of the phantom: one for an initial offset alignment and one for final isocenter localization.
With a concept similar to a hidden target test, the offset marks were used initially to set up the phantom with the lasers and field light. phantom. In order to facilitate the auto-registration algorithm, unique registration structures were designed into the faces of the phantom.
If registered correctly, the lasers and field light should be coincident with the isocenter lines.
Three models of the 6DoF phantom were printed and customized for the three different linacs equipped with 6DoF tables.
These phantoms were to be tested on a daily basis. CT imaging for each was performed using a Discovery CT590 RT (General Electric Healthcare, Chicago, IL, USA). Each phantom was placed in the leveler for a straight alignment in order to establish a corrected reference image. The leveler was also 3D printed with a low infill percentage such that the registration algorithm would not be effected within the region-of-interest (ROI) of the phantom. The thinnest slice thickness (0.625 mm) was used in order to achieve the highest spatial resolution in the scan plane. The CT images were sent to Eclipse treatment planning system (Varian Medical Systems, Palo Alto, CA, USA). A plan with a CBCT setup field and six 3.0 9 3.0 cm 2 MLC-shaped fields was generated. The six treatment fields were used to deliver a Winston-Lutz (WL) test 31 at the four cardinal gantry angles and two additional collimator angles when the gantry is at 0°.
With the leveler removed, each phantom was setup at its designated machine. Each phantom was aligned to the offset lines using the laser and field lights. A table indexer was used in order to place the phantom flush on its end for precise yaw rotation alignment.
Once aligned, a half-arc CBCT was taken. Figure 2 shows the online match of the reference CT and CBCT. A coarse, manual registration was performed only for translation adjustments. Auto-registration was followed up for rotational and fine translational adjustments.
Unique registration structures were utilized to visually verify the registration (Fig. 3). Translational and rotational shifts were then applied, with the table moving to the center of the BB at the treatment isocenter. Immediately after shifting, the six WL fields were delivered using the electronic portal imaging device (EPID). After WL delivery, laser and field light coincidence was verified with the isocenter lines indicated on the phantom. The WL test was analyzed off-line.
The daily end-to-end test was carried out for 30 days for each of the three linacs. Shifts for the three translational and three rota-   Combined data for all three linacs was also tabulated in Table 1 (N = 90). The combined registration values were 1.54 AE 0.13 cm,  The 95% confidence level was within 0.20 mm or smaller for each of the linacs. Figure 6 shows the Dmax data in a linear plot style.
After delivery of the WL test, a visual inspection of the laser and field lights were performed in order to verify that they were impinging on the 1 mm indentations of the phantom. The test was analyzed with either a pass or fail criteria. The laser and light field test passed for each linac for the 30 day period.   Gray" is the nickname of the linac, which has been denoted as "Linac 1" throughout the report. Note that the leveler from the reference CT is visible in the blended image.  Given this workflow, the WL test now inherently includes more uncertainties throughout the end-to-end process. Uncertainties

3.C | Phantom intercomparison
Example of Winston-Lutz measurement with six field delivery (G=gantry angle, C=collimator angle). The blue "+" represents the center of the BB, while the red "X" represents the center of the field size.  Furthermore, the 3D printing market in 2014 was $700 million industry, with an increased projection of $9 billion within 10 year from that year. 34 Thus, unlike the QA phantom market in radiation oncology, the options for 3D printers are relatively vaster. When choosing a 3D printer capable of having the same precision as a commercially available phantom, it is important to consider the specifications and additional features. Most importantly, the printer would need to provide submillimeter accuracy, as most TG-142 tolerances for stereotactic purposes require less than or equal to 1 mm. While the quoted specifications of the printer stated submillimeter accuracy, it was observed that the variance between each phantom and the designed offset dimensions were on the order of millimeter magnitude. This could stem from additional factors beyond specifications from nozzle armature precision. Amid our initial iterations of printing, the phantom would both warp and detach from the printer bed. This can be alleviated by having a printer with a heated bed and proper insulation. Commercial adhesive can also be applied to the bed of the printer in order to affix the first layer of material to the bed surface to prevent detachment. Other features like supporting structures, percent infill, and printer speed can also affect the quality of the print. While this is not an exhaustive list of considerations, the user needs to be aware of the capabilities of the 3D printer and gain experience using the printer in order to characterize its printing capacity. By fully understanding the 3D printer, one can fabricate the most optimal QA phantom.

| CONCLUSION
This study investigated the use of a 3D printed phantom in order to perform a streamlined, end-to-end QA test on a 6DoF table. Three individual 3D printed phantoms for three linacs were fabricated with known translational and rotational offsets from a central BB. The phantom was CT simulated in a corrected orientation as a reference.
A plan was created with a CBCT setup field followed by WL fields for mechanical and radiation isocenter verification using an EPID.
The phantom was setup to the designed offset marks, cone-beamed and 6DoF registered, delivered and analyzed the WL fields, and veri-