Three‐dimensional printing CT‐derived objects with controllable radiopacity

Abstract Purpose The goal of this work was to develop phantoms for the optimization of pre‐operative computed tomography (CT) scans of the prostate artery, which are used for embolization planning. Methods Acrylonitrile butadiene styrene (ABS) pellets were doped with barium sulfate and extruded into filaments suitable for 3D printing on a fused deposition modeling (FDM) printer. Cylinder phantoms were created to evaluate radiopacity as a function of doping percentage. Small‐diameter tree phantoms were created to assess their composition and dimensional accuracy. A half‐pelvis phantom was created using clinical CT images, to assess the printer's control over cortical bone thickness and cancellous bone attenuation. CT‐derived prostate artery phantoms were created to simulate complex, contrast‐filled arteries. Results A linear relationship (R = 0.998) was observed between barium sulfate added (0%–10% by weight), and radiopacity (−31 to 1454 Hounsfield Units [HU]). Micro‐CT scans showed even distribution of the particles, with air pockets comprising 0.36% by volume. The small vessels were found to be oversized by a consistent amount of 0.08 mm. Micro‐CT scans revealed that the phantoms' interiors were completely filled in. The maximum HU values of cortical bone in the phantom were lower than that of the filament, a result of CT image reconstruction. Creation of cancellous bone regions with lower HU values, using the printer's infill parameter, was successful. Direct volume renderings of the pelvis and prostate artery were similar to the clinical CT, with the exception that the surfaces of the phantom objects were not as smooth. Conclusions It is possible to reliably create FDM 3D printer filaments with predictable radiopacity in a wide range of attenuation values, which can be used to print dimensionally accurate radiopaque objects derived from CT data. Phantoms of this type can be quickly and inexpensively developed to assess and optimize CT protocols for specific clinical applications.


| INTRODUCTION
This work was motivated by a new Vascular and Interventional Radiology (VIR) procedure for the treatment of benign prostatic hyperplasia; Prostate Artery Embolization (PAE). 1,2 During the procedure, a microcatheter is navigated into bilateral prostatic arteries under x ray guidance. Embolization is then performed, causing the prostate gland to shrink over time, thus alleviating urinary symptoms and improving quality of life. 2 PAE is technically challenging, even for experienced VIR physicians. This is mainly related to the prostatic arterial vascular anatomy where significant anatomic variants of prostate arteries exist, in terms of their origin, number, and course. 3 In addition, the prostate arteries often have very small diameters, on the order of 0.5 to 1.5 mm. 4 While C-arm CT has sufficient spatial resolution to image these arteries, it is performed in the angiography lab during the procedure.
Understanding the anatomy and planning the optimal C-arm angles for device navigation prior to the embolization procedure has the potential to shorten procedures times, reduce cost, and reduce overall patient radiation exposure. A pre-procedure CT angiogram (CTA) can therefore be very helpful in identifying these complex pathways.
However, the inherently small prostatic arteries in some men are near the spatial resolution limit of most clinical CT scanners, and therefore may not be adequately visualized by a generic abdominal CTA protocol. The goal of the work presented here is to create patient-specific complex anatomical phantoms for use in the CTA scanner protocol optimization for PAE.
Object visibility in CT is a function of both the size and contrast of the object in question. 5 Commercial CT scanners have a multitude of data acquisition and image reconstruction parameters that affect the spatial resolution and contrast resolution of the resulting image.
In addition, the contrast media injection protocol also affects object visibility, especially for the small arteries leading to the prostate. Visibility is also a function of the image visualization format; for example, in our experience, an artery may not be visible in a direct volume rendered (DVR) format and yet be visible in a maximum intensity projection (MIP) format.
While there are several commercial phantoms available to measure CT system parameters, they can be expensive and may not present the specific combination of attributes needed to optimize a CT data acquisition and image reconstruction protocol for a specific clinical scenario. For example, in the case of prostate artery imaging, the arteries are filled with some level of contrast, while embedded in soft tissue near the pelvis, which itself has a high level of attenuation.
Tissue-equivalent materials, used in the construction of physical x ray phantoms, have been in development since the early 1900s. 6 The first generation of physical phantom models were wax-based, and included radiopaque fillers such as calcium carbonate, polyethylene, silicon dioxide, and titanium dioxide. More recently, epoxy resins have been used with the same additives, 7 but also with phenolic microspheres to adjust the model density. 8 Even before the advent of 3D printing, researchers used CT images to manufacture phantoms, making multi-layered molds for collections of urethanebased tissue-mimicking materials. 9 This prior work, however, differs from this study in that the investigators sought to mimic the xray properties of tissues at all x ray photon energies in the diagnostic spectrum using a combination of materials. Furthermore, this earlier work focused only on tissues (bone, lung and soft tissue) and not iodine-filled blood vessels, as encountered in CTA.
3D printing anatomical structures from medical image data began in the early 1990's, and has been used primarily for three purposes: surgical planning, resident and patient education, and implantable prostheses. 10 Cranio-facial surgeons use stereolithographic models derived from CT scans for surgical planning. 11 These models are used to show the relationship of complex anatomic structures that may not be fully appreciated when viewing a 2D image monitor, even for a DVR image. Surgical residents can attain a better understanding of pathological anatomy through the study of 3D anatomical models. 12 Models can also be used to better explain complex surgical procedures to patients and their families. While early attempts at prosthesis printing were problematic, mainly due to the materials involved, 10 this area nevertheless holds great promise for the future, as printing methods improve and material selections expand. 3D printing has also been used to create CT-derived molds for tissue materials that cannot be printed, because of their nanoscale structure, 13 and to create hollow vascular structures which can be incorporated into flow phantoms. 12,14 Recently, there have been attempts to use various 3D printing technologies for constructing radiopaque objects. Jahnke et al. used a standard ink-jet dot matrix printer filled with a potassium iodide solution to print radiopaque phantoms, derived from CT data, in several layers of printer paper which were then stacked to form 1 cm thick, three-dimensional phantoms. 15 18 Attenuation values for these phantoms ranged from 70 to 121 HU (at 120 kVp), which again was restricted to the range of commercially available materials. To simulate contrast-filled vessels in the liver, a dissolvable support material was used during the printing process. Once removed, this volume was filled with an iodine solution. In terms of cost, PolyJet systems start at $125,000, with material expenses ranging from $0.30/cm 3 to $0.50/cm 3 .
FDM 3D printers use a solid filament which is melted, extruded through a nozzle (positioned by three stepper motors), and deposited onto the printed object where it adheres, quickly cools, and hardens.
These printers are a potentially attractive option, due to their spatial resolution (< 1.0 mm) and the low cost of both the system (< $5,000) and the printer material (< $0.03/cm 3 ). Unfortunately, while there are an increasing number of materials available for these printers, specific attention is not routinely paid to radiopacity. The two most common materials used in these printers, polylactic acid (PLA) and acrylonitrile butadiene styrene (ABS), have radiopacities lower than water. 19 Ceh et al. were able to overcome this limitation using filaments that were manufactured by a third party, consisting of combinations of bismuth and ABS, to print both radiopaque 3D bone phantoms and x ray shielding equipment on an FDM printer. 20 It is noteworthy that bismuth has a K-edge at 90.5 keV, and therefore does not mimic tissue properties at specific x ray energies, but rather is only attenuating in aggregate. Madamesila et al. used an FDM printer and (commercially available) high impact polystyrene filament to fabricate low density phantoms for use in radiotherapy quality assurance applications. 21 Variations in attenuation were achieved using the 3D printer's infill percentage parameter, which controls the percent of an object's inner volume that is filled with printer material, the remainder containing air. For this technique, mean attenuation values ranged from 0 to À800 HU. Similar work, using PLA, was performed by Oh et al. 22 Our goal in prostate imaging is to develop a phantom, embedded in a water bath of human dimensions, which contains a pelvis and an iliac artery tree, to use in optimizing the CTA imaging protocol.
The phantom will be constructed such that iliac artery trees of various shape, dimension and opacity may be inserted into the phantom.
The purpose of the work presented here is to develop a reliable and inexpensive method of manufacturing CT phantoms by extruding 3D printer filaments composed of varying mixtures of a standard FDM 3D printer material and a radiopaque doping agent, and then using those filaments to produce patient-specific CT phantoms of the pelvis and iliac artery tree, with controlled radiopacity.
Two doping agents were evaluated: calcium carbonate and barium sulfate. Calcium carbonate has a long history of use in bone phantoms. 7,8 Barium sulfate has a K-edge at 37.4 keV, and therefore has not been used for bone. However, its K-edge is close to that of iodine (33.1 keV), making it a reasonable material to simulate iodinefilled blood vessels. The potential problem in using calcium carbonate is the volume of material required to achieve the radiopacity of bone, relative to the base material. To evaluate this issue, a mixture of 20% calcium carbonate powder and 80% ABS plastic, by weight, was extruded. Mineral oil was added to facilitate adherence of the calcium carbonate to the ABS pellets. For this mixture, 15% of the calcium carbonate did not adhere to the ABS pellets and was left at the bottom of the container. In our experience, powder that is not adhered to the pellets will settle at the bottom of the extruder. Test objects printed with this filament had an attenuation of approximately 65 HU. Furthermore, a micro-CT scan revealed a significant number of air pockets. We therefore concluded that it is not possible to 3D print material with a radiopacity approximating cortical bone (~1000 HU) using calcium carbonate and this filament creation technique.
Using the NIST XCOM website, 23 it was determined that the linear attenuation coefficient of barium sulfate at 50 keV is more than 25 times higher than that of calcium carbonate. Since barium sulfate would require much less material to achieve a similar radiopacity, we chose to proceed with using this doping agent for all phantoms, including both the iliac artery and pelvis phantoms.
Two materials were evaluated as the base material for the extruded filaments: PLA and ABS. Pellets of both material types, doped with 5% barium sulfate by weight, were extruded in a Filabot Original Filament Extruder (Filabot, Barre, VT). While an exhaustive evaluation was not performed, the ABS pellets yielded more consistent results in terms of filament extrusion, print reliability, and object durability. With PLA we had difficulty achieving a large and consistent diameter, there were more print failures, and the resulting prints were more fragile. We therefore proceeded using ABS.

2.B | Phantoms printed and scanned
Four types of phantoms were printed using the barium sulfate doped filaments. First, simple cylinder phantoms were printed to assess the HU values as a function of barium sulfate added. Next, complex artery tree phantoms were designed and printed to assess the accuracy of the printer in printing cylindrical segments of small diameter.
Third, a pelvis was printed to develop and assess a method to print large multi-piece objects derived from CT data, and to assess the use of the infill percentage parameter to produce an object with an inner core (representing cancellous bone) that had a lower HU value than the outer shell (representing cortical bone). Finally, a CT-derived iliac artery tree, including the prostate artery, was printed to assess the system's ability to print complex vessel anatomy using the radiopaque filament.
The phantoms were printed on Lulzbot TAZ 5 and TAZ 6 FDM printers (Aleph Objects, Inc., Loveland, CO). Cura software (Lulzbot version 21.04, Ultimaker, Netherlands) converted the stereolithograph files defining each phantom into G-code (printer commands), and was also used to set the printing parameters. The filament diameter parameter allows the printer to adjust the feed rate of the filament to achieve a consistent volumetric extrusion rate from the nozzle. This was set equal to the measured average filament diameter (defined above) of the filament segment created. The remaining printer parameters used are listed in Table 2  Rafael, CA) and exported into a stereolithography (STL) file. Two batches of each filament type were extruded, and four cylinders were printed from each batch, resulting in a total of eight cylinders per filament type. The cylinders were printed using an infill percentage of 100%. To evaluate the content of the cylinders prior to scanning, density measurements were taken using the immersion method described in ISO 1183-1. 24 In this method, an object's mass is measured both in air and while suspended in a water bath that is not sitting on the scale. Using these two measurements, the density is obtained as: where q is the specimen's density, m a is the mass measured in air, and m w is the mass measured with the object suspended in water.
This was then compared with the theoretical density of the mixture, computed as:

2.B.2 | Artery tree phantom
To evaluate the ability of the printer to produce completely filled, small-diameter segments, an artery tree phantom was CAD-designed (Creo V2, PTC, Needham, MA), containing segment diameters of 1.0, 2.0, and 4.0 mm (Fig. 1). The printer had a nozzle diameter of 0.5 mm, and it was questionable whether a 1.0 mm diameter branch could be printed correctly with this printer and these types of filaments. Three phantoms were printed (with 2.5%, 5% and 10% doped filaments) using an infill percentage of 100%. Ten measurements, at arbitrary locations on each phantom, for each tree segment diameter, were taken with a digital caliper, for a total of 30 measurements for each diameter. Micro-CT scans of representative sections of one phantom were taken to evaluate the internal make-up of the phantom, including air pockets.

2.B.3 | Pelvis phantom
A phantom, derived from a CT scan, of the left half of a human pelvis was constructed. The pelvis is composed of a dense cortical bone shell, with an inside of spongy cancellous bone. The 3D printer is capable of manufacturing structures with both a programmable shell thickness and a programmable infill percentage. These printing parameters were recruited to demonstrate the ability of the printer to print bone models with variable cortical bone thickness and variable cancellous bone radiopacity. The phantom was designed such that the cortical bone would be considered the shell of the 3D printed object (filled completely with filament), and the cancellous bone would be considered the inside of the object (filled with a filament support grid). When filled with water, the cancellous bone radiopacity of the phantom would be a function of the filament HU value and the infill percentage, computed as: where infill is the 3D printer infill percentage, HU filament is the HU value of the filament, and HU water is the HU value of water (= 0 HU).
The phantom was based on a SLU Hospital patient CT scan.

3.A | Extruded filament
Filaments of all doping levels were successfully extruded. Representative results of the filament diameters and lengths are shown in Fig. 5.
The resulting filaments ranged in average diameter (in their regions of consistent diameter) from 1.72 to 2.43 mm with standard deviations ranging from 0.071 to 0.095 mm. The standard deviation for one commercially produced 2.85 mm filament was measured to be 0.021 mm.
Contiguous filament segment lengths ranged from 3 to 10 m. Shorter segment lengths corresponded to larger filament diameters. Shorter segment lengths did not affect the printing of larger objects (e.g., the pelvis components) as the printer is able to pause for the insertion of a new filament segment, and then seamlessly continue.

3.B.1 | Cylinder phantoms
Once printed, the density of each of the 40 cylinders was measured.
Density measurements were able to identify one cylinder that was not consistent because of its low barium sulfate content. This difference was confirmed by the CT scan. It was found that this filament came from a segment too close to the beginning of the extrusion, where the filament still contained undoped barium sulfate pellets.
This cylinder was not included in the analysis. Micro-CT images were analyzed to assess both particle distribution and the presence of air pockets. Upon visual inspection of the images, as shown in Fig. 7 for a 5% doped cylinder, the distribution of barium sulfate particles appeared even, and there were few artifacts resulting from the printing process. One example artifact (showing a seam between the shell and area of infill) is evident in

3.B.2 | Artery tree phantom
Results of the dimensional measurements of the artery tree models are shown in Table 4. All measured diameters were slightly larger than the CAD-specified diameter, by an average of 0.08 mm. A micro-CT image of representative sections of each artery diameter is presented in Fig. 8. Another purpose of the scan was to evaluate the ability of the 3D printer to completely fill segments with small diameters. Even though the printer nozzle diameter (0.5 mm) is close to the diameter of the vessel segments, the printer was able to completely fill in the area of all three segment types, with minimal air pockets and minimal printer-induced artifact.  infill parameter, was 193 HU. These data are summarized in Table 5.

3.B.3 | Pelvis phantom
Matching example profile plots were made for one region of each slice, shown in Figs. 9(c) and 9(f). Two profile plots were made for each phantom: one crossing a water-filled section, and one crossing an infilled section.
The filament segments required to make the phantom totaled 49 m in length, had an extrusion time of 9 h, and cost $9.80. The total print time was 16 h.

3.B.4 | Prostate artery phantom
The CT scans of the two prostate artery phantoms (365 HU and 555 HU, each scanned with the pelvis phantom in a water bath) were evaluated in both axial-planar and DVR formats. Mean    Anecdotally, we found that objects printed with commercial filament were slightly more consistent than those printed with our extruded filament. It seems clear that the process presented here would benefit from a commercially extruded filament. The quality assurance steps of measuring object density and measuring filament diameter were critical to achieving predictable results.

| DISCUSSION
Micro-CT images revealed an even distribution of the radiopaque agent within the phantoms, with only very small air pockets or other artifacts of the FDM printing process, as shown in Figs. 7 and 8.
While vessel segments as small as 1.0 mm could be reliably printed, there was a small bias in the diameter (+0.08 mm). This seems to be the result of printing with a 100% infill percentage. The printer manufacturer recommends a 70% infill percentage even for load-bearing parts, and states that using a 100% infill percentage will result in oversized objects. 26  parameter. An investigation of a greater variety of infill percentages using small CAD-based phantoms would better characterize this relationship. It should also be noted that the documented infill percentage is slightly higher than the actual infill percentage, as the software does not account for the overlap in the crosshatching pattern [as shown in Fig 9(b)]. 26 We filled the phantom with water to reach clinically relevant attenuation values in the cancellous bone sections with a lower infill percentage (and therefore less filament) than would have been required for air. However, it was difficult to entirely eliminate the air from the phantom, as shown in Figs. 9(b) and 9(e). Care was taken to divide the pelvis into sections for which there would be a minimum of completely enclosed spaces, and the top and bottom of each phantom were not printed to allow water access. However, during the attachment process, the edges of each section had to be raised out of the water slightly to be fused, allowing some air to return. Future experiments will include the use of marine grade epoxy for attachment. One limitation of this phantom is that the material is not a true tissue-mimicking material for bone. Because of the K-edge in barium sulfate x ray attenuation, it does not have the linear attenuation coefficient of tissue at each of the photon energies emitted by a clinical CT scanner x ray tube. Calcium carbonate doping could provide such a property, however, we were not able to achieve usable attenuation levels with this material and our technique.
The attenuation of the prostate arteries was also affected by the CT scanning and image formation process. The smaller arteries had lower attenuation than the larger arteries, clearly the result of some combination of partial volume effects and reconstruction filtering.
Future work will seek to determine which scanning and reconstruction protocols minimize this difference, and also how these protocols affect artery visibility. These phantoms could also be used to evaluate contrast delivery. We have observed lower HU values in the prostate arteries in clinical scans, relative to those in the iliac arteries. The question is whether this is a processing artifact, as described above, or a problem with the contrast timing. This phantom could be used to determine the HU value for a small prostate artery when there is consistent contrast filling throughout the iliac artery tree, which could then be compared to clinical CT scans to potentially detect a suboptimal contrast algorithm: a combination of contrast amount, injection rate, and scan delay.
As shown in Fig. 10, the surface texture of the models produced in this work did not match that of the anatomy, as imaged by the CT scanner and visualized by the DVR technique. While 3D objects are built in discrete layers in FDM printers, we do not believe this is the cause of the issue. The printer layer height parameter of 0.25 mm would induce artifacts of a much higher spatial frequency.
The difference is more likely the result of the image segmentation technique, which could be addressed in two ways. It is possible to smooth a segmented image with Mimics prior to STL file export. This effect could be assessed prior to export by examining the edges of the segmented model. It is also possible to smooth the printed object by brushing it with acetone. As medical image interpretation is subjective, we believe this is an important issue to address going forward.
While Polyjet printers have higher spatial resolution and a wider range of materials overall, they are much more expensive and are currently not able to achieve the attenuation values required for bone and contrast-filled artery phantoms. The capital equipment cost for this work was $5100.00 and the object printing cost, including ABS pellets and barium sulfate, was $0.03/cm 3 . Finding a commercial maker of these filaments would make printing the phantoms easier and more reliable, and would reduce construction time by more than one-third.
Future work includes printing a variety of CT-derived prostate arteries in a range of radiopacities, for use in the optimization of CT scanner acquisition and reconstruction settings for optimal DVR visualization of the prostate artery. This will provide a better understanding of the specific tradeoffs between contrast dose, radiation dose and image quality for this application. More reliable and higher quality imaging of these arteries will also enable clinicians to better plan the embolization procedure, including device selection and C-arm angulations, which should lead to reduced procedure times and procedure HAMEDANI ET AL.
| 327 cost. It may also be possible to identify patients who are not candidates for the procedure, due to excessive vessel tortuosity.
In addition to PAE, there are other innovative procedures, such as Bariatric Embolization, that could benefit from this approach to pre-procedure CT image optimization, as VIR physicians continue to pursue more targeted intra-vascular therapies. These phantoms can indeed be applied to the development of any new CT imaging protocol where the anatomical complexity may not be well represented by more general purpose phantoms. CT protocols are often developed iteratively, in patients. Much of this iteration could potentially be avoided. Finally, these tools could be used to better interpret the results of a specific CT scan, as exemplified by the cortical bone attenuation in the pelvis phantom.