Three-dimensional printing accuracy analysis for medical applications across a wide variety of printers

Abstract.

Purpose

Three-dimensional (3D) printing has had a significant impact on patient care. However, there is a lack of standardization in quality assurance (QA) to ensure printing accuracy and precision given multiple printing technologies, variability across vendors, and inter-printer reliability issues. We investigated printing accuracy on a diverse selection of 3D printers commonly used in the medical field.

Approach

A specially designed 3D printing QA phantom was periodically printed on 16 printers used in our practice, covering five distinct printing technologies and eight different vendors. Longitudinal data were acquired over six months by printing the QA phantom monthly on each printer. Qualitative assessment and quantitative measurements were obtained for each printed phantom. Accuracy and precision were assessed by comparing quantitative measurements with reference values of the phantom. Data were then compared among printer models, vendors, and printing technologies; longitudinal trends were also analyzed.

Results

Differences in 3D printing accuracy across printers were observed. Material jetting and vat photopolymerization printers were found to be the most accurate. Printers using the same 3D printing technology but from different vendors also showed differences in accuracy, most notably between vat photopolymerization printers from two different vendors. Furthermore, differences in accuracy were found between printers from the same vendor using the same printing technology, but different models/generations.

Conclusions

These results show how factors such as printing technology, vendor, and printer model can impact 3D printing accuracy, which should be appropriately considered in practice to avoid potential medical or surgical errors.

1. Introduction

Three-dimensional (3D) printing has been demonstrated to be an important tool for the advancement of personalized patient care.^1–5 Using volumetric image data, such as computed tomography (CT) and magnetic resonance imaging (MRI), or computer-aided design (CAD) software, a 3D model can be designed and printed.^1–5 Medical-related applications are found in practice, research, and education, ranging from osteotomy and other surgical guides to highly detailed anatomic models for teaching.^1,6–14 Many academic hospitals now have in-house 3D printing labs, also known as point-of-care manufacturing.

However, there is a lack of thorough and standard procedures for quality assurance (QA) to ensure that the 3D printed model accurately reproduces patient-specific anatomic details, which are crucial given the applications. This becomes even more problematic given that there is high variability between 3D printers of the same technology as well as between distinct printing technologies, vendors, and printer capabilities. This is further complicated by the lack of identified and universally accepted standardization groups for 3D printing and related quality control (QC) protocols and tolerances.

Hence, individual users or 3D printing labs have developed their own basic QA procedures that are reported in the literature.^15–20 For example, Leng et al.¹⁵ developed a 3D printing QA framework including a medical 3D printing phantom as well as standard procedures from image acquisition to printing. George et al.¹⁶ investigated various factors in 3D printing that contribute to model inaccuracy and metrics to quantify the difference. Odeh et al.²⁰ investigated several measurement techniques and assessed the pros and cons of each technique. There has also been previous work studying the accuracy of 3D printers. Note that accuracy refers to the closeness to a truth value, whereas precision refers to the consistency and the closeness of data values to one another. Hatz et al.²¹ studied the accuracy of mandible models on material extrusion and powder bed fusion printers. Msallem et al.²² studied mandible model accuracy using material extrusion, powder bed fusion, vat photopolymerization, material jetting, and binder jetting printers. Salmi et al.²³ studied the accuracy of medical models with material jetting, powder bed fusion, and binder jetting printers.

However, previous studies have been limited in that they tend to focus on dental models or models without multiple complex anatomic features or very small details and have had a relatively limited selection of 3D printers. In addition, prior studies have generally not acquired longitudinal data, which could show a trend as printers are updated and begin wearing down.

Therefore, the purpose of this work was to study the differences in accuracy among a wide variety of 3D printers that are commonly used in the medical field by printing a phantom with varied structures and providing longitudinal data to fill in some of the gaps in the new, but crucial, area of 3D printing QA for medical applications.

2. Methods and Materials

A 3D printing QA phantom developed by Leng et al.¹⁵ in their 3D printing QA framework was used in this study to assess the accuracy of the studied printers. The phantom was designed by our group and has been routinely used as a QA phantom in our practice. The phantom was printed monthly on a wide variety of 3D printers (16 printers from 8 different vendors and 5 distinct printing technologies) for a period of 6 months, and both qualitative assessment and quantitative measurements were obtained. The accuracy and precision were assessed and compared among printer models, vendors, and printing technologies.

2.1. QA Phantom

Figure 1 shows the QA phantom used in this study.¹⁵ The phantom has various geometric structures with known dimensions that are specifically designed to test the various aspects of 3D printing. These include, but are not limited to, raw dimensional accuracy, shape fidelity, curvature, and surface finish. The phantom’s overall $X$, $Y$, and $Z$ dimensions were the entire phantom’s length, width, and height, and the expected values were 100, 100, and 12.5 mm, respectively. The convex and concave portions of the phantom had lengths and heights or depths. The convex length and height were 50 and 7.9 mm, respectively, and the concave portion’s length and depth were 50 and 7.5 mm, respectively. The conical part of the phantom had three dimensions studied: the top diameter, bottom diameter, and height. Figure 1 shows a rendering of the phantom from this study and the 10 dimensions. The phantom also had a vascular tree and cochlear spiral to simulate fine anatomical structures, as well as sets of line pairs with different spatial resolutions of 5, 10, and 20 lp/cm. The vascular branch was designed to have four semicylindrical segments following the largest section.

Fig. 1.

(a) QA phantom used in this study includes various geometric and anatomic structures. (b) and (c) Dimensions of individual components are illustrated.

2.2. Printers

A total of 16 3D printers covering 5 different printing technologies (material jetting, vat photopolymerization, material extrusion, binder jetting, and powder bed fusion), 8 different vendors [Stratasys, Formlabs, NewPro 3D, Ultimaker, PRUSA, 3D Systems, Hewlett-Packard (HP), and EOS], and a total of 10 distinct printer models were included in this study. Table 1 shows the abbreviated names for the printers as well as full model names and vendors. Material jetting and vat photopolymerization printers used acrylate, material extrusion printers used polylactic acid, binder jetting printers used gypsum, and the powder bed fusion printers used nylon 12. A QA process has been implemented for every printer.¹⁵ For each printer, the default parameter settings, as suggested by the manufacturer, were used in this study. Printers are regularly maintained and tested by on-site staff on a weekly and monthly basis; every 3500 print hours, or roughly every year, a technician from the manufacturer will perform maintenance procedures. Printed models are removed from the printers and supporting structures are fully removed with care to ensure that the model is not affected.

Table 1.

Printers ($N=16$) used in this study, with printing methods, vendors, and models.

Material jetting	Vat Photopolymerization	Material extrusion	Binder jetting	Powder bed fusion
Objet (Stratasys Objet 500)	Virtuous Walrus (Formlabs 3B)	Ult. A (Ultimaker S5)	ProJet (3D Systems ProJet)	HP-580 (HP Jet Fusion 580)
J55 (Stratasys J55)	Nifty Froglet (Formlabs 3B)	Ult. B (Ultimaker S5)		EOS P110 (EOS Formiga P110)
	Boundless Panda (Formlabs 3B)	PRUSA (PRUSA i3 MK3S)
	Upbeat Farrow (Formlabs 2)
	Sharp Chicken (Formlabs 3B)
	Helpful Puffin (Formlabs 3B)
	Staunch Vervet (Formlabs 3B)
	NP1 (NewPro NP1)

2.3. Measurements

Over a period of 6 months, the QA phantom was printed monthly on each of the 16 printers included in this study. Both qualitative assessment and quantitative measurements were obtained.

The concave, convex, conical, and overall dimensions were measured by hand at reference points in the structures with calipers to the nearest 100th of a millimeter (Fig. 1). The overall dimensions were measured by fitting the calipers tightly around the entire phantom (Fig. 2). The concave and convex structure lengths were measured by fitting the calipers around the structure from the top side. Heights or depths were measured using a small extension of the calipers (Fig. 2). The conical structure was measured by placing the calipers at the very edge of the structure and drawing a diameter at the top and bottom. Height was measured by placing one end of the caliper at the base and extending the other to the height of the top surface (Fig. 2). Each dimension’s measurements were recorded in a spreadsheet. Qualitative data for smoothness, evenness, and any noticeable distortion or defects were also recorded in the spreadsheet to help with the overall assessment of printing accuracy. A fit test was performed with the positive and negative structures of the phantom, which are designed to pair seamlessly into one another if printed accurately. These structures include the half-spheres, hexagonal prisms, and cone section. The result of the fit test was recorded in the spreadsheet. To quantify the accuracy of the quantitatively measured dimensions, we calculated the raw printing error (the difference between the expected and the measured value) for each measurement of each print. Mean printing errors and specific error data were primarily used in this work for the comparison of printers as opposed to relative or percent error that has been used in other works.²³

Fig. 2.

Caliper measurements of phantom structures: (a)–(c) overall $X$, overall $Y$, and overall $Z$; (d)–(f) cone top, cone bottom, and convex length; and (g)–(i) convex height, concave length, and concave height.

3. Results

3.1. Quantitative Results

The mean printing error for individual printers and dimensions ranged from $-1.75$ to 0.45 mm, with negative values meaning the measured value was less than the expected value and positive values meaning the measured value was greater than expected (Table 2). Printing errors from individual longitudinal measurements for all printers are presented as the line charts in Fig. 3. Most of the 3D printers in this study were considered accurate, with printing errors within ±1 mm. Two printers—the NewPro 3D (NP1) and the EOS Formiga P110—had average errors beyond $\pm 1\mathrm{mm}$, but only for one dimension. Both printers underprinted these specific dimensions. NP1 underprinted the convex height dimension by an average of $-1.18\mathrm{mm}$ ($-14.9\%$), and EOS P110 underprinted the cone height dimension by an average of $-1.75\mathrm{mm}$ ($-8.8\%$). NP1 had individual printing errors beyond $\pm 1\mathrm{mm}$ 14 different times across all 60 measurements. EOS P110 had individual errors beyond $\pm 1\mathrm{mm}$ twice. Note that one of those times was a significant outlier with a printing error of $-11.86\mathrm{mm}$ on a dimension with a truth value of 20 mm ($-59.3\%$). The HP Jet Fusion 580 printer had a printing error beyond $\pm 1\mathrm{mm}$ twice as well, but there were no major outliers, which kept the average printing error within 1 mm. These errors were $-1.25$ and $-1.01\mathrm{mm}$ on the concave length dimension corresponding to percent errors of $-2.5\%$ and $-1.02\%$. The Ultimaker A printer went beyond $\pm 1\mathrm{mm}$ six times, $-2.31$, $-1.26$, and $-1.37\mathrm{mm}$ on the concave height dimension corresponding to percent errors of $-30.8\%$, $-16.8\%$, and $-18.3\%$, with errors of $-1.16$, $-1.29$, and $-1.12\mathrm{mm}$ on the convex height dimension, corresponding to percent errors of $-14.7\%$, $-16.3\%$, and $-14.2\%$. Our ProJet printer went beyond $\pm 1\mathrm{mm}$ once with an error of $-1.13\mathrm{mm}$ on the concave length dimension, corresponding to a percent error of $-2.26\%$. Longitudinally, a positive trend in printing accuracy was observed for NP1 (Fig. 4).

Table 2.

Mean printing errors (mm) ± standard deviation from distinct phantoms printed at different time points.

	Objet	J55	Form 3B	Form 2	NP1	Ultimaker	PRUSA	ProJet	HP-580	EOS P110
Concave height	−0.23 ± 0.04	−0.45 ± 0.1	−0.29 ± 0.06	−0.43 ± 0.07	−0.82± 0.92	−0.64 ± 0.70	−0.48 ± 0.30	−0.41 ± 0.08	−0.16 ± 0.06	−0.47 ± 0.08
Concave length	0.01 ± 0.05	0.12 ± 0.15	0.05 ± 0.14	−0.03 ± 0.16	−0.98 ± 0.83	−0.10 ± 0.15	−0.28 ± 0.28	−0.37 ± 0.46	−0.66 ± 0.36	−0.49 ± 0.59
Cone bottom	−0.15 ± 0.09	−0.15 ± 0.12	−0.01 ± 0.11	0.04 ± 0.11	−0.58 ± 0.93	0.07 ± 0.25	−0.15 ± 0.08	−0.08 ± 0.11	−0.31 ± 0.08	−0.23 ± 0.10
Cone height	0.01 ± 0.03	0.11 ± 0.25	0.06 ± 0.15	0.01 ± 0.07	0.28 ± 0.22	0.12 ± 0.31	0.15 ± 0.13	0.24 ± 0.10	−0.02 ± 0.07	−1.75 ± 4.46
Cone top	0.06 ± 0.04	0.14 ± 0.09	0.10 ± 0.06	0.15 ± 0.15	−0.25 ± 0.76	0.19 ± 0.14	0.19 ± 0.05	0.29 ± 0.24	0.40 ± 0.08	0.35 ± 0.03
Convex height	−0.68 ± 0.23	−0.71 ± 0.23	−0.85 ± 0.24	−0.87 ± 0.10	−1.18 ± 0.85	−0.47 ± 0.50	−0.35 ± 0.28	−0.81 ± 0.03	−0.52 ± 0.02	−0.72 ± 0.09
Convex length	−0.09 ± 0.06	0.04 ± 0.13	0.06 ± 0.10	0.22 ± 0.21	−0.61 ± 0.95	−0.01 ± 0.14	−0.14 ± 0.15	0.09 ± 0.21	−0.34 ± 0.16	−0.08 ± 0.18
Overall $X$	−0.02 ± 0.02	0.12 ± 0.13	0.14 ± 0.16	−0.46 ± 0.05	−0.50 ± 0.80	0.01 ± 0.16	−0.07 ± 0.13	0.20 ± 0.04	−0.33 ± 0.13	0.00 ± 0.05
Overall $Y$	−0.03 ± 0.02	0.02 ± 0.07	0.45 ± 0.23	0.29 ± 0.19	−0.84 ± 1.02	−0.06 ± 0.16	−0.16 ± 0.14	0.22 ± 0.07	−0.23 ± 0.19	−0.01 ± 0.05
Overall $Z$	−0.05 ± 0.05	0.06 ± 0.10	−0.01 ± 0.04	−0.01 ± 0.06	−0.19 ± 0.84	0.07 ± 0.14	−0.01 ± 0.21	0.12 ± 0.09	−0.23 ± 0.13	0.14 ± 0.03

Fig. 3.

Line charts showing individual measurements of printing error for different printers, grouped by printer model.

Fig. 4.

Example of printing accuracy trend over time (NP1). First measurements were typically the worst.

3.2. Qualitative Results

Qualitatively, most of the tested printers performed satisfactorily meaning that no significant or obvious faults were detected on visual inspection. All phantoms were successfully printed without failed parts. There was warping of the smallest line pairs in the phantoms produced by material jetting and vat photopolymerization printers. The smallest line pairs were distorted nearly every time on the phantoms made by Objet, J55, all Formlabs 3B, Formlabs 2, and NP1. The medium line pairs were distorted less often than the small (though medium line pair distortion was still quite prominent). There was a lack of surface and curvature smoothness for the convex portion of the phantoms produced by all the material extrusion printers. Curvature smoothness was poorly reproduced by the Ultimaker printers which also consistently failed the hex portion fit test. Surface smoothness was poorly reproduced by the PRUSA printer and only produced three out of four total branches in the vascular tree as opposed to the full four being produced nearly every time by every other printer. Powder bed fusion printers also produced artifacts. For phantoms printed by HP-580, there were consistent raised edges on the positive features of the phantom as well as wrinkles on the side and back faces. Wrinkled surfaces were also present on phantoms printed by the EOS P110 printer, and the phantoms consistently failed the conical and hexagonal fit tests (Fig. 5).

Fig. 5.

Surface defects: (a) shows raised edges and (b) shows wrinkles.

4. Discussion

In this study, printing accuracy was assessed for a variety of 3D printers commonly used in medical applications. A printer was defined as being accurate if the error between actual and measured values was less than $\pm 1\mathrm{mm}$. According to this criterion, material jetting and vat photopolymerization printers were the most accurate, i.e., highest performing printers. Although other tolerances could be employed, e.g., percent deviation, absolute deviation was seen as being the most direct measure of spatial fidelity. Additional trends discovered were that every printer, on average, underprinted the concave and convex height dimensions more severely than every other dimension (e.g., Overall $X$, $Y$, $Z$, cone), and that the NewPro NP1 was consistently less precise than the other printers with higher standard deviations. The data also showed that there were differences in accuracy between printers of the same printing technology but from different vendors. This difference was seen between our Formlabs printers and NP1, both of which employ vat photopolymerization printing technology. The study also detected differences in accuracy, albeit minor, between printers of the same printing technology and vendor, but different models. This was observed in the material jetting printers, Objet 500 and J55, both of which were manufactured by Stratasys. Objet was more accurate than J55 for 8/10 of the dimensions. This was also observed for certain dimensions in the Formlabs printers. The Form 2 printer was less accurate for three dimensions compared with the Form 3B printers.

In a previous work by Ravi et al., printers using vat photopolymerization were found to have printing errors always within 1 mm, which is similar to the results of this work. Hatz et al. found that their selective laser sintering (SLS) printer had a mean printing error across all measurements of $-0.036\mathrm{mm}$ with a standard deviation of 0.114 mm, and their material extrusion printer had a mean error across all measurements of $-0.055\mathrm{mm}$ with a standard deviation of 0.227 mm. The mean errors reported by Hatz et al. were smaller than what was found in this work. The average error across all dimensions for our powder bed fusion printers, comparable to Hatz’s SLS printers, was $-0.28\mathrm{mm}$; the average error for our material extrusion printers, comparable to Hatz’s FFF printers, was $-0.11\mathrm{mm}$. This discrepancy may be due to the different objects that were printed in each study and different measurement techniques. Salmi et al. reported relative errors in the form of a percentage and found that the mean errors of their PolyJet, SLS, and “3DP” printers were 0.18%, 0.79%, and 0.67%, respectively. They found the lowest error with their PolyJet printers, comparable to our material jetting printers, which is similar to our findings, and the worst error with their SLS printers, comparable to our powder bed fusion printers. However, the percent errors reported by Salmi et al. seem to be lower than the percent errors in this study when calculated. Msallem et al. found the mean error ± standard deviation of their EOSINT P385 and Ultimaker 3 to be $-0.07\pm 0.08\mathrm{mm}$ and $-0.01\pm 0.16\mathrm{mm}$, respectively, compared to our finding that our powder bed fusion printers and material extrusion printers had average printing errors and standard deviations of $-0.28\pm 1.04\mathrm{mm}$ and $-0.10\pm 0.36\mathrm{mm}$, respectively. Msallem et al. also found the mean errors and standard deviations of their Formlabs 2, Objet 30 Prime, and ProJet CJP 660 Pro to be $0.23\pm 0.39\mathrm{mm}$, $0.17\pm 0.13\mathrm{mm}$, and $0.09\pm 0.11\mathrm{mm}$, respectively, compared to our finding that the mean errors and standard deviations of our Formlabs 2 printer, material jetting printers, and ProJet printer were $-0.15\pm 0.35\mathrm{mm}$, $-0.09\pm 0.27\mathrm{mm}$, and $-0.05\pm 0.39\mathrm{mm}$, respectively.

For phantoms printed by NP1, the first phantom measurement across all dimensions was extremely inaccurate but improved after the first month. This was likely primarily due to the gel membrane used for the first measurement being defective, as it would deteriorate even when not being used. The following measurements had newer, nondefective membranes. This consistent inaccuracy for the first phantom contributes to the worse average printing error.

Consistent qualitative defects included warping of line pairs on material jetting and vat photopolymerization printers and wrinkles and raised edges on powder bed fusion printers. The warping is most likely due to limitations in spatial resolution with the smallest line pairs having a resolution of $20\mathrm{lp}/\mathrm{cm}$. Wrinkling or “elephant skin” on the powder bed fusion printers is likely due to improper control of the bed’s temperature. Raised edges are likely a result of the fluid-like behavior of a layer when it is fused. The layer tends to rise up on the edges during this process. The EOS P110 printer also consistently failed the positive–negative fit test. This is because the negative portions tend to shrink, and the positive portions tend to expand using this printing technology.

There are several limitations related to this study. First, all measurements were made by hand using calipers. This manual step introduces human error due to the precision of the individual making the measurement as well as the individual’s ability to precisely identify and measure the predetermined measurement points on the phantom. Second, measurement of points along curved surfaces created additional challenges when measuring distances using a caliper, further increasing the potential for measurement error. To reduce these errors, other metrology technology, such as surface scanning, can be used by generating full 3D digital models of the actual printed phantom, which can be compared to the original file and offer a more holistic view of the differences between file and fabrication. Despite the potential for additional measurement errors introduced by human measurements that manifest as increased variance of the measured differences, the results do illustrate real and objective differences between individual printer technologies and models. The described protocols therefore provide a useful starting point for the development of QC standards and protocols and wider adoption by the medical 3D printing community. Moreover, the phantom described in this work was not designed to study the ability of 3D printing to characterize inner structures. To quantify these, not only is a phantom needed but also specific measurement techniques. Standard methods, e.g., caliper-based measurements and surface scanning, cannot visualize internal structures, thereby facilitating the need for imaging techniques, e.g., x-ray or CT. This, in turn imposes specific requirements for the phantom and printing materials, i.e., printed inner structures need appropriate contrast to be distinguished among different structures. Thus, the phantom was not designed to test inner structures, and instead was designed to quantify the precision and accuracy of various geometric and anatomic structures that can be easily and quickly measured externally. Another limitation is that the phantom used in this study is only one size and is therefore used to characterize the printer’s performance within the portion of the build volume of the printer where the phantom is placed. As a result, it does not characterize the entire volume. Given that the phantom is designed to detect temporal changes in the printer, it is assumed that any changes in printer performance will be reflected throughout the build volume. Thus, sampling a small subvolume in this manner is assumed to reflect overall printer performance. In addition, the printing of a small volume phantom is necessary to ensure that the test can be performed quickly, and cost effectively, given the frequency of testing required for routine QC evaluation. Future studies may include investigating the effect of build size and position and the stability of phantoms over extended periods of time.

5. Conclusion

3D printing has significant potential for advancing patient care with its plethora of applications, from patient models to surgical guides. However, there are many variations in 3D printers that affect the print accuracy. This study has demonstrated that the accuracy of 3D printing can vary depending on the printing technology, manufacturer, and model. As new printing technologies and advancements are introduced, it is crucial to be aware of the varying accuracy discovered in this work due to the possible detrimental consequences of inaccuracies when 3D printed objects are used in surgery, preoperative planning, custom medical device creation, and the impactful applications to come.

Biographies

Adam Wentworth received his MS degree in Materials Science & Engineering (2011) from the University of Connecticut. He is a senior clinical engineer in the Anatomic Modeling Unit at the Mayo Clinic in Rochester, Minnesota, United States. He has over 10 years of 3D printing experience, has prototyped novel medical devices at the Massachusetts Institute of Technology, holds a professional certification in Solidworks, and a research interest in automation of patient-specific design.

Victoria Sears holds her BS and MS degrees in Bioengineering from the University of Michigan, United States. She is a biomedical engineer within Mayo Clinic’s Anatomic Modeling Unit (AMU) in the Department of Radiology. As part of the clinical care team, she collaborates with physicians, radiologists, and technicians to create accurate 3D printed anatomic models and patient-specific osteotomy guides. She is the AMU’s lead of virtually planning cranio-maxillofacial surgeries.

Andrew Duit received his AAS degree in mechatronics from Alexandria Technical & Community College, in 2019. He is currently a healthcare engineering technician for Mayo Clinic’s Anatomic Modeling Unit in Rochester, Minnesota, United States with a special focus on material extrusion, powder bed fusion 3D printing, and a general focus on 3D printer maintenance and operation.

Eric Erie received his AAS degree in Biomedical Equipment Technology (2015) from Dakota County Technical College. He is currently a healthcare engineering technician for Mayo Clinic’s Anatomic Modeling Unit in Rochester, Minnesota, United States with special focus on photopolymer 3D printing and a general focus on 3D printer maintenance and operation.

Kiaran McGee received his BSc degree in applied physics (1989) from the University of Technology, Sydney Australia, MS degree (1992) in medical physics from the University of Manitoba, Winnipeg, Canada, and PhD (1999) from Mayo Graduate School. He is a professor of medical physics at Mayo Clinic, Rochester, Minnesota, United States. He has authored over 75 peer-reviewed articles. His research interests include MR-based methods for measuring the intrinsic mechanical properties of tissue, development of MR-based methods for radiation therapy treatment simulation, the application artificial intelligence for synthesizing multiple imaging signals, and the application of 3D printing to medicine.

Shuai Leng received his BS degree in engineering physics (2001) from Tsinghua University, and a PhD in medical physics (2008) from the University of Wisconsin, Madison. He is a professor of medical physics at Mayo Clinic in Rochester, Minnesota, United States. He has authored over 230 peer-reviewed articles. His research interest is technical development and clinical application of x-ray and CT imaging, artificial intelligence, and 3D printing in medicine.

Biographies of the other authors are not available.

Disclosures

Institution (Mayo Clinic) owns patent related to 3D printing quality assurance. The authors have nothing else to disclose.