Dimensional accuracy of 3D printed vertebra Kent Ogden a, Nathaniel Ordwayb, Dalanda Dialloa, Gwen Tillapaugh-Faya, Can Aslanc ∗
a
Department of Radiology, bDepartment of Orthopedic Surgery, SUNY Upstate Medical University, Syracuse, NY 13210; cDepartment of Biomedical and Chemical Engineering, Syracuse University, 900 S Crouse Ave, Syracuse NY 13210 ABSTRACT 3D printer applications in the biomedical sciences and medical imaging are expanding and will have an increasing impact on the practice of medicine. Orthopedic and reconstructive surgery has been an obvious area for development of 3D printer applications as the segmentation of bony anatomy to generate printable models is relatively straightforward. There are important issues that should be addressed when using 3D printed models for applications that may affect patient care; in particular the dimensional accuracy of the printed parts needs to be high to avoid poor decisions being made prior to surgery or therapeutic procedures. In this work, the dimensional accuracy of 3D printed vertebral bodies derived from CT data for a cadaver spine is compared with direct measurements on the ex-vivo vertebra and with measurements made on the 3D rendered vertebra using commercial 3D image processing software. The vertebra was printed on a consumer grade 3D printer using an additive print process using PLA (polylactic acid) filament. Measurements were made for 15 different anatomic features of the vertebral body, including vertebral body height, endplate width and depth, pedicle height and width, and spinal canal width and depth, among others. It is shown that for the segmentation and printing process used, the results of measurements made on the 3D printed vertebral body are substantially the same as those produced by direct measurement on the vertebra and measurements made on the 3D rendered vertebra.
Keywords: 3D printing, orthopedic modeling, dimensional accuracy, vertebral modeling
1. PURPOSE The purpose of this work was to compare the dimensional accuracy of a 3D printed vertebra with direct measurements made on the dissected specimen used to produce the 3D model. Additionally, measurements were made on a 3D software rendered display of the vertebra for comparison to the physical measurements.
2. INTRODUCTION Additive manufacturing (AM) originated in the mid-1980’s with the invention of stereolithography by Charles W. Hull1. The original process used a liquid polymer that was hardened by UV light as the three dimensional object was constructed in layers. The build took place in a reservoir of the liquid on a platform that was lowered into the liquid as additional layers were added at the surface. Since then, many technologies for additive manufacturing have been developed. Examples of these technologies are those based on particulate material that may be sintered using a laser or bound with liquid binders; molten material deposition that is finding widespread application in ‘3D Printers’, which extrude melted material such as polylactic acid (PLA) or ABS filament from a heated extruder jet; and laminated material deposition processes, among others2. In recent years, AM technology has seen increasing use in tissue engineering which has as one goal the printing of living human organs for transplantation2-5. AM technologies are also finding increased use in medical application such as surgical planning, especially in the orthopedic surgery environment but also for ∗
[email protected];
phone 315-‐464-‐5083;
fax 315-‐464-‐8789
other organ systems6-9. A common feature of many of these applications is the use of medical imaging data such as CT and MRI scans for extraction of anatomical structures10. When deriving surgical planning 3D models from scan data, it may be critical that the resulting manufactured object have good geometric accuracy. An example would be the sizing of pedicle screws for attachment of spine hardware based on the dimensions of 3D printed vertebrae. There are many factors that affect the accuracy of the final object, and these may be divided into three categories: (1) Scan acquisition and reconstruction parameters, (2) 3D model extraction techniques, and (3) 3D printing accuracy. In this work, the effect of modifying parameters for 3D vertebral model extraction is examined. In particular, segmentation algorithms generally utilize a set of thresholds to determine the soft tissue - bone boundary. Upper thresholds are generally set to exceed the maximum Hounsfield unit (HU) value for bone and do not generally impact on the bone – soft tissue boundary. Lower thresholds discriminate bone from soft tissue and will have a direct impact on the location of these boundaries and the resulting dimensions of the printed object. It is the effect of changing the lower threshold that is examined here.
3. METHODS As part of a research project into the mechanical properties of a variety of pedicle screws used to attach orthopedic hardware to vertebral bodies, a cadaver spine was scanned on a General Electric Discovery 690 PET/CT scanner (GE Medical Systems, Waukesha WI) at 0.625 mm slice thickness and with pixels approximately 0.31 mm square. The spine was scanned at 120 kV tube potential, 300 effective mAs/rotation, and a pitch of 0.52. The L1 vertebra was then grossly dissected with a scalpel and ronguer to remove most soft tissues. The remaining soft tissue was macerated by soaking in an enzymatic detergent for 24hrs and then removed with a curette. The specimen was then soaked and kept in a blood analog solution to keep the calcified tissue intact and 'wet' for all measurements. The CT scan was downloaded to a local computer for processing and extraction of the surface model. The data was processed using Analyze 11.0 (AnalyzeDirect Inc., Overland Park, KS). The imported data was interpolated to cubic voxels approximately 0.31 mm3. The Image Edit tool in Analyze was used to segment the L1 vertebra from the volume, and most of the segmentation was performed using the Auto Trace function in the Image Edit module. The auto trace was initially configured to use a lower threshold value of 200 HU. Figure 1 below shows a segmented axial slice through L1.
Figure 1. Representative CT slice through L1 and the segmented bone.
After segmentation, the L1 object was filtered by a binary median filter using a 3x3x3 kernel in a ‘jack’ configuration. This was done to smooth the small jagged edges that were evident in some of the segmented images. The L1 object was then converted to a binary volume to use for surface extraction. Figure 2 shows the effect of the median filter on the boundary of the segmented structure.
Figure 2. Original boundary (top) and median filtered boundary (bottom) for a representative slice through the vertebral body. The median filter does not appreciably shift the location of the boundary in areas where the boundary is relatively smooth. The surface of the vertebra was extracted using the Surface Extraction module in Analyze. The Adaptive Deformation algorithm was used with the number of iterations set to 30, cube edge of 1 and Time Step of 0.01. Using the minimum cube edge size ensured that finer details would be captured by the surface extraction. The extracted surface mesh was exported as a stereolithography (.stl) file for printing by the 3D printer. The file was then printed on a Replicator® 2 printer (Makerbot, Brooklyn NY) using ‘standard’ quality settings. Total fabrication time was approximately 2 hours. The resulting printed vertebra is shown with the original cadaver vertebra in Figure 3. The segmentation, surface extraction, and printing process described above were repeated using segmentation lower thresholds of 175, 150, and 125 HU.
Figure 3. 3D printed L1 vertebra with original cadaver vertebra. Some bone tissue loss is seen on the articular process in the cadaver vertebra, this was due to experimental work with pedicle screws that fractured the bone there. This did not affect measurements made on the vertebra. Measurements were made by a single observer of 15 dimensions of both the cadaver and printed vertebrae. These dimensions are shown in table 1 with a description. Table 1. List of dimensions measured. Dimension
UEW
UED
LEW
LED
Description
Upper Endplate Width
Upper Endplate Depth
Lower Endplate Width
Lower Endplate Depth
Dimension
SCD
PDW Left PDH Left
Description
Spinal Canal Depth
Pedicle Pedicle Pedicle Diameter Diameter Diameter Width Width Left Height left Right
PDW Right
VBHp VBHp VBHa Left Right Veterbral Vertebral Vertebral Body Body Height Body Height posterior Height posterior Right anterior Left
PDH Right Pedicle Diameter Height Right
SCW Spinal Canal Width
VBa to SPp Vertebral Body Transverse anterior to Process Spinous Width Process posterior TPW
Each measurement was repeated three different times over the course of several days to minimize measurement bias due to learning effects. The CT scan was also used to generate a standard 3D rendering of the L1 vertebra using Aquarius Intuition® software (Terarecon, Foster City CA). The rendered image was rotated to approximate the lateral, cranial, and caudal views and measurements were made using the available ruler tools. Figure 4 below shows 3D
rendered views of the CT scan with typical measured values shown on the images. The measurements were repeated three times by a single observer.
Figure 4. 3D rendered display of L1 vertebra and representative measured values of several vertebra dimensions.
4. RESULTS Table 2 below shows the average and sample standard deviations for the three sets of measurements at a lower threshold HU value of 200. Table 2. Mean values of measured dimensions for the cadaver, rendered, and 3D printed vertebrae with lower threshold = 200 HU. Dimension (all in mm) UEW UED LEW LED VBHp Left VBHp Right VBHa SCW SCD PDW Left PDH Left PDW Right PDH Right TPW VBa to SPp
Cadaver 48.3 ± 0.12 31.5 ± 0.81 49.5 ± 0.06 31.0 ± 0.22 26.1 ± 0.27 26.0 ± 0.33 26.1 ± 0.02 24.3 ± 0.03 21.1 ± 0.13 9.1 ± 0.02 15.8 ± 0.12 10.7 ± 0.19 15.7 ± 0.04 77.1 ± 0.01 84.6 ± 0.12
3D Rendered 47.1 ± 0.38 31.1 ± 0.12 47.8 ± 0.61 30.7 ± 0.30 26.0 ± 0.31 25.3 ± 0.26 26.8 ± 0.40 24.5 ± 0.46 19.5 ± 0.26 9.1 ± 0.50 16.6 ± 0.44 10.5 ± 0.69 15.8 ± 0.15 76.8 ± 0.50 83.8 ± 0.68
3D Printed 48.1 ± 0.09 31.4 ± 0.15 49.5 ± 0.05 31.7 ± 0.35 27.1 ± 0.08 27.3 ± 0.42 27.2 ± 0.05 24.0 ± 0.01 19.8 ± 0.22 9.4 ± 0.15 17.2 ± 0.07 11.3 ± 0.09 17.0 ± 0.14 76.8 ± 0.16 84.4 ± 0.14
An ANOVA analysis was conducted on the measurements for each dimension to compare for significance in the measurement differences. The results for the ANOVA analysis are shown in Table 3. Table 3. Results of ANOVA testing for significance in the differences between pairs of measurements for the 3D model segmented using a threshold of 200 HU. ‘Yes’ indicates that the differences were significant at the 95% confidence level, ‘No’ that the differences were not significant. Dimension UEW UED LEW LED VBHp Left VBHp Right VBHa SCW SCD PDW Left PDH Left PDW Right PDH Right TPW VBa to SPp
Cadaver – 3D Rendered 3D Rendered - Printed Yes Yes No No Yes Yes No Yes No Yes No Yes Yes No No No Yes No No No Yes Yes No No No Yes No No No No
Cadaver - Printed No No No No Yes Yes Yes No Yes No Yes No Yes No No
The Cadaver-3D Rendered measurements were significantly different for 5 of 15 measurements (~33%), the 3D Rendered-Printed measurements were significantly different for 7 of 15 measurements (~47%), and the Cadaver-Printed measurements were significantly different for 6 of 15 measurements (40%). To determine the effect of changing the segmentation threshold, the results for the vertebrae dimension measurements were compared using a one-way ANOVA. The results are shown in Table 4.
Table 4. ANOVA results for vertebrae dimensions from models segmented with different lower threshold values. Dimension LED VBHp Left VBHp Right VBHa SCW SCD PDH Left PDW Right PDH Right UEW TPW PDW Left VBa to SPp UED LEW
Cadaver vs. 125 HU Model Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes No No No
Cadaver vs. 150 HU Model Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes No No No No No
Cadaver vs. 175 HU Model Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes No No No No No
Cadaver vs. 200 HU Model Yes Yes Yes Yes Yes Yes Yes Yes Yes No No No No No No
In Table 4 the results are sorted so that the total number of non-significant differences is easy to appreciate. It is clear from these results that as the lower threshold is decreased from 200 to 125 HU, the number of measurements that are significantly different from the cadaver vertebra values increases.
5. DISCUSSION In this work we examined the accuracy of dimensional measurements of a 3D printed vertebra vs direct measurements on the ex-vivo vertebra and measurements made on 3D renderings of the vertebra. We compared the resulting measurements using a one-way ANOVA to determine whether the differences were statistically significant. In general, we found that as the lower threshold for segmentation of the vertebra was decreased, the dimensions of the resulting printed vertebra increased and the number of measured dimensions that differed significantly from the cadaver vertebra increased. At a lower threshold of 200 HU, the measurements were similar to the 3D rendered vertebra measurements, with the 3D rendering having higher accuracy. It seems likely that increasing the lower threshold for vertebra segmentation above 200 HU would increase the number of measured dimensions that agree with the cadaver vertebra measurements. However, the semiautomated segmentation begins to leave significant voids in the segmentation where the HU values of the bone drop below 200 HU. This results in a significant effort being required to manually repair and finish the segmentation.
3D printed anatomical structures have the potential to provide powerful tools for analysis, surgical planning, and selection of appropriate hardware in orthopedic surgical applications. Care must be taken to ensure that the parts produced by 3D printing exhibit the accuracy necessary for these applications. In this work, we have shown that it is possible to produce 3D printed anatomy with an accuracy that is comparable to measurements made in a 3D rendered version of the anatomy, and that is equivalent to the original anatomy within the limits of measurement accuracy for many of the dimensions measured
6. REFERENCES [1] Hull, C. W. "Apparatus for production of three-dimensional objects by stereolithography." U.S.
Patent No. 4,575,330. 11 Mar. 1986. [2] Gibson, I., Rosen, D.W., and Stucker, B. [Additive manufacturing technologies: rapid
prototyping to direct digital manufacturing]. Springer, (2010). [3] Leukers, B., et al. "Hydroxyapatite scaffolds for bone tissue engineering made by 3D printing."
Journal of Materials Science: Materials in Medicine 16(12), 1121-1124 (2005). [4] Fedorovich, N. E., et al. "Hydrogels as extracellular matrices for skeletal tissue engineering: state-of-the-art and novel application in organ printing." Tissue engineering 13(8), 1905-1925 (2007). [5] Mironov, V., et al. "Organ printing: computer-aided jet-based 3D tissue engineering." TRENDS
in Biotechnology 21(4), 157-161(2003). [6] McGurk, M., Amis, A., Potamianos, P., Goodger, N. “Rapid prototyping techniques for
anatomical modelling in medicine.” Ann R Coll Surg Engl 79, 169–174 (1997). [7] Giannatsis, J., and Dedoussis, V. "Additive fabrication technologies applied to medicine and health care: a review." The International Journal of Advanced Manufacturing Technology 40(1-2), 116-127 (2009). [8] Zein, N. N., et al. "Three-‐dimensional print of a liver for preoperative planning in living donor liver transplantation." Liver Transplantation 19(12), 1304-1310 (2013). [9] Giacomo, G. AP Di, et al. "Clinical application of stereolithographic surgical guides for implant placement: preliminary results." Journal of periodontology 76(4), 503-507 (2005). [10] Rengier, F., et al. "3D printing based on imaging data: review of medical applications." International journal of computer assisted radiology and surgery 5(4), 335-341, (2010).
