The Accuracy of a Method for Printing Three-Dimensional Spinal Models

Background To study the morphology of the human spine and new spinal fixation methods, scientists require cadaveric specimens, which are dependent on donation. However, in most countries, the number of people willing to donate their body is low. A 3D printed model could be an alternative method for morphology research, but the accuracy of the morphology of a 3D printed model has not been determined. Methods Forty-five computed tomography (CT) scans of cervical, thoracic and lumbar spines were obtained, and 44 parameters of the cervical spine, 120 parameters of the thoracic spine, and 50 parameters of the lumbar spine were measured. The CT scan data in DICOM format were imported into Mimics software v10.01 for 3D reconstruction, and the data were saved in .STL format and imported to Cura software. After a 3D digital model was formed, it was saved in Gcode format and exported to a 3D printer for printing. After the 3D printed models were obtained, the above-referenced parameters were measured again. Results Paired t-tests were used to determine the significance, set to P<0.05, of all parameter data from the radiographic images and 3D printed models. Furthermore, 88.6% of all parameters of the cervical spine, 90% of all parameters of the thoracic spine, and 94% of all parameters of the lumbar spine had Intraclass Correlation Coefficient (ICC) values >0.800. The other ICC values were <0.800 and >0.600; none were <0.600. Conclusion In this study, we provide a protocol for printing accurate 3D spinal models for surgeons and researchers. The resulting 3D printed model is inexpensive and easily obtained for spinal fixation research.


Introduction
The study of anatomy is an important part of medical education. Furthermore, with a sound knowledge of anatomy, we can design new surgical fixation techniques. A good example of spinal surgery is the pedicle screw fixation technique, which was rarely used before 1907 [1]. However, as more anatomic features of this technique were described, spine surgeons began to attempt the technique, and it has since become widely used in thoracic, lumbar, and sacral regions [2,3].
To study the morphology of the human spine and new spinal fixation methods, scientists require cadaveric specimens, which are dependent on donation. However, in most countries, the number of people willing to donate their body is low [4]. Indeed, some religions and faiths discourage people from donating their body, and as a result, access to cadaveric specimens is very limited in some countries.
Without sufficient numbers of cadaveric specimens, some researchers learn new techniques only by studying radiographic images [5][6][7][8]. Computed tomographic analysis may be sufficient for morphometric study, but the use of cadaveric specimens is preferable and more reliable [9][10][11]. Moreover, cadaveric research is necessary for some studies.
3D printing is a process for making a 3D-printed model of almost any shape from a 3D digital model or other electronic data source [12]. The 3D printing methods include selective laser melting, laser sintering, fused deposition modeling, stereolithography, laminated object manufacturing and fused filament fabrication [13,14]. Many materials are used in these different 3D printing techniques; e.g. thermoplastics (PLA and ABS) are commonly used for the fuse filament fabrication technique, where as titanium alloys and cobalt chrome alloys are used for the selective laser melting technique. Researchers can choose different materials according to the 3D printing technique to be used and the properties, cost and color of materials that they prefer [12,15,16].
We previously reconstructed a 3D digital spinal model from CT scan data, and it has been shown that this 3D digital model is morphologically accurate [17]. Accordingly, an accurate 3D model printed from a 3D-reconstructed digital model would allow for studying the morphological features of the 3D printed model, which may solve the problem of scarce cadaveric specimens.

Materials and Methods
This study was performed following the Declaration of Helsinki principles and was approved by the Institutional Review Board (IRB) of The Second Affiliated Hospital of Wenzhou Medical University. Written informed consent was obtained from all participants.
Forty-five computed tomography (CT) scans of cervical, thoracic or lumbar spines of patients (mean age 42.5±7.7 years (range 31-54 years))were obtained using the Star PACS system (INFINITT, Seoul, South Korea) of our hospital. The included cervical, thoracic and lumbar spines lacked spinal disease, as shown by CT scans for health examination or because the patients had presented with oral, anterior neck, cardiac, pulmonary or abdominal diseases. Patients with any spinal abnormality, such as fracture, scoliosis or tumor, were excluded.
We measured the following parameters using the Star PACS System, which was proven to achieve accurate measurements in previous studies [8,18]: The CT scan data were then imported in DICOM format into Mimics software v10.01 (Materialise, Leuven, Belgium)for 3D reconstruction. The threshold value was set at "Bone (CT)",  "226-Max", which is optimal for bone reconstruction. After the 3D digital images were calculated and reconstructed, we removed the bone, which was not needed, and every vertebra was then separated. The data were then saved in STL format and imported into Cura software. After a 3D digital model was formed, we saved it in Gcode format and exported it to a3D printer (3D ORTHO Waston Med Inc. Changzhou, Jiangsu, China) to print the objects. The scale was set at 1:1, and PLA (Polylactic acid:(C 3 H 4 O 2 ) n ), with a molecular weight of 5000-700000 according to the product instruction, was used as the print material. After the 3D-printed models were obtained, the above-mentioned parameters were measured again.

Statistical analysis
The data were analyzed using the SPSS software (version 17.0, SPSS Inc., Chicago, IL, USA). Comparisons of the radiographic image and 3D-printed model data were made using paired ttests, with the level of significance set at P<0.05. If P>0.05, the Intraclass Correlation Coefficient (ICC) was calculated to assess how strongly the data from radiographic images and 3D printed models resembled each other.

Results
Forty-four parameters of the cervical spine, 120 parameters of the thoracic spine, and 50 parameters of the lumbar spine were measured.
For  16.91±1.60, and 17.24±1.97 mm, respectively). We found that the 3D printed model data also showed similar trends (Table 4 and Table 5).
Furthermore, 88.6% of all parameters of the cervical spine (Tables 1, 2 and 3), 90% of all parameters of the thoracic spine (Table 4), and 94%ofall parameters of the lumbar spine (Table 5) were >0.800. The other ICC values were <0.800 and >0.600, and none were <0.600.

Discussion
Surgeons are constantly exploring novel internal or external fixation techniques to improve healthcare for patients. When they identify "new ideas", surgeons should ideally test their feasibility using cadaveric models. Unfortunately, the rate of body donation for use in research is very low [4]. Furthermore, in most developing countries, particularly those with strong religious beliefs or without higher education, the rate of donation is lower than in developed countries [19,20].
In China, universities and medical schools have been faced with an ongoing shortage of cadavers for education and research because of aspects of the Chinese culture [21];the donation rate has also been found to be low in Greece [20]. As a result, for most surgeons, it is difficult to obtain sufficient cadaveric specimens on which to test their "new ideas". In addition, most cadaveric specimens are stored in formalin, a storage technique that will change the shape of the bone if the specimens are stored for long periods. This is unacceptable for experiments that require accurate data.
To allow surgeons to test the feasibility of newly developed fixation techniques, we must provide accurate spine bone models. With the development of 3D digital reconstruction, it is now possible to test new fixation techniques on 3D digital images. Puchwein et al [17]studied the morphometry of the odontoid peg and its impact on ventral screws (one screw or two screws) using3D digital images; similar methods could resolve the problem caused by a lack of available cadavers. Unfortunately, digital images sometimes do not provide sufficiently accurate data. We previously imitated trans-pedicle, trans-disc oblique lumbar interbody fixation using 3D digital images. For L1/2, L2/3 and L3/4 screws, the data from 3D digital images and cadavers were similar, but the data were different for L4/5 and L5/S1 screws because the screw angles were blocked by iliac bone and by part of the L5 inferior articular process [22].  3D printing techniques can use 3D digital images to print 3Dmodels, creating the possibility of printing a morphologically accurate 3Dmodel. In our study, 214 parameters from C1 to L5were measured. The method for measuring data using radiographic images was a little different from that using the 3D-printed model because the largest width values of the vertebral body and pedicle were not in the same image; therefore, systematic error could not be avoided. To minimize this error as much as possible, we chose the best matched images for measurements, and forty-five spines were included to minimize individual error. The results showed no significant difference between the data from radiographic images and from 3D-printed models. Our results showed that 88.6% of the ICC values for the parameters of the cervical spine, 90% of the ICC values for the parameters of the thoracic spine, and 94%of the ICC values for the parameters of the lumbar spine were >0.800. These results prove the strong resemblance between data from radiographic images and 3D-printed models. The mean age of the patients from which the 45 CT scans were takenwas42.5±7.7 years (range from 31 to 54 years old), and all of the patients were adults with normal spinal structure. To decrease the error as much as possible, patients with spinal diseases were excluded. The printed material (PLA) is not very expensive, costing approximately$150 per kilogram ($0.15 per gram). One atlas or axis model is approximately 5-10 grams, whereas other cervical through lumbar vertebrae models range from10 to 35 grams. This cost will be reduced as the method becomes more widely used.

Limitations of this study
The 3D machine used in this study cannot print a model larger than 15 cm Ã 15 cm Ã 25 cm. However, if the 3D model is printed at a scale lower than 1:1, the systematic error would be increased; therefore, we chose to print each vertebra separately. Because of this, we could not measure the some parameters between two segments, such as foramen height, in this study.  Although we can provide an accurate printed spinal model using our protocol, this is simply a bony model, without any soft tissues, nerves or vessels. If the screw perforates the cortical bone and extends outside the bone into soft tissue, it will be defined as failure; therefore, the current 3D-printed model is still not suitable for some techniques that are needed to study the relationship between a screw and soft tissue. However, most spinal fixation techniques, including pedicle screw fixation, odontoid screw fixation, atlantoaxial transarticular screw fixation, lateral screw fixation, and pedicle rib screw fixation, all of which are known to be safe if the screw does not perforate more than 2 mm outside the cortex, could be studied using3Dprinted models.
In the future, if we are able to use different materials to print discs, facet joints and ligaments, it may become possible to conduct biomechanical studies directly on 3Dprinted models.

Conclusion
In this study, we provide a protocol for printing accurate 3D spinal models that can be used by surgeons and researchers. This 3D-printed model is inexpensive and can easily be obtained for spinal fixation research.