Abstract
Purpose
Generate and describe the process of creating a 3D printed, rapid prototype temporal bone model from clinical quality CT images.
Materials and methods
We describe a technique to create an accurate, alterable, and reproducible rapid prototype temporal bone model using freely available software to segment clinical CT data and generate three different 3D models composed of ABS plastic. Each model was evaluated based on the appearance and size of anatomical structures and response to surgical drilling.
Results
Mastoid air cells had retained scaffolding material in the initial versions. This required modifying the model to allow drainage of the scaffolding material. External auditory canal dimensions were similar to those measured from the clinical data. Malleus, incus, oval window, round window, promontory, horizontal semicircular canal, and mastoid segment of the facial nerve canal were identified in all models. The stapes was only partially formed in two models and absent in the third. Qualitative feel of the ABS plastic was softer than bone. The pate produced by drilling was similar to bone dust when appropriate irrigation was used.
Conclusion
We present a rapid prototype temporal bone model made based on clinical CT data using 3D printing technology. The model can be made quickly and inexpensively enough to have potential applications for educational training.
1
Introduction
Mastoidectomy is a common procedure used for cochlear implantation and multiple otologic diseases. It requires comprehensive anatomical knowledge of the temporal bone and surgical drill dexterity in order to avoid damaging delicate adjacent structures like the facial nerve, internal carotid artery, cochlea, and sigmoid sinus. Otolaryngologists must master complicated surgical procedures involving the anatomy of these structures, which can only be achieved through extensive practice of surgical technique. Mistakes in otolaryngology can be severely detrimental to the patient, making every opportunity to practice and perfect one’s skill highly beneficial.
The current approach to surgical training of residents involves practicing on cadaveric specimen. Residents are given access to a limited number of cadavers to dissect in temporal bone laboratories. Although cadaveric specimens are highly valuable for teaching and training, they are highly regulated and assert several obstacles. The availability of cadavers is presently dependent on voluntary donation or the attainment of unclaimed bodies . The limited availability and high demand for cadavers thus create an ethical dilemma further complicated by cultural, religious, legal, and socioeconomic factors. However, even if the supply of cadavers were not a concern, there would still be the issue of reproducibility. There is considerable variance in the population and pathology of the temporal bone making repetition on a standard model problematic. Furthermore, cadavers have the inherent risk of infectious disease transmission and costs associated with the appropriate disposal of biologic waste.
Advances in technology have allowed for alternative techniques to supplement cadaveric dissection for surgical training. Surgical simulation using CT and MRI data and computer-aided models has existed in many institutions for the past few years, creating a virtual operating room . Rapid prototyping (RP) through 3D printing has been the latest technological advance to serve as an adjunct to cadaver-based instruction. 3D printing is rapidly becoming popular in the field of medicine due to its accuracy, affordability, and speed of production . The applications of 3D printing in the medical field have included the production of presurgical models for planning, patient-specific prostheses, and models for surgical simulation and education . There have been several publications examining the potential role of 3D printing the temporal bone for the purposes of medical education of the Otolaryngology student, however these studies used high resolution CT and/or MRI scans from cadaveric specimens for model generation . These models vary in their complexity and preciseness. Preliminary data show that these models may have a viable role in surgical training; however there are still some unresolved and unexplored issues. There have been few studies examining the generation of a rapid prototype model using clinical CT scans. In this study we seek to address whether a truly expeditious rapid prototype temporal bone model can be created through the use of clinical CT data. We will discuss the potential applications as well as the limitations of this model.
2
Materials and methods
2.1
Manufacture of prototypes
Imaging data from CT scans performed for clinical purposes were used for this study. The resolutions of the scans were 0.227 mm × 0.227 mm × 0.5 mm or better. The right (Model A) and left (Model B) temporal bones of two different patients were selected based on being well pneumatized with no clinical disease in the selected ear. The CT data of a third patient (Model C) were those from a patient with a hyperpneumatized mastoid with a dehiscent superior semicircular canal. The freely available software program, ITK-SNAP , was used to convert the CT data into a 3D model. Using the program’s automated segmentation feature, the CT scans were selectively segmented using intensity thresholds to isolate bone. This was done by selecting a starting voxel in a region corresponding to temporal bone and setting the intensity parameters with a lower and upper threshold. Each threshold range was made specific to the CT scan because bone density varied from patient to patient. We then reviewed the automatic segmentations and an initial manual segmentation was performed to minimize artifacts and include any desired areas of bone not included by the automated segmentation.
Post-processing of the model consisted of 3D surface mesh formation and further manual segmentation. Using the surface mesh feature, the automated segmented initial versions were converted into a 3D surface mesh and smoothed using a Gaussian filter with a standard deviation of 0.8 mm and max approximation error of 0.03 mm. We found that these parameters allowed for the optimal balance to produce a smooth surface while maintaining satisfactory resolution. Manual segmentation was specifically required for the following: ossicles, mastoid air cells, and drain hole insertion. We found that the more intricate and delicate structures of the mastoid air cells and ossicles did not appear of high enough quality on CT for automatic segmentation and required human judgment ( Fig. 1 .) Manual segmentation consisted of including bony structures that were not initially segmented and de-segmenting soft-tissue and undesired non-bone structures. This was the most time intensive step in the formation of the rapid prototype, requiring inspection of each axial slice and many reiterations of the 3D surface mesh. Additional post processing modifications included the formation of an intentional drain hole. Drain holes were made by deliberately de-segmenting voxels forming a 3 mm hole in the region of the sigmoid sinus to allow for sufficient drainage of the resin scaffolding encased in the air cells of the mastoid.
The final virtual models were converted into STL files and subsequently printed by the Dimensions SST 1200es Printer. This printer used Fused Deposition Modeling Technology which is a type of Fused Filament Fabrication that prints out material by laying down consecutive layers of ABS thermoplastic with a layer thickness 0.254 mm. For free hanging structures, the printer deposited a removable support material made of resin to act as scaffolding. Once the rapid prototype model was complete, it was placed in a heated detergent wash for 8-12 h to remove the dissolvable resin scaffolding.
2.2
Evaluation of prototypes
Once each rapid prototype model was complete, we evaluated each model on several parameters. For each model, the presence and quality of the following structures were assessed: malleus, incus, stapes, oval window, round window, promontory, horizontal semicircular canal, and mastoid segment of the facial nerve canal. The diameter of the bony external auditory canal at the most lateral edge was measured with a micrometer and compared against CT data. An experienced surgeon performed mastoidectomy on the prototype to visualize the above anatomic landmarks and qualitatively evaluate the efficacy of the model for dissection and response to surgical drilling.
2
Materials and methods
2.1
Manufacture of prototypes
Imaging data from CT scans performed for clinical purposes were used for this study. The resolutions of the scans were 0.227 mm × 0.227 mm × 0.5 mm or better. The right (Model A) and left (Model B) temporal bones of two different patients were selected based on being well pneumatized with no clinical disease in the selected ear. The CT data of a third patient (Model C) were those from a patient with a hyperpneumatized mastoid with a dehiscent superior semicircular canal. The freely available software program, ITK-SNAP , was used to convert the CT data into a 3D model. Using the program’s automated segmentation feature, the CT scans were selectively segmented using intensity thresholds to isolate bone. This was done by selecting a starting voxel in a region corresponding to temporal bone and setting the intensity parameters with a lower and upper threshold. Each threshold range was made specific to the CT scan because bone density varied from patient to patient. We then reviewed the automatic segmentations and an initial manual segmentation was performed to minimize artifacts and include any desired areas of bone not included by the automated segmentation.
Post-processing of the model consisted of 3D surface mesh formation and further manual segmentation. Using the surface mesh feature, the automated segmented initial versions were converted into a 3D surface mesh and smoothed using a Gaussian filter with a standard deviation of 0.8 mm and max approximation error of 0.03 mm. We found that these parameters allowed for the optimal balance to produce a smooth surface while maintaining satisfactory resolution. Manual segmentation was specifically required for the following: ossicles, mastoid air cells, and drain hole insertion. We found that the more intricate and delicate structures of the mastoid air cells and ossicles did not appear of high enough quality on CT for automatic segmentation and required human judgment ( Fig. 1 .) Manual segmentation consisted of including bony structures that were not initially segmented and de-segmenting soft-tissue and undesired non-bone structures. This was the most time intensive step in the formation of the rapid prototype, requiring inspection of each axial slice and many reiterations of the 3D surface mesh. Additional post processing modifications included the formation of an intentional drain hole. Drain holes were made by deliberately de-segmenting voxels forming a 3 mm hole in the region of the sigmoid sinus to allow for sufficient drainage of the resin scaffolding encased in the air cells of the mastoid.
