Data and software
CT data were obtained from a patient scanned at the Radiology Department of the Hospital of Lithuanian University of Health Sciences. The patient was scanned with a Toshiba Aquilion One multi-slice spiral tomograph with preset protocol (0.25 mm step, 0.5 mm slice thickness, gantry angle, 0°.). Axial slices were exported using FC30 convolution kernel. The data were anonymized using DICOM Anonymizer by Sha He (2008) . The computer used was a PC with an Intel i5–3570 3.40 GHz CPU, 16 GB RAM, NVIDIA GeForce GT640 1 GB RAM video card, standard optical mouse, and running 64-bit Windows 10 operating system. The programs used for digital work are listed in Table 1.
CT data-set review
First the quality of the CT scan data was assessed. This consisted of checking whether the ROI (region of interest) was as required, if number and thickness of the slices were sufficient, and if there were any scanning artifacts of various nature.
The DICOM file header also had to be checked to determine if the parameters used during scanning were correct. The parameters include the protocol used, gantry angle, slice thickness, slice step, and convolution kernel. If the gantry angle is set greater than 0.0° and the reconstructing software is unable to make the necessary corrections, then shear distortions of the model may occur . The start and the end of a CT scan images series often contains defective slices that have to be discarded on dataset review (Fig. 1a).
If the data obtained from the CT contains errors (e.g., metal-induced scatter or movement artifacts), it has to be determined if these errors affect the anatomical region required for future modeling (Figs. 1b, 2a, b).
If so, then a decision has to be made whether to reacquire the CT scan or to take a bone impression instead. If CT artifacts do not affect the area of the proposed implant, then one can proceed to the segmentation step. The image set also needs to be assessed for motion artifacts, which can only be seen in the reconstructed sagittal projection or 3D volume rendering.
CT data import and virtual model production
Slicer 3D software was used to create the 3D model of the jaw. Bone tissue segmentation was performed, as explained in the online Slicer 3D tutorial . The most important step is correct choice of level thresholding for the tissue of interest. The specific threshold at which the program will build a surface is influenced by many factors: CT scanner and its software settings (convolution kernel), CT volume cropping, bone density of the actual patient, and most importantly, operator selection. Human decision for threshold selection is very important and is the sum of anatomy knowledge and work experience, which has no substitute.
Following selection of the desired threshold for volume label (or mask) (Fig. 3),
the surface model of the selected anatomy is created (Fig. 4a).
This is usually a polygon mesh that can be saved as a number of digital formats, such as .stl, .obj, .ply, and .vtk. In this case, .stl (STL) was used because of compatibility with all currently used programs. The STL file format describes 3D objects as meshes made of stitched triangles. Segmentation programs of all kinds, including commercial software, are prone to errors while constructing mesh models of anatomical structures. Created mesh (i.e., virtual model of the jaw) often contains non-manifold edges, holes, inverted triangles, and other errors that hinder further processing steps. Human bone with its trabecular and cavernous structures is difficult for computers to convert to manifold (closed surface) mesh structures, which is required by modeling software. This is very evident when a program tries to recreate the trabecular bone structure (Figs. 4f and g). The situation mandates the use of specialized error correcting programs such as Netfabb. Even then, the program only corrects software errors introduced by segmentation software, but not holes in the bone anatomical structure. Fenestrations in the virtual bone model may be part of the actual anatomy of the patient; they also may be artifacts of thresholding. Anatomical holes of models are to be corrected manually (holes closed) or avoided during modeling (Figs. 4g and h).
The number of abutments has to be minimized to make fewer perforations of mucosa. Implant abutment in the oral environment is a very special place—it is a direct gateway for organisms from the internal environment to the external environment (oral cavity). This perforation of mucosa is the place where the organism lacks its natural barrier against microorganisms present in the oral cavity. None of the current dental implants form a firm implant-epithelial junction that can reliably prevent microorganism invasion into the bone. It has been shown than bacterial invasion leads to loosening of the implant-bone bond in the long term. This condition is even more prominent with subperiosteal dental implants of which only 30–40% of the whole surface make contact with the bone and the rest is surrounded by soft tissues. Thus, chances of bacterial invasion should be minimized by reducing the number of mucosa perforations. One must avoid creating narrow spaces, crevices, and concavities, especially near abutments. Pits and narrow spaces are harbors for possible oral fluid microorganism contamination and proliferation. In addition, crevices impede tissue liquid circulation and thus hinder regenerative processes. It is important to respect blood supply and not hinder periosteum to bone contact, such as with large metal areas. Overall, it is recommended to use as little metal as possible to achieve masticatory force distribution goals.
One should plan dental function and aesthetic in advance, before designing the actual implant. First, virtual models have to be put into centric relation by means of existing dentition intercuspation or using x-ray markers in the removable dentures that are being used as scanning aides (Figs. 4b and c). Home work also includes examining diagnostic models and wax-ups that aid in planning abutments emergence profiles and their relation to planed prostheses. Using virtual teeth in the software during the process of implant modeling also aids proper placement of abutments (Figs. 5b and c).
In addition, consider the lateral component of jaw movements—create anatomical and mechanical means of implant fixation. When creating fixed cementable abutments, pay attention to parallelism of prospective abutments with current partnering abutments (teeth or implants) and the relation to opposing dentition (antagonists).
The modeling process of a custom dental implant per se is the de novo creation process. This is in contrast to reconstructive surgery modeling, which often uses contra-lateral mirroring to obtain correct anatomical structures. Thus, new objects of intricate shape must be created, which in turn must closely adapt to current anatomical structures. This task requires suitable digital tools. Modeling for medical applications is still a new and emerging field of digital design. There is a range of software built for the dental prosthetic field and numerous programs for implant planning and surgical guide creation. Although some of these programs offer planing osteotomies and bone grafting, the majority are predominantly limited to use of standard screw type implants from different manufacturers. Consequently, they are not useful for modeling custom structures. Numerous software packages have been created for making fixed and removable dental prostheses via CAD/CAM. It appeared that the most straightforward route would be to use applications designed for modeling a partial removable dentures because of the similarity of subperiosteal implants and the framework of partial removable denture. However, after testing several software packages, this route proved to be unusable because the programs do not allow jaw model modification, are limited to preset structures, or require heavy customization to be useful.
3D modeling tools
Depending on software capabilities and designer approach, subperiosteal dental implants can be created using several different tools (techniques). The applicable techniques and corresponding programs are listed below:
° Digital clay modeling.
▪ Sculpt clay with brush tool (Meshmixer, Blender, FreeForm).
▪ Sculpt clay applying profile along surface curve (VRMesh, FreeForm, Sculpt).
° Selected surface extrusion (Meshmixer, VRMesh).
° Curve fitting with pipe or tube (VRMesh, Freeform, Sculpt).
In this case, modeling was conducted via the addition-subtraction method using the mesh brush sculpting technique. Two copies of a jaw model were used. The the first copy was used to model the implant itself by bulging virtual clay, which is similar to the manner in which a dental technician lays wax on the refractory model for partial production of a removable framework. The second copy was used for subtraction from first—to form the inner (bone facing) surface of the implant.
Models for the maxilla and mandible should be put as close as possible to centric relation, which makes abutment planning easier (Figs. 4b and c). Placing virtual teeth helps with the planning of future occlusion and abutments (Figs. 5b and c).
Firstly, abutments are placed according to the current bone anatomy and future prosthetic plan. Bone quality and anatomic form must be considered while designing the implant. Load bearing and retentive elements should be planned only where at least 0.8 mm of cortical bone is present. Retentive elements (loops, puds) were placed at specified locations and all above-mentioned elements were merged with one copy of the jaw models using a Boolean addition operation (Figs. 6b and c, Figs. 7a and d).
During the third step, interconnecting structures (straps) are modeled into saddle-like web connections using various brushes of the “SCULPT” tool in Meshmixer (Figs. 7b and c). Abutments, retention loops, and peddles are connected by straps. The overall structure of the subperiosteal implant is usually designed to be 1.5–1.8 mm thick, which is an empirically defined value. During modeling, the implant is shaped slightly thicker by 0.2–0.3 mm to allow for later reduction and smoothing of its surface. Reduction is necessary because of the nature of the DMLS process, which results in printed parts with rough and oxidized surfaces. Supports also leave rough surface once removed; thus, extra thickness of metal must be included to enable reduction of the material.
The Boolean subtraction operation is an essential part of custom implant additive modeling that results in a final product. The input for this computation consists of two models. The first model is an anatomical model of the bone and the second is the same anatomical model, but with modeled implant structure bulging from it. When the computer calculates the difference between these models - the resultant structure is the implant (Figs. 6, 7).
Implant contains small residual artifacts that have the appearance of small snowflake-like structures left behind after the Boolean operation. These small objects appear because of inconsistencies accumulated in virtual models during earlier manipulations. Such structures have to be removed manually and errors corrected using the above-mentioned software (Figs. 7e, 8a and b).
Model mesh repair is a very important part of the modeling process, from the start to the end. Initially, it is necessary to repair the virtual model of the jawbone before any modifications to the virtual models (before modeling) is performed. Further, it is necessary to repair models before and after each subtraction or addition operation when needed, because software cannot commence Boolean operations on meshes containing errors. Finally, designed implant models also need to be checked for errors and repaired if needed, before feeding the mesh file to the machine code generation program. There are many types of mesh errors that can occur in complex models of human anatomy—bones in particular. These include, but are not limited to, flipped triangles, holes, non-manifold edges, duplicate faces, and duplicate vertices. Usually, mesh repair is a multi-step process that corrects various types of errors.
There are various freely available programs available for mesh error repair. They include the following:
▪ Netfabb Basic (comprehensive list of errors repaired, automated scripts).
▪ MeshLab (no automated scripts, need manual selection for each type of errors to be corrected).
▪ 3Data expert lite (limited repair capabilities in free version).
▪ GOM Inspect V8 SR1 (some error repair capability).
▪ Meshmixer (automated error repair, several repair options, not sufficient even for program internal needs).
Netfabb contains pre-compiled repair scripts for correcting errors of different levels of complexity. This option facilitates user operations with mesh repair. Netfabb Private is delivers three pre-compiled repair scripts for mesh repair. These three scripts are of increasing comprehensiveness for error detection and correction(Table 2).
The “default repair” script of the Netfabb program was used at all steps where it was necessary (Figs. 8d and e).
On completion of modeling, the implant design and jaw model STL files (Figs. 9a, b, c) were sent to a printing facility (Orthobaltic UAB, Lithuania).
The jawbone model, created earlier, was printed in polyamide using an SLS printer ESO P 396, in 60 mkm layers. The implant was produced using the EOS printer EOSINT M 280 by DMLS process in Ti64 alloy (100mkm layers), which is known to be biocompatible and exhibits the required mechanical properties. The implant was printed and then annealed in an argon environment. Supports were removed, and the implant was finished by a technician. The inner and outer surfaces of the implant were checked for voids or pimples. These were removed when encountered by abrasive instruments. Implant fit was checked against a plastic jaw model. In addition, implant outer surface was shaped to the desired profile with tungsten carbide cutters and blasted with aluminum oxide 150 mkm at 5 bar (Figs. 9d, e, f). Complete preparation of the implant for surgery includes more steps, which are metal surface treatments (polishing, blasting, etching, etc.), packaging and sterilization, but those are not in the scope of this article.