If you don't remember your password, you can reset it by entering your email address and clicking the Reset Password button. You will then receive an email that contains a secure link for resetting your password
If the address matches a valid account an email will be sent to __email__ with instructions for resetting your password
Transparent facial orthoses (TFOs) are commonly used for the treatment of craniomaxillofacial trauma and burns to prevent hypertrophic and keloid scarring. A TFO is typically customized to the patient’s facial contours and relies on a precise fit to ensure good rehabilitative performance. A smart method of TFO design and manufacture is needed which does not require an experienced prosthetist, allowing for rapidly produced, well-fitting TFOs. Whether the rapid application reduces the final level of patient scarring is unclear.
The purpose of this clinical study was to determine whether a scalable, automated design-through-manufacture pipeline for patient specific TFO fabrication would be successful.
Material and methods
The automated pipeline received a 3-dimensional (3D) facial scan captured from a depth sensitive mobile phone camera. The scan was cleaned, aligned, and fit to a template mesh, with a known connectivity. The resultant fitted scan was passed into an automated design pipeline, outputting a 3D printable model of a custom TFO. The TFOs were fabricated with 3D printing and were both physically and digitally evaluated to test the fidelity of a digital fit testing system.
A total of 10 individuals were scanned with 5 different scanning technologies (STs). All scans were passed through an automated fitting pipeline and categorized into 2 groups. Each ST was digitally fitted to a ground truth scan. In this manner, a Euclidean distance map was built to the actual facial geometry for each scan. Heatmaps of 3D Euclidean distances were made for all participant faces.
The ability to automatically design and manufacture a custom fitted TFO using commercially available 3D scanning and 3D printing technology was successfully demonstrated. After considering equipment size and operational personnel requirements, vat polymerization (VP) technology was found to be the most promising route to TFO manufacture.
Previous TFO fabrication methods used 3D printing to produce vacuum formed molds or required clinicians for adjustments and application. A fully automated digital solution needing no specialist clinician preapplication was developed and tested, and the findings are presented for the first half of this pipeline: data capture to TFO design. From this, more extensive clinical trials with this system can be performed, potentially providing a new, rapid, automated method of designing and manufacturing TFOs.
Transparent facial orthoses (TFOs) are used to treat craniomaxillofacial trauma and burns, preventing hypertrophic and keloid scarring and ensuring the maintenance of face symmetry and shape during recovery.
Less than 2000 Pa will not reach the capillary layer and therefore not effect collagen production. Between 2000 and 3333 Pa has been shown to reduce collagen production and has a negligible risk. Above 3333 Pa can increase the effect and is claimed to give more rapid results, but higher pressure can also cause harm, such as blistering, abnormal bone growth, and necrosis.
To account for this, a topical silicone elastomer sheet or gel can be included between the TFO and facial surface, but the inclusion of a silicone layer reduces TFO transparency, compromising a clinician’s ability to visually observe fit. However, recent studies have found that this can still be determined by using computational models.
used 3D facial scans and computer numerical control (CNC) milling machines to manufacture the vacuum forming molds. More recently, additive manufacture (AM) has been used to directly print a TFO, eliminating the multistep fabrication process of casting and vacuum forming.
Coupled with 3D scanning, this provided a noninvasive, streamlined, manufacturing process well suited for custom applications. Despite this, a major barrier to adoption was the recursive labor-intensive design process to create the essential 3D computer-aided design (CAD) models required for each patient. However, by employing parametric design processes, automated design mass customization (MC) pipelines can be produced for patient-specific products.
Establishing an automated design process for custom-fit TFOs which is accurate (achieves good fit) and robust (serves a wide demographic) would reduce the time from problem identification to intervention, enable remote manufacture, remove logistical barriers, and create significant cost savings.
This study aimed to demonstrate a scalable MC pipeline for patient-specific TFO fabrication. The study hypothesis was that from raw 3D facial data captured from a mobile phone, an autonomous data analytics pipeline with an integrated CAD application programming interface (API) could generate custom-fitting TFOs more rapidly than current design and fabrication methods.
Material and METHODS
A total of 10 volunteers of varying ages, sex, and ethnicity (summarized in Table 1) were recruited at Imperial College London (UK) and Kings College London (UK) in March 2021 following a protocol (19IC5167) approved by the Imperial College Research Ethics Committee. A 4-step process (Fig. 1) was used to convert the data of each volunteer into a custom-fit TFO. In Step 1 (data acquisition), 5 different 3D scanners were tested to collect the face geometries. All scans were collected on the same day and under the same environmental conditions (setting). Participants were asked to sit in an upright position, looking forward with a neutral expression. The scanning technologies (STs) included 2 iPhone apps, 1 handheld structured light scanner, and 2 stereophotogrammetry systems. Table 2 summarizes the details of equipment, software program, and output file. Figure 2 illustrates the 3D meshes obtained by each method on 1 participant. Collected 3D scans were submitted into the pipeline.
The fitting process (Fig. 3) comprised 4 stages of fitting the template mesh to the face data: cleaning (removal of extraneous data such as the background, hair, and clothing), course rigid alignment, iterative closest point alignment, and noniterative closest point alignment. In both rigid and nonrigid alignment, the connectivity remained the same, but the distance between the vertices changed in the nonrigid alignment, removing heterogeneity across different raw facial meshes in terms of orientation, location, and mesh structure (vertex indexing and triangulation).
Step 3 (region of interest [RoI] extraction) entailed ensuring that the fitted mesh corresponded to the template mesh with respect to vertex index and connectivity, enabling rapid extraction of topographical data. In Li et al,
the RoI was an egg-shaped profile around the nose and mouth. For this TFO design, the RoI was instead the full face without the eyes, nostrils, and mouth.
For Step 4 (CAD modeling) an open-source CAD platform (Blender) was utilized in a headless mode to seamlessly interface into the pipeline. Extracted facial data were extruded by 2 mm and smoothed, reducing sharp edges from the extrusion process, improving comfort around the eyes, nose, and mouth holes, and optical transparency (Fig. 4).
A technology appraisal of commercially available machines was conducted to identify suitable 3D printers that meet the requirements for TFO manufacture. Traditional manufacture involves cast and mold making, vacuum forming, and numerous hand tool tasks (such as cutting, grinding, and polishing)
with much equipment that requires storage, maintenance, and personnel to operate. Minimizing the number and size of equipment required for TFO manufacture was advantageous; reducing the total volume and weight of the equipment reduced the space requirements, increased the potential for mobile or redistributed manufacturing systems, reduced the number of personnel required, and created cost savings.
TFO mask prototypes were fabricated on a liquid crystal display (LCD) 3D printer (LC Dental; Photocentric Ltd) using a biocompatible and transparent resin. CAD casts were prepared for printing on a slicing software program (Photocentric Studio, v184.108.40.206; Photocentric Ltd). Figure 5A shows a TFO ready for printing (with support structures).
A TFO material must be biocompatible and is typically rigid. Polyethylene terephthalate glycol (PETG) and polycarbonate are popular for traditional methods because of their transparency and manufacturability, with typical Young moduli of 2100 MPa and 2200 MPa, respectively.
MED610 is the proposed material for any clinical trials, but all TFO prototypes in this study were manufactured from a highly transparent photopolymerized resin (ultraviolet [UV] Digital Light Processing [DLP] Poliglass; Photocentric Ltd), with a Young modulus of 2100 MPa and tensile strength of 40 MPa. This material is not biocompatible but was used because of its transparency and similarity to the mechanical properties of MED610 and because the research team had access to a compatible 3D printer.
Printer parameters were set to the manufacturer’s settings for this material. Once printed, the parts were removed from the print bed and were washed with a cleaning solution (SKU: RCL30RD01 resin cleaner; Photocentric Ltd) in an ultrasonic bath (Photocentric Wash 15; Photocentric Ltd) for 20 minutes at 50 °C. The TFOs were left to dry for 1 hour before polymerization at 60 °C for 60 minutes in the polymerization chamber (Photocentric Cure M+; Photocentric Ltd). Support structures were left attached during the polymerization process to minimize shrinkage (Fig. 5B). After polymerization, the supports were removed with side cutters (Side cutters 3766D; SensoPlus). Abrasion with a rotary tool (DREMEL 4000 F013400045; Robert Bosch GmbH) with 60-grit band attachment was used to remove the support structures left by the side cutters. Once the TFO was clear of large defects, both surfaces of the TFO were manually abraded with grits from 240-2500 (240-2500 grit sandpaper; LANHU) under a constant stream of water. The result of manual abrasion on the TFO is shown in Figure 5C. Next, a general-purpose clear coat spray paint (2k Clear Glamor Spray; SprayMax) was applied twice on both sides of the mask from a 20-cm distance. The result of the clear coat is shown in Figure 5D. Typically, a TFO with support structures required approximately 100 mL of photopolymer resin (40% of which was for support) and took less than 4 hours to completely print 1800 layers at a 100-μm layer height. Postprocessing also took less than 4 hours: 2.5 hours to wash, dry, and polymerize the TFO, less than 30 minutes to remove support and abrade the surface, and finally 1 hour to set the clear coat spray and obtain a glossy transparent finish on the TFO. The 3D facial scan capture took less than 1 minute, and the automated design process was approximately 2 minutes. Therefore, 3D facial scan to completed TFO manufacture took less than 8 hours.
Because of COVID19 restrictions, physical fit testing on the participants was not possible, and TFO prototypes were fit tested on facial replicas. The replicas consisted of a rigid sublayer that represented the skull structure and a flexible top layer mimicking the skin and soft tissues of the human face. The skull sublayer was 3D printed (Fortus 400MC; Stratasys) from ABS M30i (Stratasys). The elastomeric top layer was cast in a 3D-printed and smoothed mold from silicone (Platsil Gel-OO M41002/M41003; Mouldlife). The replica halves were designed using scans from a stereophotogrammetry machine (Di4D Snap 6200; Dimensional Imaging Ltd). Because of their geometric accuracy, these were used as a digital representation of actual face shape and as the ground truth scans. The combination of a skull-like sublayer and a soft silicone top (OO-30 hardness) created a facial replica with varying mechanical properties like an actual human face (such as stiff along the nasal walls and soft on the checks), as well as producing a highly accurate topographical representation of the participants face. Limitations of the skull sublayer were its rigidity that did not replicate the free-moving mandible. Figure 6 presents the replica components and production process.
Ten individuals were scanned. All scans were passed through the automated fitting pipeline. Since the purpose of the study was to create a mobile phone integrated solution, the resultant fitted meshes were categorized into 2 groups (mobile phone based and traditional) based on the ST type. The traditional scans were used to test the fidelity of the mobile phone STs. For each point on the processed scan, a nearest neighbor search was conducted on the ground truth scan. In this manner, a Euclidean distance map was built for each scan. For each study participant, an average of each ST was calculated. Table 3 presents the average root mean squared error (RMSE) and maximum distances for each ST (average of 10). The RMSE was calculated by first formatting the data using a k-nearest neighbors (KNN) search. This paired datapoints in the first mesh to their closest point in the second mesh. The RMSE was calculated after this.
Table 3Average distance measurements of raw scans against base truth face scan
In all STs, a noticeable difference was found between the RMSE and maximum Euclidean distances. Upon inspection, this clinical issue was caused by different STs not covering the exact same proportion of the face as the ground truth scan. The large distances were found around the edge of the face, between the 2 edges, like concentric circles. This is why RMSE was used to evaluate quality. The Scandy Pro technology presented the best RMSE for mobile STs. Therefore, these scans were used for the TFO design.
Figure 7 shows Euclidean distance maps for participants 1, 4, 8, and 10 between the TFO contact surface and ground truth scan. These participants represent a breadth of features, leading to areas of high or low Euclidean distances.
Table 4 presents the average RMSE and maximum distances for each participant’s TFO against their ground truth scans. The values are given to an accuracy of 1×10-4 m, since the iPhone TrueDepth scanner is only accurate to 0.1 mm at close distances and human skin only detects spatial displacements in this same order of magnitude. A noticeable difference was found between the RMSE and maximum distances of the raw scan and processed TFO, with the average RMSE decreasing by 37% (0.53 mm) and average maximum distance decreasing by 54% (5 mm). This difference was because only the desired facial regions were evaluated. Some scans included the upper body, which skewed the data to larger distances (Fig. 2 bottom right).
Table 4Average distance measurements of digital fit test, for each participant’s TFO, against base truth face scan. TFOs unaltered
Max Euclidean Distance (mm)
Relative standard deviation
RMSE, root mean squared error; TFOs, transparent facial orthoses.
Table 4 presents an RMSE distance of 0.59 to 1.17 mm and a maximum Euclidean distance of 2.13 to 6.29 mm. These same scans were also evaluated with areas of hair not included as the TFO would not be in contact with these areas. This resulted in RMSE distances of 0.48 to 0.94 mm and maximum distances of 1.07 to 2.89 mm (Table 5).
Table 5Average distance measurements of digital fit test without hair against base truth face scan. TFOs altered to remove features such as eyebrows and fringes
Max Euclidean Distance (mm)
Relative standard deviation
RMSE, root mean squared error; TFOs, transparent facial orthoses.
The results of this evaluation supported the hypothesis that a fully autonomous system for TFO manufacture would be successful. The first challenge was selecting the optimal data capture method. Previous studies used existing or self-fabricated photogrammetry devices,
selecting the highest quality capture device. The present authors recognized the increased prevalence of smart mobile phones with high quality depth sensitive cameras and available scanning applications (ScandyPro; Scandy, FaceApp; Bellus 3D). The data acquired confirmed that noncontact face shape capture with new mobile STs was as effective as traditional photogrammetry technologies, providing an affordable option that could be used without causing discomfort.
With the greatly reduced capture time and little to no need for preparation and setup, mobile 3D image capture technology stood out as a technological improvement.
Fabricating the TFOs was straightforward, with no print failures during manufacture and minimal postprocessing. As only an isopropanol bath and UV oven were required to clean and polymerize the TFOs, minimal technician input was required to deliver TFOs to patients. A full 3D printing method offered a more streamlined solution than current combined 3D printing and traditional processes.
Figure 7 presents heatmaps for the selection of participants. Figure 7A illustrates a good average fit. However, the top right of the participant’s brow has a spike in Euclidean distance. This is an example of hair, captured in the scan, interfering with the process. This heatmap demonstrated the need for a strict scanning protocol for clinical settings, where patient faces need to be cleared of any hair or artifacts that could disrupt the scan. Figure 7B illustrates the same issue since Scandy Pro could not pick up individual hairs, creating a raised region on the right eyebrow. Figure 7C was overall an excellent fit. However, a different facial expression in the 2 compared scans meant the left side of the face’s cheek and mouth had a greater offset than the right side (red area). Figure 7D was the best result, with the only problem area being around the nostrils. The improved result was because of a similar facial expression for the 2 technologies. In all participants, the RMSE distance between the TFO and ground truth scan was less than 1 mm.
Printed TFOs were fit tested on silicone face replicas. Figure 8A illustrates the mask placed on the mimic with no pressure, and Figure 8B illustrates the same, but with pressure applied. Contact areas were clearly visible around the cheekbone, indicating qualitatively how well the 3D printed TFO fit the replica and, by extension, the patient. A similar effect would be observed when applying the TFO to a face through skin blanching.
While the TFOs generally had a good fit, under visual inspection, some air gaps were visible. In a single participant, a large discrepancy was caused by warping during postprocessing. Support structure optimization was conducted to minimize distortion. A digital image analysis system for physical fit-testing could be developed and tested against participant trials. This would provide live feedback on fit quality and enable rapid geometric iteration with no specialist.
The participant group was predominantly White, and all participants were under the age of 40; however, the aim of this study was to produce an automated pipeline which produces the best possible fit. With wearable devices, a prevalent issue of demographic bias causes some groups to be less biometrically suited, leading to impaired product function. A study looking at facial half masks by Li et al
a well-fitting TFO does not necessarily mean a good clinical outcome. To attain the optimal pressure of 2000 to 3333 Pa, the TFO was pulled into the face by straps but also by offsetting the surface in convex areas such as around the nose, chin, and brow.
Traditional processes include manual offsetting by a clinician or the inclusion of silicone gels. The present study did not include pressure offsets, as the main objective was to achieve a good fit to the original scan with a fully automated pipeline. Wei et al
proposed pressure calculations based on the depth of soft tissue beneath the mask, and a similar approach could be incorporated into the pipeline developed in this study to automate the pressure offsets. To achieve the same result, a surface offsetting function was built to move the surface perpendicular to the vertex normal to increase or decrease the pressure. Once pressure calculations are implemented, this functionality will provide clinical structural modifications.
Based on the findings of this clinical study, the following conclusions were drawn:
A proof-of-concept system was demonstrated to automatically design and manufacture a custom-fitting TFO using commercially available 3D scanning and 3D printing technology and showed promise for future clinical implementation.
The total time taken from acquiring a patient’s 3D facial scan to TFO application was less than 8 hours.
After considering equipment size, transportation, and operational personnel requirements, vat polymerization (VP) technology was found to be the most promising route to successful TFO manufacture. However, a core challenge remains in terms of the availability of certified biocompatible materials fit for prolonged skin contact with the desired mechanical properties.
The automated design process was successful on 100% of the submitted scans, providing an average fit under 1 mm against the ground truth scan in all situations when digitally fitted and utilizing only AM.
Once pressure distribution geometry alterations are implemented, this will become the first system to require no specialist clinician.
Further studies should include a greater demographic and implement autonomous pressure evaluation with structural changes.
The authors thank Mansoor Khan and Sarah Karmel for their support and advice throughout the project. The authors also like to thank Mark Cutler for his expert opinion of TFO design and considerations. Finally, we would like to thank Photocentric for all their kind support over the duration of the project.
Harnessing the Transparent Face Orthosis for facial scar management: A comparison of methods.
Funding: Supported by the EPSRC under Grant EP/T014970/1 as part of the feasibility study: “A novel design-through-manufacture approach to Transparent Face Orthosis for deployed medical environments” and was administered by the Redistributed Manufacturing in Healthcare Network (RiHN).