Abstract
Three-dimensional (3D) soft tissue prediction is replacing two-dimensional analysis in planning for orthognathic surgery. The accuracy of different computational models to predict soft tissue changes in 3D, however, is unclear. A retrospective pilot study was implemented to assess the accuracy of Dolphin 3D software in making these predictions. Seven patients who had a single-segment Le Fort I osteotomy and had preoperative (T 0 ) and >6-month postoperative (T 1 ) cone beam computed tomography (CBCT) scans and 3D photographs were included. The actual skeletal change was determined by subtracting the T 0 from the T 1 CBCT. 3D photographs were overlaid onto the T 0 CBCT and virtual skeletal movements equivalent to the achieved repositioning were applied using Dolphin 3D planner. A 3D soft tissue prediction (T P ) was generated and differences between the T P and T 1 images (error) were measured at 14 points and at the nasolabial angle. A mean linear prediction error of 2.91 ± 2.16 mm was found. The mean error at the nasolabial angle was 8.1 ± 5.6°. In conclusion, the ability to accurately predict 3D soft tissue changes after Le Fort I osteotomy using Dolphin 3D software is limited.
Planning for orthognathic surgery has historically been facilitated by two-dimensional (2D) soft tissue prediction of simulated osseous movements. As the availability of cone beam computed tomography (CBCT) and three-dimensional (3D) photography has increased, many software packages have added capability for 3D prediction. The ability to accurately simulate skeletal movements in 3D is invaluable in helping surgeons plan orthognathic procedures, inform patients of the expected results of their operations, and teach trainees.
Current software can reliably simulate hard tissue movements of the maxilla and mandible in 3D. A linear correlation between hard and soft tissue changes has been established in 2D, but these relationships have not been completely determined in 3D. In addition, prior studies of 3D prediction outcomes for orthognathic surgery have not incorporated colorized 3D photographs, an essential component to creating life-like predictions for treatment planning and patient education ( Fig. 1 ).
Three broad categories of computational modelling methods have been applied to facial soft tissue morphing: mass spring model (MSM), finite element model (FEM), and mass tensor model (MTM). Each has advantages and weaknesses, and no method has been accepted as the gold standard. These approaches have been shown to produce soft tissue predictions for orthognathic surgery that are accurate to within 0.27–1.17 mm. All of these modelling methods require a large amount of graphics and computational resources and are therefore difficult to apply in real time.
In order to accommodate real-time soft tissue prediction during virtual treatment planning and consultations with patients and families, Dolphin 3D Imaging (Dolphin Imaging & Management Solutions, Chatsworth, CA, USA), the market leader in orthognathic surgical planning, uses a landmark-based photographic morphing algorithm that was developed for 2D prediction and has been extrapolated to 3D. This system requires the user to plot 79 landmarks on the CT volume (42 bony, 37 soft tissue) and generates adjustable curves connecting these points, similar to the tracing of a lateral cephalometric radiograph ( Fig. 2 ).
The purpose of this study was to assess the accuracy of soft tissue prediction for Le Fort I osteotomy (LFI) using Dolphin 3D software. The hypotheses were (1) that Dolphin 3D would produce clinically useful 3D photographic predictions for LFI osteotomies, and (2) that Dolphin 3D would more accurately predict soft tissue changes for midline structures than for lateral facial points because the algorithms for 3D soft tissue morphing are extrapolated from experience with 2D changes in the midline. The specific aim was to measure differences in the predicted 3D soft tissue image compared to the actual result at multiple midline and lateral facial points in a series of patients after LFI osteotomy.
Materials and methods
To address the study question, a retrospective case series of patients who had a single-segment LFI osteotomy was implemented. This was designed as a pilot study with strict inclusion criteria in order to test the concept of 3D photographic prediction using Dolphin 3D software. This study was approved by the Institutional Review Board of the Center for Applied Clinical Investigation at Boston Children’s Hospital.
Study sample
The study population included patients who had a single-segment LFI osteotomy at Boston Children’s Hospital from March 2008 to June 2014. To be included, subjects had to have both preoperative (T 0 ) and at least 6-month postoperative (T 1 ) CBCTs and 3D photographs, and had to have completed orthodontic treatment by the time of the T 1 records. Patients were excluded if they had (1) craniofacial anomalies including cleft lip/palate, (2) additional operations at the time of LFI such as malar implants, mandibular or chin osteotomies, (3) multi-segment LFI osteotomies, (4) orthodontic appliances in place at the time of T 1 records, or (5) inadequate records. This study was limited to single-segment LFI osteotomy in order to facilitate the evaluation of soft tissue differences in one facial region without the influence from other osseous changes. All patients underwent pre- and postoperative orthodontic treatment and had fixed orthodontic appliances in place at the time of LFI osteotomy. All T 1 3D photographs were taken after orthodontic appliances had been removed.
Image acquisition and preparation
CBCTs were obtained using either an i-CAT scanner (3D Imaging Systems, Imaging Sciences International Inc., Hatfield, PA, USA) or a Planmeca Promax 3D Max scanner (Planmeca USA, Inc., Roselle, IL, USA), using standard CBCT exposure settings. Prior to obtaining each image, the patients were placed in natural head position and asked to close the teeth and relax the facial muscles. All CBCT volumes were captured from at least the glabella to the hyoid bone.
3D photographs were acquired using a 3D VECTRA M3 imaging system and VECTRA software version 5.5 (Canfield Scientific, Inc., Fairfield, NJ, USA). This system comprises five pod-mounted cameras that are positioned in an arc around the subject. System calibration was performed before each image capture. The patient was asked to fixate on a point at eye level on the wall in their direct vision prior to the image capture in order to establish natural head position. Images were saved as .OBJ files.
Each CBCT volume was digitally segmented into relevant anatomical units (maxilla, mandible) using Mimics software (Materialise Inc., Leuven, Belgium) ( Fig. 3 ). Each segmented unit was then converted to an STL file. The T 1 maxillary and mandibular segments were aligned to the T 0 CBCT using the skull base as a reference point. The segments were saved in these registered positions as new STL files and exported. These STL files, along with photographic and CBCT data, were imported into Dolphin 3D software version 11.8 (Dolphin Imaging & Management Solutions).
In Dolphin 3D, virtual bone cuts mimicking the true LFI osteotomies were created on the T 0 CBCT volumes using the virtual surgical planning module. The T 0 3D photographs were linked to the T 0 CBCT volumes using cranial and forehead landmarks. Seventy-nine points (42 bony, 37 soft tissue) were plotted onto the T0 linked CBCT/3D photograph as prompted by the Dolphin software, and curves connecting these points were generated automatically ( Fig. 2 ). These curves were adjusted to accurately align with the 3D photograph.
The T 0 maxillary segment was virtually moved to match the T 1 position. The T 0 mandibular segment was then auto-rotated as necessary to align with the T 1 position. A 3D soft tissue prediction image (T P ) was generated using the native Dolphin 3D algorithm. The T P image was exported to Mimics and aligned to the T 1 3D photograph using the nasion and ears as reference landmarks.
Image analysis
Differences between the T P and T 1 facial images (error) were measured for each subject at 14 points (six midline and eight lateral) and at the nasolabial angle ( Table 1 , Fig. 4 ). Twelve of these points are standard anthropometric or cephalometric landmarks. Two lateral points were created for this study: (1) lateral ala (LA), defined as the intersection of lines tangent to endocanthion (en) and subalare (sbal), and (2) maxillary buttress (MB), found at the intersection of lines tangent to exocanthion (ex) and sbal ( Fig. 4 A). Linear errors exceeding 2 mm were considered clinically significant, as this magnitude of difference has been suggested as the threshold for a visually perceptible facial difference.
