Patients and CBCT Scans
Pre- and posttreatment CBCT images of 42 female patients were retrieved from the archives of National Engineering Laboratory for Digital and Material Technology of Stomatology and Beijing Key Laboratory of Digital Stomatology. The patients were divided into 2 groups according to the treatment plan. The patients who received nonextraction treatment were divided into the nonextraction group and those who received extraction of four premolars were divided into extraction group.
The inclusion and exclusion criteria were: (1) female patients between 18 to 35 years old whose body weight change during the treatment period was less than 2 kg, (2) CBCT images which included the whole masseter muscle and corresponding facial area, (3) the absence of a posterior crossbite, (4) without mandibular deviation (the deviation of menton point less than 3mm), (5) without a history of facial surgery or trauma, and (5) no systematic disease.
CBCT scans were taken by Newtom VGi (Quantitative Radiology, Verona, Italy) with the following settings: field of view, 24×19 cm; 90 kV; 6.0 mA; scan time, 15 s; and voxel size, 0.3 mm. Pretreatment and posttreatment CBCT scans were superimposed using Dolphin 11.8 Premium (Dolphin Imaging & Management Solutions, Chatsworth, CA, USA) by frontal base area (Fig. 1). The reoriented posttreatment CBCT was exported in Digital Imaging and Communications in Medicine (DICOM) format. Pretreatment and reoriented posttreatment CBCT scans were used for segmentation, reconstruction and measurements.
The study was approved by the Institutional Review Board of Peking University School and Hospital of Stomatology (PKUSSIRB-201944062).
Cephalometric radiographs were reconstructed from pretreatment and reoriented posttreatment CBCT images. ANB and FMA (Frankfurt-mandibular angle) were selected and measured to represent the sagittal and vertical skeletal relationship, respectively (Fig. 2).
Segmentation of the Masseter Muscle
A self-developed generative adversarial network (GAN)-based framework was used for noise reduction and automatic segmentation of masseter muscles from CBCT scans. The framework was developed by the Department of Machine Intelligence, Key Laboratory of Machine Perception (MOE), Peking University. To ensure the accuracy of segmentation, a layer by layer manual check was performed using ITK-SNAP 3.6.0 (http://www.itksnap.org) based on the automatic segmentation result. Pre- and posttreatment scans were placed in parallel and manually edited at the same time (Fig. 3) to ensure the consistency of anatomic structures. The left and right masseter muscle were separately exported and saved as a Stereo-Lithography Interface (STL) format.
Reconstruction of the Masseter Muscle and Facial 3D Models
Threshold segmentation was used to generate facial surface 3D models. Pretreatment and reoriented posttreatment CBCT were imported into Mimics Research 22.0 (Materialise NV, Leuven, Belgium). The initial models of bilateral masseter muscles were imported and transferred to unify their own space coordinates with the systematic coordinates of the Mimics software. Then, the 3D models of the craniofacial bone structure, outer layer of the facial surface and the masseter muscle were calculated and saved as STL format.
Measurement of the Thickness of the Facial Soft Tissues
3D models of pre- and posttreatment craniofacial bones, facial surfaces and masseter muscles were imported into Geomagic Studio 14.0 (3D Systems Inc., Morrisville, NC, USA). The Frankfurt horizontal plane (FH plane in Fig. 4a, b) was generated by the fitted plane of bilateral Obitale points and Porion points. The sagittal plane was defined by the plane perpendicular to the FH plane and passing through the anterior nasal spine (ANS) point and posterior nasal spine (PNS) point. The boundaries of the masseter area of facial soft tissue (MAS) was defined using the reference lines as follows: upper boundary- the tangent line passing the lower margin of the zygomatic arch; lower boundary- the tangent line passing the lower margin of the mandibular body; anterior boundary- the tangent line passing the anterior margin of the masseter muscle; and posterior boundary- the line connecting the middle of the articular tubercle and the posterior point of the gonial angle (Fig. 4a). These lines were projected on the sagittal plane forming four reference planes vertical to the sagittal plane. The MAS was cut out from the facial model by the four reference planes (Fig. 4b). The average deviation calculated between the pre- and posttreatment MAS was used to represent the thickness change on each side (Fig. 4c). The mean value of the left and right sides was used to represent the change in thickness of MAS in one patient.
The masseter muscle was cut by a plane formed by its own superoinferior axis and anteroposterior axis (Fig. 5a) calculated by principal component analysis (PCA) and complied by MATLAB R2018b (MathWorks Inc., Natick, MA, USA) to ensure that only the lateral half surface of the masseter muscle was used for comparison (Fig. 5b). The average deviation calculated between the pre- and posttreatment lateral surface of masseter muscle was used to represent the thickness change on the left and right sides, respectively (Fig. 5c). The mean value of the left and right sides was used to stand for the thickness change of the masseter muscle in one patient.
Between the skin surface of the face and lateral surface of the masseter muscle lies the fat tissue (FT). In this study, FT thickness was defined as the average deviation between the lateral surface of the masseter muscle and the corresponding facial surface. The change of FT thickness was calculated by subtracting the posttreatment FT thickness from the pretreatment FT thickness.
The patients were divided into the extraction group and nonextraction group according to their treatment plan. The thickness changes of MAS, MM and FT were compared with zero. Pretreatment age, treatment duration, ANB, FMA and the thickness changes of facial soft tissues (MAS, MM and FT) were compared between the extraction and nonextraction groups using a t-test if the data were normally distributed; otherwise, the Mann-Whitney test was used. Spearman’s correlation analysis was conducted among the abovementioned variables. In collinearity diagnostics, no significant multicollinearity was detected, so a linear regression model was used to confirm the correlations.
The sample size was calculated by Power Analysis and Sample Size (PASS) 15.0.1 software (NCSS LLC, UT, USA). To achieve a power of 0.80 in comparison between the two groups, at least 20 patients should be included in each group. And regression analysis required about 44 samples so that the power was about 0.81, therefore the two groups were analyzed as a whole (n=42).
The intraclass correlation coefficient (ICC) was calculated to assess the reliability of the measurements. Since the calculation of the thickness change of the masseter muscle was conducted automatically by the computer program and software, which did not include random error, it was not repeated. The measurement of MAS, ANB and FMA was repeated once two weeks later by the same examiner and a second examiner, and the intra- and interexaminer ICCs were both larger than 0.99 for MAS measurement and between 0.98-0.99 for ANB and FMA measurement.
The statistical analysis was conducted using SPSS Statistics 23.0 (IBM Co., Armonk, NY, USA) at a significance level of 0.05.