Advertisement

Evaluating the accuracy of automated orthodontic digital setup models

Open AccessPublished:December 23, 2022DOI:https://doi.org/10.1053/j.sodo.2022.12.010

      Abstract

      Orthodontic tooth setup provides the orthodontist insight into the possibilities and limitations of treatment and enables visualization of the final goal of treatment. Lately, artificial intelligence techniques have been applied to the digital setup softwares to automate the process of tooth segmentation and alignment. The purpose of this study was to evaluate the accuracy and efficiency of automated digital setup softwares. The diagnostic digital impression data of 30 patients (mean age, 29.8 ± 14.4 y) who underwent non-extraction orthodontic treatment from 2020 to 2022 were collected. Three automated digital setup software programs, Autolign (Diorco, Yongin, Korea), Outcome Simulator Pro (Align Technology, Inc., Arizona, USA), and Ortho Simulation (Medit, Seoul Korea) were tested and compared to the manual setup model using the Maestro 3D Dental Studio software (AGE Solutions S.r.l., Pontedera, Italy). The linear and angular transposition with respect to six dimensions of tooth movement, including occlusal-gingival, facial-lingual, and mesial-distal bodily movement, as well as crown rotation, mesial-distal crown tip, and facial-lingual crown torque, were calculated. The mean error for linear and angular movement ranged 0.39 – 1.40 mm and 3.25 – 7.80 °, respectively. The effectiveness of automated digital setup systems varies among software, tooth type, and the dimension of movement. Although the setup results have shown improvement in the occlusion indices, further manual adjustments may be needed for clinical use, such as indirect bonding and clear aligner therapy.

      Introduction

      An orthodontic diagnosis is a crucial step, wherein the patient is examined using different modalities, such as radiographs, computed tomography, photos, and dental impressions.
      • Proffit W.R.
      • Fields H.W.
      • Larson B.E.
      • et al.
      Contemporary Orthodontics.
      In addition to the image data, patients’ health records and questionnaires regarding orofacial functions are collected to classify the type of malocclusion and plan for treatment. Tooth impressions provide clinicians with detailed information on the dental arches, such as molar relationships, overjet and overbite, tooth angulation and inclination, and arch asymmetries, which are considered in treatment planning. Also, dental models may be used to simulate the orthodontic treatment by separating individual teeth from the dental casts and aligning them in the desired position. This procedure, known as the “orthodontic tooth setup,” which was originated by Kesling
      • Kesling H.D.
      The diagnostic setup with consideration of the third dimension.
      provides the orthodontist insight into the possibilities and limitations of treatment and enables visualization of the final goal of treatment. The orthodontic tooth setup may be helpful, especially in cases that require critical decision-making, such as patients with a congenitally missing tooth, complex malocclusions with severe skeletal discrepancies, and borderline cases for extraction or non-extraction treatment. Orthodontic tooth setup models can also be used to fabricate orthodontic appliances such as indirect bonding systems and clear aligners.

      Digital setup

      A digital setup technique using digital impression data is replacing the conventional method using plaster casts. Digital setup eliminates the process of working with the plaster models, which is time-consuming and labor-intensive. Additionally, the digital setup provides accurate information on the changes in tooth position and arch forms, allows 3D superimposition to visualize the treatment changes, and enables precise assessment of the intra-arch and inter-arch relationships.
      • Sousa M.V.
      • Vasconcelos E.C.
      • Janson G.
      • et al.
      Accuracy and reproducibility of 3-dimensional digital model measurements.
      . Also, different versions of the setups can be made and compared to help determine the final treatment plan.
      Digital models can be manipulated to obtain a digital tooth setup model by segmenting and repositioning the teeth in the desired position, which is similar to the conventional setup models but only in the virtual space. There are numerous softwares available for analyzing the digital impression data and constructing a setup model. The reliability of digital models has been proven to be as effective and accurate as conventional models.
      • Sousa M.V.
      • Vasconcelos E.C.
      • Janson G.
      • et al.
      Accuracy and reproducibility of 3-dimensional digital model measurements.
      • Fleming P.S.
      • Marinho V.
      • Johal A.
      Orthodontic measurements on digital study models compared with plaster models: a systematic review.
      • Lippold C.
      • Kirschneck C.
      • Schreiber K.
      • et al.
      Methodological accuracy of digital and manual model analysis in orthodontics - a retrospective clinical study.

      Manual vs. automated digital setup

      Although intraoral scanners can capture the tooth anatomy with greater accuracy than conventional alginate impressions, the interproximal area is not captured because of the stereo-vision system of the scanner head.
      • Richert R.
      • Goujat A.
      • Venet L.
      • et al.
      Intraoral scanner technologies: a review to make a successful impression.
      . As a result, the teeth are connected as one body, like the plaster model. Therefore, each tooth has to be separated by a software. After setting the occlusal plane, the individual tooth is aligned by translating and rotating the tooth (Fig. 1).
      Fig 1
      Fig. 1The digital setup consists of tooth segmentation, orientation of occlusal plane, and tooth alignment by translation and rotation. Maestro 3D Dental Studio (v3, AGE Solutions S.r.l., Pontedera, Italy) requires segmentation of teeth, defining the occlusal plane, and aligning individual teeth manually.
      Lately, artificial intelligence techniques have been applied to the digital setup softwares to automate the process of tooth segmentation and alignment.
      • Vaid N.R.
      Artificial intelligence (AI) driven orthodontic care: a quest toward utopia?.
      . Autolign software (Diorco.Co., Yongin, South Korea), automatically segments the teeth and detects the facial axes of the clinical crown (FACC), the most prominent portion of the central lobe on the facial surface for the incisors and premolars, and buccal groove for molars.
      • Andrews L.F.
      Straight wire.
      After adjusting the FACC of each tooth and setting the occlusal plane and desired arch form (Fig. 2), tooth alignment is done automatically by the software. This type of software, which requires the user to review the reference axes of each tooth, set the occlusal plane, and determine the arch form before alignment, may be considered semi-automated setup software.
      Fig 2
      Fig. 2Autolign (Diorco.Co., Yongin, South Korea), an example of semi-automated setup software, automatically segments the teeth and detects the facial axes of the clinical crown (FACC). After adjusting the FACC of each tooth and setting the occlusal plane and desired arch form, the alignment is done automatically by the software.
      A fully automated setup software automatically segments and aligns the teeth from the intraoral scan data in an end-to-end manner. Outcome Simulator Pro software (Align Technology, Inc., Arizona, USA), an example of a fully automated setup software, is embedded in the iTero Scanner and provides an automated tooth setup in several minutes following the intraoral scan without having the clinician define the FACC, occlusal plane, and the arch form (Fig. 3).
      Fig 3
      Fig. 3Outcome Simulator Pro software (Align Technology, Inc., Arizona, USA) is a fully automated setup software, which is embedded in the iTero scanner. From the intraoral scan data, after checking facial axes of the clinical crown, it creates setup results in an end-to-end manner.
      Ortho Simulation (Medit, Seoul, South Korea) is another example of fully automated setup software. Similar to the Outcome Simulator Pro, Ortho Simulation is embedded in Medit's intraoral scanner software and only accepts scans acquired by Medit's intraoral scanners, the i500 and i700. After importing the scan data, the clinician has to define the upper and lower dental midlines. Then, the software recognizes tooth numbers and performs the setup (Fig. 4).
      Fig 4
      Fig. 4Ortho Simulation (Medit, Seoul, South Korea) is another example of fully automated setup software, which is embedded in i500. After importing the intraoral scanning data, the clinician has to adjust the upper and lower dental midlines. Then, the software segments and aligns teeth.
      With the advancement of artificial intelligence technologies, digital setup software is rapidly evolving to automate previously manual tasks such as tooth segmentation and alignment. Previous research found that digital setup models manually completed by clinicians had a high level of reliability.
      • Barreto M.S.
      • Faber J.
      • Vogel C.J.
      • et al.
      Reliability of digital orthodontic setups.
      ,
      • Im J.
      • Cha J.Y.
      • Lee K.J.
      • et al.
      Comparison of virtual and manual tooth setups with digital and plaster models in extraction cases.
      . To our knowledge, however, no study has assessed the performance of the automated setup softwares. Therefore, this study aimed to evaluate the quality of the setup model by using a fully automated setup model to ascertain the possibility of applying the setup results to clinical practice. The purpose of this study was to evaluate the accuracy and efficiency of automated digital setup softwares and compare it with the manual digital setup models performed by a clinician.

      Materials and methods

      This study was approved by the institutional review board of Asan Medical Center (2022-1051). Due to the retrospective nature of the study, informed consent from the patients was waived.
      The diagnostic digital impression data of 30 patients (mean age, 29.8 ± 14.4 y) who underwent non-extraction orthodontic treatment from 2020 to 2022 were collected. The intraoral scan data were obtained either by using the Trios 3 Pod (3Shape, Copenhagen, Denmark) or the iTero Element 5D (Align Technology, Inc., Arizona, USA). The inclusion criteria were: i) permanent dentition; ii) treated by non-extraction; iii) skeletal Class I (ANB 0 to 4); iv) mild to moderate crowding (arch length discrepancies of 2–5 mm); and v) less than 3 missing teeth. Exclusion criteria were: i) craniofacial anomalies and ii) presence of prostheses.

      Digital setup software

      We performed a manual digital setup (MS) using Maestro 3D Dental Studio (v3, AGE Solutions S.r.l., Pontedera, Italy), which is considered the gold standard. The patient's initial digital impression data was imported. The occlusal plane was adjusted, and the teeth were segmented by identifying the mesiodistal proximal contact points. The teeth were leveled and aligned by molar distalization, arch expansion, and incisor proclination as needed considering the molar and canine relationship and the incisor inclination.
      To assess the performance of the automated digital setup models, three commercial software programs, Autolign (semi-automated setup, SS), Outcome Simulator Pro (fully automated setup 1, FS1), and Ortho Simulation (fully automated setup 2, FS2), were used.
      For Autolign, initial intraoral scan data was imported first. The software automatically segmented teeth and detected the FACC and mesiodistal axis, then the investigator (HW) reviewed and adjusted the reference lines as needed. Using the ideal arch template, the arch width and form were selected considering the patient's initial arch form. The “Auto Align” function aligned the teeth along the pre-determined arch form. After the automated alignment, no manual adjustments were made.
      For Outcome Simulator Pro, the teeth must be scanned with the iTero intraoral scanner. For the patients whose diagnostic intraoral scan was done with the Trios 3, the scan data was 3D printed by using an FDM printer, the Cubicreator 4 (Cubicon, Sungnam, South Korea). Then the 3D-printed models were scanned using an iTero scanner. After importing the scan, the software automatically segmented the teeth, detected the FACCs, and aligned the teeth. After the initial automated setup, no manual adjustments were made.
      The Ortho Simulation software also required the scanning of the teeth using the Medit's intraoral scanner. Therefore, the 3D-printed models were scanned using the i500 (Medit, Seoul, South Korea) scanner. After importing the scan data into the software, the upper and lower dental midlines were identified by the user. Then the software automatically detected and segmented the teeth, but some adjustments for the segmentation were needed. After segmentation, automated alignment was done. After the initial automated alignment, no manual adjustments were made.

      Accuracy and efficiency analysis

      The linear and angular transposition with respect to six dimensions of tooth movement, including occlusal-gingival, facial-lingual, and mesial-distal bodily movement, as well as crown rotation, mesial-distal crown tip, and facial-lingual crown torque, were calculated. The linear translation movement was calculated in reference to the center of the clinical crown. The angular changes were analyzed using the long axis of the tooth that passes through the center of the crown parallel to the FACC. Positive values indicate occlusal, facial, and mesial for linear movement and mesial-out rotation, mesial angulation, and facial inclination for angular movement. To estimate the error of the automated setup model, the difference in the values of linear and angular tooth movement in the MS and the three automated setup software (SS, FS1, and FS2) was calculated. Because there were positive and negative values for each dimension of movement, the mean error was calculated using the absolute values to prevent underestimating the error.

      Quality of the setup

      To evaluate alignment quality, the setup models were assessed using the Peer Assessment Rating (PAR) index, which quantifies objectively the malocclusion using five categories: Displacement, Buccal occlusion, Overjet, Overbite, and Centerline.
      • Richmond S.
      • Shaw W.C.
      • Roberts C.T.
      • et al.
      The PAR Index (Peer Assessment Rating): methods to determine outcome of orthodontic treatment in terms of improvement and standards.
      ,
      • Richmond S.
      • Shaw W.C.
      • O'Brien K.D.
      • et al.
      The development of the PAR Index (Peer Assessment Rating): reliability and validity.
      .

      Statistical analysis

      Descriptive statistics were used to describe baseline demographic and clinical characteristics. The differences in tooth positions between SS and MS, FS1 and MS, and FS2 and MS were tested using the paired t-test. The Bonferroni correction was applied to multiple comparisons. A p-value of <0.05 was considered to indicate significance. The statistical analysis was performed using the Statistical Package for the Social Sciences software (version 22.0, IBM, Armonk, NY).

      Results

      Mean errors of linear and angular tooth movement as a result of automated setup are shown in Fig. 5.
      Fig 5
      Fig. 5Linear and angular movement errors as a result of automated setup by using (A) Autolign (semi-automated setup, SS), (B) Outcome Simulator Pro (fully automated setup, FS1), and (C) Ortho Simulation (fully automated setup, FS2). For linear movement errors, positive values indicate an automated setup position that is more extrusive, buccal, and mesial; negative values indicate an automated setup position that is more intrusive, lingual, and distal. For angular movement errors, positive values indicate automated setup position more mesially rotated, mesially angulated, and buccally inclined; Negative values indicate automated setup position more distally rotated, distally angulated and lingually inclined.
      Compared to the MS models, the setup models from SS showed intrusion of the maxillary premolars and molars and extrusion of the mandibular premolars and molars. Incisors and molars were positioned lingually, and premolars were positioned more buccally than in the MS models. All teeth were located distally. For angular movements, mandibular incisors and premolars were angulated mesially more than the MS. Maxillary molars were angulated mesially, and mandibular molars were angulated distally. All teeth were inclined lingually.
      In the FS1 models, maxillary incisors were extruded and positioned labially. Mandibular incisors were also located labially. All teeth were positioned mesial to those in the MS models. Vertical and bucco-lingual translative movement of upper and lower molars were smaller than SS and FS2. For angular movements, all teeth were distally rotated. Maxillary incisors and premolars were distally angulated, and the other tooth groups showed mesial angulation. Maxillary incisors were lingually inclined, and lower incisors were labially inclined. Maxillary molars were buccally inclined, and mandibular molars were lingually inclined.
      In the FS2 models, maxillary incisors were intruded, and maxillary molars were extruded. All teeth were positioned lingually. Both maxillary and mandibular molars were located mesially. Angular changes have shown distal rotation in all the teeth, with mesial angulation of the maxillary and mandibular incisors and mesial and distal angulations for the maxillary and mandibular molars, respectively. The incisors showed a more labial inclination.
      The mean absolute errors for linear and angular movement between the automated setup software and the MS and the paired comparison of the automated setup software are shown in Table 1.
      Table 1Mean absolute error of tooth movement discrepancies between the automated setup software and the manual setup software, and the differences among the automated setup softwares.
      1. Autolign (SS)2. Outcome

      Simulator (FS1)
      3. Ortho

      simulation (FS2)
      Difference (1-2)Difference (1-3)Difference (2-3)
      Vertical, mmUpper incisors0.75 ± 0.611.11 ± 0.721.30 ± 1.10
      Upper premolars1.20 ± 0.670.31 ± 0.330.77 ± 0.60
      Upper molars0.95 ± 0.580.30 ± 0.380.91 ± 0.68
      Lower incisors0.67 ± 0.500.61 ± 0.490.78 ± 0.60
      Lower premolars0.68 ± 0.530.38 ± 0.450.65 ± 0.48
      Lower molars0.73 ± 0.520.22 ± 0.320.73 ± 0.63
      Total0.81 ± 0.590.54 ± 0.590.88 ± 0.77-0.29 ± 1.09
      p < 0.016, indicating a statistically significant difference between softwares after Bonferroni correction. MS, manual setup (Maestro 3D); SS, semi-automated setup (Autolign); FS1, fully automated setup 1 (Outcome Simulator Pro), FS2, fully automated setup 2 (Ortho Simulation).
      -0.05 ± 1.580.24 ± 1.50
      p < 0.016, indicating a statistically significant difference between softwares after Bonferroni correction. MS, manual setup (Maestro 3D); SS, semi-automated setup (Autolign); FS1, fully automated setup 1 (Outcome Simulator Pro), FS2, fully automated setup 2 (Ortho Simulation).
      Buccal-lingual, mmUpper incisors0.75 ± 0.610.86 ± 0.740.80 ± 0.54
      Upper premolars0.72 ± 0.570.75 ± 0.660.78 ± 0.63
      Upper molars0.75 ± 0.580.45 ± 0.550.79 ± 0.55
      Lower incisors0.77 ± 0.580.91 ± 0.840.94 ± 0.67
      Lower premolars0.93 ± 0.680.68 ± 0.611.05 ± 0.80
      Lower molars0.87 ± 0.620.39 ± 0.551.40 ± 0.93
      Total0.79 ± 0.610.71 ± 0.710.95 ± 0.72-0.41 ± 1.29
      p < 0.016, indicating a statistically significant difference between softwares after Bonferroni correction. MS, manual setup (Maestro 3D); SS, semi-automated setup (Autolign); FS1, fully automated setup 1 (Outcome Simulator Pro), FS2, fully automated setup 2 (Ortho Simulation).
      0.24 ± 1.83
      p < 0.016, indicating a statistically significant difference between softwares after Bonferroni correction. MS, manual setup (Maestro 3D); SS, semi-automated setup (Autolign); FS1, fully automated setup 1 (Outcome Simulator Pro), FS2, fully automated setup 2 (Ortho Simulation).
      0.65 ± 1.67
      p < 0.016, indicating a statistically significant difference between softwares after Bonferroni correction. MS, manual setup (Maestro 3D); SS, semi-automated setup (Autolign); FS1, fully automated setup 1 (Outcome Simulator Pro), FS2, fully automated setup 2 (Ortho Simulation).
      Mesial-Distal,

      mm
      Upper incisors0.86 ± 0.690.68 ± 0.510.69 ± 0.54
      Upper premolars1.16 ± 0.870.94 ± 0.830.97 ± 0.64
      Upper molars1.38 ± 1.010.85 ± 0.881.19 ± 0.85
      Lower incisors1.14 ± 1.130.55 ± 0.540.91 ± 0.87
      Lower premolars1.62 ± 1.260.87 ± 1.030.98 ± 0.88
      Lower molars1.59 ± 1.270.97 ± 1.241.23 ± 1.06
      Total1.25 ± 1.070.78 ± 0.850.97 ± 0.83-1.31 ± 1.89
      p < 0.016, indicating a statistically significant difference between softwares after Bonferroni correction. MS, manual setup (Maestro 3D); SS, semi-automated setup (Autolign); FS1, fully automated setup 1 (Outcome Simulator Pro), FS2, fully automated setup 2 (Ortho Simulation).
      -0.83 ± 2.38
      p < 0.016, indicating a statistically significant difference between softwares after Bonferroni correction. MS, manual setup (Maestro 3D); SS, semi-automated setup (Autolign); FS1, fully automated setup 1 (Outcome Simulator Pro), FS2, fully automated setup 2 (Ortho Simulation).
      0.48 ± 1.73
      p < 0.016, indicating a statistically significant difference between softwares after Bonferroni correction. MS, manual setup (Maestro 3D); SS, semi-automated setup (Autolign); FS1, fully automated setup 1 (Outcome Simulator Pro), FS2, fully automated setup 2 (Ortho Simulation).
      Rotation, °Upper incisors5.60 ± 4.034.64 ± 3.736.24 ± 5.25
      Upper premolars4.43 ± 3.714.99 ± 4.665.07 ± 6.49
      Upper molars4.99 ± 4.342.36 ± 3.2511.02 ± 6.56
      Lower incisors6.28 ± 5.215.05 ± 3.897.18 ± 5.72
      Lower premolars7.43 ± 5.235.50 ± 4.5710.18 ± 9.27
      Lower molars4.71 ± 3.742.81 ± 3.698.29 ± 6.7
      Total5.63 ± 4.554.32 ± 4.127.80 ± 6.892.25 ± 7.74
      p < 0.016, indicating a statistically significant difference between softwares after Bonferroni correction. MS, manual setup (Maestro 3D); SS, semi-automated setup (Autolign); FS1, fully automated setup 1 (Outcome Simulator Pro), FS2, fully automated setup 2 (Ortho Simulation).
      4.59 ± 17.84
      p < 0.016, indicating a statistically significant difference between softwares after Bonferroni correction. MS, manual setup (Maestro 3D); SS, semi-automated setup (Autolign); FS1, fully automated setup 1 (Outcome Simulator Pro), FS2, fully automated setup 2 (Ortho Simulation).
      2.33 ± 16.5
      p < 0.016, indicating a statistically significant difference between softwares after Bonferroni correction. MS, manual setup (Maestro 3D); SS, semi-automated setup (Autolign); FS1, fully automated setup 1 (Outcome Simulator Pro), FS2, fully automated setup 2 (Ortho Simulation).
      Tip, °Upper incisors4.69 ± 4.783.90 ± 3.296.14 ± 4.99
      Upper premolars4.52 ± 3.024.17 ± 3.004.87 ± 4.6
      Upper molars4.55 ± 3.742.50 ± 3.795.74 ± 4.59
      Lower incisors3.58 ± 2.853.26 ± 2.826.03 ± 5.02
      Lower premolars4.48 ± 3.733.44 ± 2.684.98 ± 4.77
      Lower molars4.61 ± 3.971.89 ± 3.167.00 ± 5.72
      Total4.37 ± 3.783.25 ± 3.215.83 ± 5.00-1.20 ± 5.96
      p < 0.016, indicating a statistically significant difference between softwares after Bonferroni correction. MS, manual setup (Maestro 3D); SS, semi-automated setup (Autolign); FS1, fully automated setup 1 (Outcome Simulator Pro), FS2, fully automated setup 2 (Ortho Simulation).
      -1.33 ± 10.34
      p < 0.016, indicating a statistically significant difference between softwares after Bonferroni correction. MS, manual setup (Maestro 3D); SS, semi-automated setup (Autolign); FS1, fully automated setup 1 (Outcome Simulator Pro), FS2, fully automated setup 2 (Ortho Simulation).
      -0.09 ± 9.43
      Torque, °Upper incisors8.58 ± 5.794.40 ± 4.135.02 ± 4.61
      Upper premolars6.55 ± 4.843.93 ± 3.065.00 ± 3.75
      Upper molars7.97 ± 6.43.30 ± 4.927.91 ± 5.88
      Lower incisors6.02 ± 4.825.58 ± 4.777.23 ± 5.42
      Lower premolars5.55 ± 4.544.87 ± 3.966.27 ± 5.04
      Lower molars8.89 ± 6.463.34 ± 4.916.29 ± 4.99
      Total7.26 ± 5.634.35 ± 4.426.26 ± 5.09-3.94 ± 7.21
      p < 0.016, indicating a statistically significant difference between softwares after Bonferroni correction. MS, manual setup (Maestro 3D); SS, semi-automated setup (Autolign); FS1, fully automated setup 1 (Outcome Simulator Pro), FS2, fully automated setup 2 (Ortho Simulation).
      -2.76 ± 12.34
      p < 0.016, indicating a statistically significant difference between softwares after Bonferroni correction. MS, manual setup (Maestro 3D); SS, semi-automated setup (Autolign); FS1, fully automated setup 1 (Outcome Simulator Pro), FS2, fully automated setup 2 (Ortho Simulation).
      1.17 ± 9.89
      p < 0.016, indicating a statistically significant difference between softwares after Bonferroni correction. MS, manual setup (Maestro 3D); SS, semi-automated setup (Autolign); FS1, fully automated setup 1 (Outcome Simulator Pro), FS2, fully automated setup 2 (Ortho Simulation).
      low asterisk p < 0.016, indicating a statistically significant difference between softwares after Bonferroni correction.MS, manual setup (Maestro 3D); SS, semi-automated setup (Autolign); FS1, fully automated setup 1 (Outcome Simulator Pro), FS2, fully automated setup 2 (Ortho Simulation).
      The lowest linear and angular absolute discrepancies were observed in FS1 in all directions, except for the upper incisors and upper premolars. The absolute mean linear error ranged from 0.54 mm to 1.25 mm. Linear errors among the softwares were statistically significant (p < 0.05), except for vertical errors between SS and FS2. Absolute mean angular errors ranged from 3.25° to 7.78°, with a significant difference among the softwares except for crown tip errors between FS1 and FS2.
      Figs. 6 and 7 show the Bland-Altman plots of the tooth position errors, which illustrate the tendency of error for each software according to the type of tooth movement. The X-axis means the average of the tooth movements, and the Y-axis means the difference in tooth movement between softwares. Negative values indicate an automated tooth setup position with the facial surface rotated more distally, with a smaller mesial tip, or with less buccal crown torque than the manual digital setup (MS).
      Fig 6
      Fig 6Bland-Altman plots of linear tooth movement discrepancies between automated setup software and the manual digital setup software. (MS, manual setup; SS, semi-automated setup; FS1, fully automated setup 1, FS2, fully automated setup 2.)
      Fig 7
      Fig. 7Bland-Altman plots of angular tooth movement discrepancies between automated setup software and manual digital setup software. (MS, manual setup; SS, semi-automated setup; FS1, fully automated setup 1, FS2, fully automated setup 2.)
      Most of the tooth positions were within the range of 95% confidence levels. An upward-sloping trend was observed in all the plots, which indicates that as the average tooth movement increased, the tooth movement error increased.
      The PAR indices for setup quality assessment are shown in Table 2. Mean PAR index of the initial models was 26.2. The PAR indices of SS, FS1, FS2, and MS (the reference model) were 1.60 ± 1.13, 3.97 ± 5.40, 2.47 ± 1.74, and 0.77 ± 1.89 respectively. Buccal occlusion category accounted for the differences among the automated setup softwares.
      Table 2The mean Peer Assessment Ratings indices of initial models and the setup models by using manual, semi-automated, and fully automated software.
      Displacement

      (X 1)
      Buccal occlusion

      (X 1)
      Overjet

      (X 6)
      Overbite

      (X 2)
      Centerline

      (X 4)
      Total
      Initial10.67 ± 3.771.63 ± 1.521.80 ± 1.191.00 ± 1.170.37 ± 0.5626.17 ± 8.69
      Maestro 3D (MS)0.20 ± 0.610.17 ± 0.380.03 ± 0.180.03 ± 0.180.03 ± 0.180.77 ± 1.89
      Autolign (SS)0.10 ± 0.311.40 ± 1.100.00 ± 0.000.03 ± 0.180.00 ± 0.001.60 ± 1.13
      Outcome Simulator Pro (FS1)0.03 ± 0.181.60 ± 1.330.37 ± 0.810.10 ± 0.310.00 ± 0.003.97 ± 5.40
      Ortho simulation (FS2)1.03 ± 1.101.30 ± 0.750.00 ± 0.000.07 ± 0.250.07 ± 0.252.47 ± 1.74
      MS, manual setup (Maestro 3D); SS, semi-automated setup (Autolign); FS1, fully automated setup 1 (Outcome Simulator Pro), FS2, fully automated setup 2 (Ortho Simulation).
      The mean setup time is shown in Table 3. The mean setup time of MS, SS, FS1, and FS2 were 19.80 ± 3.05, 12.00 ± 2.15, 1.11 ± 0.18, and 3.11 ± 1.01 respectively.
      Table 3The mean setup time of software. The mean setup time of MS, SS, FS1, and FS2 were 19.80 ± 3.05, 12.00 ± 2.15, 1.11 ± 0.18, and 3.11 ± 1.01 respectively.
      Setup softwareTime (min)
      Maestro 3D (MS)19.80 ± 3.05
      Autolign (SS)12.00 ± 2.15
      Outcome Simulator Pro (FS1)1.11 ± 0.18
      Ortho simulation (FS2)3.11 ± 1.01
      MS, manual setup; SS, semi-automated setup; FS1, fully automated setup 1, FS2, fully automated setup 2.

      Discussion

      A digital setup model can visualize the final occlusion of orthodontic treatment goals, making accurate diagnosis and precise treatment planning possible. It can be used in conjunction with other types of patient data, such as clinical photos and radiographs, to determine the extraction, the tooth to be extracted if it is decided to extract, and the plan for anchorage to achieve a favorable outcome. The digital setup model can be combined with other 3D modalities, such as CBCT, in complex cases, such as tooth impaction, to visualize the impacted tooth to be aligned. It may also be used for virtual simulation surgery for skeletal malocclusion patients undergoing orthognathic surgery.
      • de Waard O.
      • Baan F.
      • Bruggink R.
      • et al.
      The prediction accuracy of digital orthodontic setups for the orthodontic phase before orthognathic surgery.
      The latest CAD/CAM technologies for digital setup have been augmented by applying AI algorithms to automate the tedious tasks that are required to be done by users. Tooth segmentation is an essential step for the reconstruction of a 3D tooth model, and digital models employ tooth segmentation methods wherein the recording of mesiodistal width and the proximal surface of the tooth is dependent on surface data, and areas that are not scanned may not be recorded accurately.
      • Im J.
      • Kim J.Y.
      • Yu H.S.
      • et al.
      Accuracy and efficiency of automatic tooth segmentation in digital dental models using deep learning.
      . Close contacts between adjacent teeth, misalignment, and crowding can make automated tooth segmentation time-consuming and difficult.
      • Lian C.
      • Wang L.
      • Wu T.H.
      • et al.
      Deep multi-scale mesh feature learning for automated labeling of raw dental surfaces from 3D intraoral scanners.
      . There have been advancements in automated segmentation algorithms using machine learning and deep learning. Recent studies have been done for automated tooth segmentation
      • Im J.
      • Kim J.Y.
      • Yu H.S.
      • et al.
      Accuracy and efficiency of automatic tooth segmentation in digital dental models using deep learning.
      • Lian C.
      • Wang L.
      • Wu T.H.
      • et al.
      Deep multi-scale mesh feature learning for automated labeling of raw dental surfaces from 3D intraoral scanners.
      • Kim T.
      • Cho Y.
      • Kim D.
      • et al.
      Tooth segmentation of 3D scan data using generative adversarial networks.
      • Wu T.H.
      • Lian C.
      • Lee S.
      • et al.
      Two-stage mesh deep learning for automated tooth segmentation and landmark localization on 3D intraoral scans.
      • Zhao Y.
      • Zhang L.
      • Liu Y.
      • et al.
      Two-stream graph convolutional network for intra-oral scanner image segmentation.
      . using deep learning techniques that have shown improved segmentation from an intraoral scanned image.
      • Zhao Y.
      • Zhang L.
      • Liu Y.
      • et al.
      Two-stream graph convolutional network for intra-oral scanner image segmentation.
      . Im et al
      • Im J.
      • Kim J.Y.
      • Yu H.S.
      • et al.
      Accuracy and efficiency of automatic tooth segmentation in digital dental models using deep learning.
      . compared automatic segmentation using landmark-based method (Ortho analyzer), tooth designation method (Autoalign), and deep learning and found the highest success rate (97.26%) using deep learning. Kim et al
      • Kim T.
      • Cho Y.
      • Kim D.
      • et al.
      Tooth segmentation of 3D scan data using generative adversarial networks.
      . proposed an automated segmentation tool for recreating missing data of interdental areas for an intraoral scanner using generative adversarial networks and reported a higher than conventional method using plaster casts. Wu et al
      • Wu T.H.
      • Lian C.
      • Lee S.
      • et al.
      Two-stage mesh deep learning for automated tooth segmentation and landmark localization on 3D intraoral scans.
      . employed mesh deep learning for automated tooth segmentation on 3D intraoral scans with a high segmentation accuracy measured with a dice similarity coefficient (DSC) of 0.964 ± 0.054.
      Contrary to the vast amount of research for automated tooth segmentations, there are few reports in the academic field on the algorithms for automated tooth alignment and their accuracy. As the fully automated software claims to be used for consultation purposes only, we aimed to evaluate how well the automated setup software worked and look into the possibilities for using it for clinical purposes such as making clear aligners and indirect bonding jigs.
      The commercial software we have used for the assessment of automated alignment has shown different results. The semi-automated setup software used in this study, Autolign, performed automatic alignment based on the FACC and the mesiodistal axis of each tooth that was automatically detected and fine-tuned by the user. After the user defined the occlusal plane, the teeth were automatically arranged along an ideal line of occlusion, and the curve of Spee was flattened. There are options that the software follows to perform the setup, such as the shape of the arch (normal, wide, and tapered), and these rules set by clinicians will determine the final setup result. Tooth angulation and inclination were set as ideal values according to the preset prescription; therefore, in non-extraction cases, crowding was resolved primarily by molar distalization. An expansion could be done by setting a wide arch form, and manual adjustments for interproximal reduction or incisor proclination are possible after the initial alignment. We believe that the rule-based algorithm of the Autolign, which uses the ideal arch form, accounts for a greater error in tooth position than the two fully automated setup softwares that preserved the patients’ original arch forms.
      The Outcome Simulator Pro software also detected the FACC of the teeth automatically but was more accurate and did not allow manual adjustments. It was done in an end-to-end manner, indicating that there were no additional options to be inputted for the process of automated setup. However, it tended to resolve crowding primarily by interproximal reduction from all teeth from mesial to the first molar to the mesial of the contralateral first molar and focused on matching the upper and lower dental midlines. This may be associated with the fact that this software was designed for consultation purposes only, so the incisor alignment is important. Also, upper incisors were extruded and lower incisors were intruded in most cases to form a proper overbite. The second molars were not fully corrected (Fig. 8). Additionally, the software focused on the incisor occlusion, and the first molar occlusion was not always corrected to Angle's Class I.
      Fig 8
      Fig. 8The Outcome Simulator Pro software did not fully correct the occlusion of second molars.
      Ortho Simulation software intrudes the upper incisors, which is different from other software. All teeth were lingually inclined and mesialized. As a result, the arch form became narrower than it was before the treatment. The reason for the lingual and mesial movement of the teeth was due to the decreased mesiodistal widths of the teeth during the segmentation stage (Fig. 9). Contrary to the Outcome Simulator Pro software, Ortho Simulation corrected the inclination of the second molars.
      Fig 9
      Fig. 9Ortho Simulation software could not recognize the mesial and distal borders of the teeth.
      The total setup time was drastically reduced when using the fully automated setup software. The semi-automated software, Autolign, showed a mean setup time of 12 min, while Outcome Simulator Pro and Ortho Simulation showed a mean setup time of 1.1 and 3.1 min, respectively. Increased time in the Autolign software was due to the process of confirming the FACC and mesiodistal axis of each tooth and setting the occlusal plane orientation and the arch form. Despite the short setup time for fully automated softwares, manual adjustments are required afterward, which should be considered when using it in clinical practice. The manual setup using Maestro 3D Dental Studio showed a mean time of 19.8 minutes, and this was done by an experienced clinician. Therefore, it may take longer for new users.
      The quality of the setup results was favorable for all the automated setup softwares, with the PAR indices ranging from 1.6 to 4.0. Richmond et al
      • Richmond S.
      • Shaw W.C.
      • O'Brien K.D.
      • et al.
      The development of the PAR Index (Peer Assessment Rating): reliability and validity.
      . stated that to demonstrate a high standard of treatment for a practitioner, the mean percentage reduction in weighted PAR index should be greater than 70% for the “greatly improved” category, and more than 30% for the “improved” category. The automated setup results in this study showed a reduction in the PAR indices by 94%, 85%, and 91% for Autolign, Outcome Simulator, and Ortho Simulation, respectively, and this falls under the "greatly improved" category according to Richmond.
      There are a few limitations to the study. First, the MS model may not be considered the ideal setup depending on the clinician and their treatment philosophies. Also, we used the 3D printed diagnostic models of patients and scanned them for Outcome Simulator Pro and Ortho Simulation, and this has affected the accuracy of detecting the reference planes such as FACC and mesiodistal proximal contact points for the automated setup software. Another limitation is that we used the amount of tooth movement generated by each software, and there might be a difference between the center of the crown and the reference axis detected from the software, and this might have affected the analysis of tooth movements. Since the software has a default option of non-extraction for the digital setup, we only included the patients who were indicated for the non-extraction treatment and who had moderate crowding and no sagittal skeletal discrepancies. Therefore, the inclusion of complex cases with skeletal discrepancies may reveal a greater amount of error. However, since the software does not have patients' clinical information, such as skeletal patterns, functional problems, periodontal conditions, and facial soft tissues, it would be feasible to assess the capabilities of aligning and leveling the arches only. Also, it has no information on the biological limitation of tooth movement. Therefore, it is the role of the clinician to diagnose the patient and make decisions about the treatment. The automated setup software should be used to support the clinicians by visualizing the treatment outcome based on the objectives set by the clinician.

      Conclusion

      The effectiveness of automated digital setup systems varies among software, tooth type, and the dimension of movement. The time efficiency has been considerably improved. Although the setup results have shown improvement in the PAR indices, further manual adjustments may be needed for clinical use, such as indirect bonding and clear aligner therapy.

      References

        • Proffit W.R.
        • Fields H.W.
        • Larson B.E.
        • et al.
        Contemporary Orthodontics.
        6th ed. Mosby Elsevier, St Louis2018
        • Kesling H.D.
        The diagnostic setup with consideration of the third dimension.
        Am J Orthod. 1956; 42: 740-748
        • Sousa M.V.
        • Vasconcelos E.C.
        • Janson G.
        • et al.
        Accuracy and reproducibility of 3-dimensional digital model measurements.
        Am J Orthod Dentofacial Orthop. 2012; 142: 269-273
        • Fleming P.S.
        • Marinho V.
        • Johal A.
        Orthodontic measurements on digital study models compared with plaster models: a systematic review.
        Orthod Craniofac Res. 2011; 14: 1-16
        • Lippold C.
        • Kirschneck C.
        • Schreiber K.
        • et al.
        Methodological accuracy of digital and manual model analysis in orthodontics - a retrospective clinical study.
        Comput Biol Med. 2015; 62: 103-109
        • Richert R.
        • Goujat A.
        • Venet L.
        • et al.
        Intraoral scanner technologies: a review to make a successful impression.
        J Healthc Eng. 2017; 8427595
        • Vaid N.R.
        Artificial intelligence (AI) driven orthodontic care: a quest toward utopia?.
        Semin Orthod. 2021; 27: 57-61
        • Andrews L.F.
        Straight wire.
        Concept Appl. 1989; : 7-33
        • Barreto M.S.
        • Faber J.
        • Vogel C.J.
        • et al.
        Reliability of digital orthodontic setups.
        Angle Orthod. 2016; 86: 255-259
        • Im J.
        • Cha J.Y.
        • Lee K.J.
        • et al.
        Comparison of virtual and manual tooth setups with digital and plaster models in extraction cases.
        Am J Orthod Dentofacial Orthop. 2014; 145: 434-442
        • Richmond S.
        • Shaw W.C.
        • Roberts C.T.
        • et al.
        The PAR Index (Peer Assessment Rating): methods to determine outcome of orthodontic treatment in terms of improvement and standards.
        Eur J Orthod. 1992; 14: 180-187
        • Richmond S.
        • Shaw W.C.
        • O'Brien K.D.
        • et al.
        The development of the PAR Index (Peer Assessment Rating): reliability and validity.
        Eur J Orthod. 1992; 14: 125-139
        • de Waard O.
        • Baan F.
        • Bruggink R.
        • et al.
        The prediction accuracy of digital orthodontic setups for the orthodontic phase before orthognathic surgery.
        J Clin Med. 2022; 11: 6141
        • Im J.
        • Kim J.Y.
        • Yu H.S.
        • et al.
        Accuracy and efficiency of automatic tooth segmentation in digital dental models using deep learning.
        Sci Rep. 2022; 12: 9429
        • Lian C.
        • Wang L.
        • Wu T.H.
        • et al.
        Deep multi-scale mesh feature learning for automated labeling of raw dental surfaces from 3D intraoral scanners.
        IEEE Trans Med Imaging. 2020; 39: 2440-2450
        • Kim T.
        • Cho Y.
        • Kim D.
        • et al.
        Tooth segmentation of 3D scan data using generative adversarial networks.
        Appl Sci. 2020; 10: 490
        • Wu T.H.
        • Lian C.
        • Lee S.
        • et al.
        Two-stage mesh deep learning for automated tooth segmentation and landmark localization on 3D intraoral scans.
        IEEE Trans Med Imaging. 2022; 41: 3158-3166
        • Zhao Y.
        • Zhang L.
        • Liu Y.
        • et al.
        Two-stream graph convolutional network for intra-oral scanner image segmentation.
        IEEE Trans Med Imaging. 2022; 41: 826-835