Ok A Review of The Use of Artificial Intelligence in Orthodontics
Ok A Review of The Use of Artificial Intelligence in Orthodontics
Ok A Review of The Use of Artificial Intelligence in Orthodontics
https://dergipark.org.tr/omujecm
1. Introduction
Digital data processing technologies in medical and dental can be integrated into dental imaging systems to identify even
fields have gained attention in the last two decades. Utilization the smallest deviations which human eye cannot recognize.
of digital technology, especially artificial intelligence (AI) With this outstanding ability, it can easily be used to make
technology, can help to reduce the cost and duration of accurate diagnosis of cephalometric landmarks (Tong et al.,
treatment, the need for human expertise and the number of 1989).
medical error cases. This approach also has a revolutionary
Artificial intelligence-based software systems have
potential in public health scenarios in developing countries.
significant and modificative role in the field of orthodontics
Artificial intelligence, which was brought forward by and they are considered as the future of dental applications. For
McCarthy in 1956, can be described as the behavior of the non- this reason, we aimed to review the literature on the use of AI
biologic beings which has the capacity to perceive complex technology in the orthodontic field (Table 1). Artificial
environments, learn and react accordingly (Nilsson and intelligence is used in every area of orthodontics from patient
Nilsson, 1998). Artificial intelligence does not necessarily communication and diagnosis to treatment processes.
mimic the human brain, it is rather a problem-solving tool Orthodontic software programs which use AI technologies are
which has its own set of rules. Studies have been conducted to based on “machine learning” technology. “The machine” uses
achieve human-like behaviors with AI and it has been found raw data to collect information from a database in machine
that computers exceed human results in many parameters learning technology. These software programs can analyze
(Faber et al., 2019). Artificial intelligence technology has been diagnostic dental radiographs and photos, also they can give
used in a wide spectrum from differential diagnosis and guidance to the dentists, during 3D intraoral scanning, to reach
radiographic interpretation to restorative treatment in dental an ideal model easily (Kattadiyil et al., 2014). The use of AI
field (Khanna, 2010). Dental management software, which can be divided into two main application areas in orthodontics
uses AI to gather and store the patient data, is available in the in particular: diagnosis and treatment (Fig. 1).
market. In this point, artificial intelligence can be used to
2. Artificial intelligence and orthodontic diagnosis
generate complete detailed virtual databases which are easily
Patient data, carefully obtained from an adequate database
accessible. Interactive and voice recognizing interfaces help
containing a detailed list of the patient's problems, form the
dental clinicians to easily complete some complex tasks.
basis of correct and accurate orthodontic diagnosis. The
Software with AI technology can document the necessary data
orthodontic diagnostic database can be obtained from written
and transfer them to the clinician faster and more efficiently
or verbal interview data; clinical examination and examination
than its human counterparts (Kannan, 2017). With its unique
of patient records including dental impressions, radiographs,
learning ability, AI can be trained to perform different tasks. It
and diagnostic photographs (Proffit et al., 2018).
* Correspondence: bsakdeniz@hotmail.com
Akdeniz and Tosun / J Exp Clin Med
Table 1. Current literature on the use of artificial intelligence in orthodontic
158
Akdeniz and Tosun / J Exp Clin Med
does not eliminate the need for an expert clinician for analysis asserted that the developed software had a high success rate
and decision-making steps (Yagi et al., 2010). The automation (over 90%) in differential diagnosis of cephalometric
systems which use AI and machine learning technologies landmarks. The automated tracing module was integrated into
remarkably have decreased the evaluation workload and a recent web-based software. The web-based software can also
prevented the diagnostic variations (Murata et al., 2017). detect soft tissue profile in profile photographs and with its
orthognathic surgery planning module, it can simulate possible
Different algorithms of AI systems were tested in several
soft tissue changes after planned orthognathic treatment.
studies in the orthodontic field. All these algorithms needed a
big database of patient examination records as input. The 4. Estimation of growth and development
results showed that the use of AI during diagnosis reduced the Timing is one of the main components of orthodontic
need for an expert clinician and the number of diagnostic treatment. Growth and development can be estimated by
errors. The researchers concluded that the AI applications were anthropometric indicators like chronologic age, menarche,
promising in orthodontic field (Kim et al., 2009; Yagi et al., vocal changes, height increase and skeletal maturation (skeletal
2010; Auconi et al., 2011;; Niño-Sandoval et al., 2016; Wang age) (Hägg and Taranger, 1982). Radiographs are widely used
et al., 2016; Murata et al., 2017). for detection of skeletal maturation indicators (Hägg and
Taranger, 1980). Deep learning (a machine learning algorithm
Noroozi et al. (2006) described a software which used
that uses multiple layers to progressively extract higher level
“fuzzy logic” concept. The software took graphical and
features from the raw input) and AI technologies were used by
numeric patient data as input and could recommend treatment
several authors to automate the age estimation by examining
plan for non-surgical orthodontic patients. Fuzzy logic enables
hand and wrist radiographs. With deep learning ability, AI
the software work with the nominal parameters. Human brain
systems can evaluate the radiographs after the input of a vast
is naturally accustomed to these “fuzzy” parameters. The
database consists of race, age, and gender. Results show that
authors asserted that the software program could suggest
the AI systems can evaluate the skeletal maturity with a
treatment options even for the specific situations like missing
performance like a radiologist (Lee et al., 2017; Spampinato et
teeth.
al., 2017; Iglovikov et al., 2018; Larson et al., 2018).
3. Automated cephalometric tracing
Maturation levels of cervical vertebrae are also used for
Tracing of cephalometric radiographs can either be done
assessment of skeletal maturity. Kök et al. (2019) compared
manually or digitally with computer aid. Although the use of
seven different, widely used AI algorithms to estimate cervical
computers for cephalometric tracing aims to save time by
vertebrae maturation levels. Artificial Neural Networks (ANN)
reducing tracking errors and increasing the diagnostic value of
algorithm, which is a mathematical model of human nervous
cephalometric analysis, the inconsistency in identifying
system formed by artificial nerve cells, showed better results.
anatomical landmarks is still a major source of random error
The authors concluded that ANN could be used in the future
(Miller et al., 1971).
applications for determining cervical vertebrae stage.
In order to overcome this problem, efforts have been made
5. Facial proportions
to automate cephalometric analysis with the aim of reducing
Evaluation of facial proportions includes measurement of
errors and the time required for analysis (Hutton et al., 2000).
ratios and linear lengths between facial structures. Although
Levy-Mandel et al. (1985) conducted the first study on lateral cephalometric radiographs and profile photographs are
automatic extraction of anatomical landmarks on lateral widely used for linear assessments, it is difficult to perform
cephalometric radiographs. They preprocessed the image with sensitive measurements because of the magnification
an edge-detector and knowledge-based line-following differences. Ratios and angular measurements are independent
algorithm, involving a production system with organized sets of dimensions and generally used for photographic assessment.
of rules and a simple interpreter, was subsequently applied.
Measurements of “ideal” facial proportions are currently
Automated cephalometric tracing was subsequently studied by
used by surgeons and orthodontists to comprehend the ideals
several other researchers and proved to perform as successfully
of beauty and reproduce aesthetically “beautiful” proportions
as expert dentists and could be used to accelerate the
(Harrar et al., 2018). However, the classical rules of ideal facial
cephalometric diagnostic phase (Tanikawa et al., 2009, 2010;
aesthetics have some deficiencies in reflecting the beauty
Mario et al., 2010; Banumathi et al., 2011; Gupta et al., 2015;
perception of the population because facial beauty is a very
Montúfar et al., 2018a, 2018b; Kunz et al., 2020). Although AI
subjective concept and there is not widely used and validated
systems have not been utilized for fully automated
set of rules for facial aesthetics, which is approved by the
cephalometric tracing yet, they have reached the maturity to be
population. (Knight and Keith, 2005; Yin et al., 2014).
used in some existing cephalometric software programs to
suggest possible locations of anatomical structures. Today, AI applications do not only perform basic tasks
such as optical facial recognition, but they are also matured
Lee et al. (2020) used deep convolutional neural network-
enough to simulate much complex cognitive tasks including
based analysis for automated cephalometric tracing. Authors
159
Akdeniz and Tosun / J Exp Clin Med
analysis and interpretation of facial data. Studies in this field variables. On the other hand, it has some shortcomings, such
showed that AI systems seemed to be promising tools to build as limitations in identifying all possible outcomes and
a validated formula for the human perception of facial establishing a linear relationship between variables and their
attractiveness (Patcas et al., 2019; Yu et al., 2014). outcomes (Zarei et al., 2006).
6. Artificial intelligence and orthodontic treatment planning Artificial neural networks were cited as good candidates to
Extraction decision develop a predictive model for orthodontic therapy, thanks to
Planning phase is the most significant and critical part of their ability to detect complex non-linear relationships between
orthodontic treatment. Extractions should be carefully planned inputs and outputs. Artificial neural networks were shown to
due to their irreversible nature. Clinicians come to the stage of have the ability to learn and generalize beyond the situations
deciding to extractions after combining the patient data derived they were faced with (Zarei et al., 2006).
from clinical evaluations, diagnostic photographs, dental
There are studies in the literature which showed that the
models and radiographs with their clinical expertise. Although
treatment results of Class II and Class III patients could be
practitioners with less experience can learn from the decisions
simulated by utilizing artificial neural networks technique. The
of their more experienced colleagues, the lack of a standard
researchers conclude that the neural networks technique is a
assessment method for the decision-making process requires a
promising tool which can be used for simulation of different
different approach. Neural networks were used to mimic
malocclusion models (Zarei et al., 2006; Auconi et al., 2015).
human decision-making process for orthodontic extractions.
Simulation of orthodontic treatment has gained
Sagittal, vertical and molar relationships, tooth inclinations,
popularity by clear aligner systems produced by a digital
overjet, overbite, protrusion index, soft tissue characteristics
process.
and patient complaints were given as input. Artificial
intelligence system can then guide the clinician to decide the Moving the teeth with “tooth positioning appliances”
extraction, based on the analysis fed from the mentioned through sequential stages which are formed by “set-ups” on
inputs. Studies showed that artificial intelligence systems can plaster models was a concept introduced by Kesling (1945).
assist clinicians by preventing errors in decision step and can The major drawback of this technique was that there was a need
provide 80 to 90% accuracy when making an orthodontic to manually subdivide the movement into small increments by
extraction decision (Jung and Kim, 2016; Xie et al., 2010). different plaster set-ups for each increment (Faltin et al., 2003).
160
Akdeniz and Tosun / J Exp Clin Med
was successful in simulating simpler treatment plans, there analysis towards an informed clinical aesthetic practice. Aesthetic
were significant differences between the simulation and Plast. Surg. 42, 137-146.
clinical results in more complex treatments. The ClinCheck 11. Hutton, T.J., Cunningham, S., Hammond, P., 2000. An evaluation
software showed extremely limited reliability when it came to of active shape models for the automatic identification of
cephalometric landmarks. Eur. J. Orthod. 22, 499-508.
simulation of extraction therapy. ClinCheck models failed to
accurately reflect patients' final occlusion in complex 12. Iglovikov, V.I., Rakhlin, A., Kalinin, A.A., Shvets, A.A., 2018.
Paediatric bone age assessment using deep convolutional neural
treatments. networks. In deep learning in medical image analysis and
multimodal learning for clinical decision support. Springer,
7. Conclusion
Quebec, pp. 300-308.
It is quite clear that AI technology has a significant impact on
13. Jung, S.K., Kim, T.W., 2016. New approach for the diagnosis of
the dental field, and so far, there have been major investments
extractions with neural network machine learning. Am. J. Orthod.
in this field. Although early attempts showed apparent Dentofac. Orthop. 149, 127-133.
deficiency, improvement in AI area is accelerating. Artificial
14. Kannan, P.V., 2017. Artificial intelligence applications in
intelligence can be a useful and practical tool for minimizing healthcare. Asian Hosp. Healthc. Manag. 30, 5.
errors and improving patient care.
15. Kattadiyil, M.T., Mursic, Z., AlRumaih, H., Goodacre, C.J., 2014.
One of the most common criticisms against AI technology Intraoral scanning of hard and soft tissues for partial removable
dental prosthesis fabrication. J. Prosthet. Dent. 112, 444-448.
stems from the fear that corporate initiatives will exclude
expert clinicians from the healthcare system and reduce 16. Khanna, S., 2010. Artificial intelligence: Contemporary
applications and future compass. Int. Dent. J. 60, 269-272.
treatment costs by using AI systems. Furthermore, it is difficult
to say that this is an unnecessary fear because recent 17. Kim, B.M., Kang, B.Y., Kim, H.G., Baek, S.H., 2009. Prognosis
prediction for class III malocclusion treatment by feature wrapping
developments show that attempts in this direction have already
method. Angle Orthod. 79, 683-691.
started. Although it is still clear that AI is not likely to replace
18. Knight, H., Keith, O., 2005. Ranking facial attractiveness. Eur. J.
clinicians in the near future, the increasing use of digital 3D
Orthod. 27, 340-348.
technologies in orthodontics shows that AI technology, which
19. Kunz, F., Stellzig-Eisenhauer, A., Zeman, F., Boldt, J., 2020.
helps in interpretation of complex data, will also keep
Artificial intelligence in orthodontics: Evaluation of a fully
attracting increasing attention. automated cephalometric analysis using a customized
convolutional neural network. J. Orofac. Orthop. der
References Kieferorthopädie. 81, 52.
1. Auconi, P., Caldarelli, G., Scala, A., Ierardo, G., Polimeni, A., 20. Larson, D.B., Chen, M.C., Lungren, M.P., Halabi, S.S., Stence,
2011. A network approach to orthodontic diagnosis. Orthod. N.V, Langlotz, C.P., 2018. Performance of a deep-learning neural
Craniofac. Res. 14, 189-197. network model in assessing skeletal maturity on pediatric hand
2. Auconi, P., Scazzocchio, M., Cozza, P., McNamara Jr, J.A., radiographs. Radiology. 287, 313-322.
Franchi, L., 2015. Prediction of Class III treatment outcomes 21. Lee, H., Tajmir, S., Lee, J., Zissen, M., Yeshiwas, B.A., Alkasab,
through orthodontic data mining. Eur. J. Orthod. 37, 257-267. T.K., Choy, G., Do, S., 2017. Fully automated deep learning system
3. Banumathi, A., Raju, S., Abhaikumar, V., 2011. Diagnosis of for bone age assessment. J. Digit. Imaging. 30, 427-441.
dental deformities in cephalometry images using support vector 22. Mario, M.C., Abe, J.M., Ortega, N.R.S., Del Santo Jr, M., 2010.
machine. J. Med. Syst. 35, 113-119. Paraconsistent artificial neural network as auxiliary in
4. Buschang, P.H., Ross, M., Shaw, S.G., Crosby, D., Campbell, cephalometric diagnosis. Artif. Organs. 34, E215-E221.
P.M., 2014. Predicted and actual end-of-treatment occlusion 23. Miller, R., Dijkman, D., Riolo, M., Moyers, R., 1971. Graphic
produced with aligner therapy. Angle Orthod. 85, 723-727. computerization of cephalometric data.
5. Faber, J., Faber, C., Faber, P., 2019. Artificial intelligence in 24. Montúfar, J., Romero, M., Scougall-Vilchis, R.J., 2018a.
orthodontics. APOS Trends Orthod. 9, 201-–205. Automatic 3-dimensional cephalometric landmarking based on
6. Faltin, R.M., de Almeida, M.A.A., Kessner, C.A., Faltin, K.J., active shape models in related projections. Am. J. Orthod.
2003. Efficiency, three-dimensional planning, and prediction of the Dentofac. Orthop. 153, 449-458.
orthodontic treatment with the Invisalign System: Case report. R. 25. Montúfar, J., Romero, M., Scougall-Vilchis, R.J., 2018b. Hybrid
Clin. Ortodon. Dent. Press 2, 61-71. approach for automatic cephalometric landmark annotation on
7. Gupta, A., Kharbanda, O.P., Sardana, V., Balachandran, R., cone-beam computed tomography volumes. Am. J. Orthod.
Sardana, H.K., 2015. A knowledge-based algorithm for automatic Dentofac. Orthop. 154, 140-150.
detection of cephalometric landmarks on CBCT images. Int. J. 26. Murata, S., Lee, C., Tanikawa, C., Date, S., 2017. Towards a fully
Comput. Assist. Radiol. Surg. 10, 1737-1752. automated diagnostic system for orthodontic treatment in dentistry.
8. Hägg, U., Taranger, J., 1980. Menarche and voice change as 2017 IEEE 13th Int. Conf. e-Science 1-8.
indicators of the pubertal growth spurt. Acta Odontol. Scand. 38, 27. Nilsson, N.J., Nilsson, N.J., 1998. Artificial intelligence: A new
179-186. synthesis. Morgan Kaufmann.
9. Hägg, U., Taranger, J., 1982. Maturation indicators and the 28. Niño-Sandoval, T.C., Perez, S.V.G., González, F.A., Jaque, R.A.,
pubertal growth spurt. Am. J. Orthod. 82, 299-309. Infante-Contreras, C., 2016. An automatic method for skeletal
10. Harrar, H., Myers, S., Ghanem, A.M., 2018. Art or Science? An patterns classification using craniomaxillary variables on a
evidence-based approach to human facial beauty a quantitative Colombian population. Forensic Sci. Int. 261, 159-e1.
161
Akdeniz and Tosun / J Exp Clin Med
29. Patcas, R., Bernini, D.A.J., Volokitin, A., Agustsson, E., Rothe, R., 36. Wang, X., Cai, B., Cao, Y., Zhou, C., Yang, L., Liu, R., Long, X.,
Timofte, R., 2019. Applying artificial intelligence to assess the Wang, W., Gao, D., Bao, B., 2016. Objective method for evaluating
impact of orthognathic treatment on facial attractiveness and orthodontic treatment from the lay perspective: An eye-tracking
estimated age. Int. J. Oral Maxillofac. Surg. 48, 77-83. study. Am. J. Orthod. Dentofac. Orthop. 150, 601-610.
30. Proffit, W.R., Fields, H.W., Larson, B., Sarver, D.M., 2018. 37. Xie, X., Wang, L., Wang, A., 2010. Artificial neural network
Contemporary orthodontics. Elsevier Health Sciences. modeling for deciding if extractions are necessary prior to
orthodontic treatment. Angle Orthod. 80, 262-266.
31. Spampinato, C., Palazzo, S., Giordano, D., Aldinucci, M.,
Leonardi, R., 2017. Deep learning for automated skeletal bone age 38. Yagi, M., Ohno, H., Takada, K., 2010. Decision-making system for
assessment in X-ray images. Med. Image Anal. 36, 41-51. orthodontic treatment planning based on direct implementation of
expertise knowledge. 2010 Annu. Int. Conf. IEEE Eng. Med. Biol.
32. Tanikawa, C., Yagi, M., Takada, K., 2009. Automated 2894-2897.
cephalometry: System performance reliability using landmark-
dependent criteria. Angle Orthod. 79, 1037-1046. 39. Yin, L., Jiang, M., Chen, W., Smales, R. J., Wang, Q., Tang, L.,
2014. Differences in facial profile and dental esthetic perceptions
33. Tanikawa, C., Yamamoto, T., Yagi, M., Takada, K., 2010. between young adults and orthodontists. Am. J. Orthod. Dentofac.
Automatic recognition of anatomic features on cephalograms of Orthop. 145, 750-756.
preadolescent children. Angle Orthod. 80, 812-820.
40. Yu, X., Liu, B., Pei, Y., Xu, T., 2014. Evaluation of facial
34. Tong, W., Nugent, S.T., Jensen, G.M., Fay, D.F., 1989. An attractiveness for patients with malocclusion: A machine-learning
algorithm for locating landmarks on dental X-rays. Images twenty- technique employing Procrustes. Angle Orthod. 84, 410-416.
first century. Proc. Annu. Int. Eng. Med. Biol. Soc. 552-554.
41. Zarei, A., El-Sharkawi, M., Hairfield, M., King, G., 2006. An
35. Vecsei, B., Joós-Kovács, G., Borbély, J., Hermann, P., 2017.
intelligent system for prediction of orthodontic treatment outcome.
Comparison of the accuracy of direct and indirect three- 2006 IEEE Int. Jt. Conf. Neural Netw. Proc. 2702-2706.
dimensional digitizing processes for CAD/CAM systems an in
vitro study. J. Prosthodont. Res. 61, 177-184.
162