|

Received: 6 June 2021    Accepted: 29 June 2021

DOI: 10.1111/ocr.12517

REVIEW ARTICLE

Artificial intelligence in orthodontics: Where are we now?

A scoping review

Anna Monill-­González1  | Laia Rovira-­Calatayud1  | Nuno Gustavo d’Oliveira2  |

Josep M. Ustrell-­Torrent3

1
Faculty of Medicine and Health Sciences,
University of Barcelona, Barcelona, Spain Abstract
2
Department of Odontostomatology -­ Objective: This scoping review aims to determine the applications of Artificial
Orthodontics. Coordinator of the Master
Intelligence (AI) that are extensively employed in the field of Orthodontics, to eval-
of Orthodontics, Faculty of Medicine and
Health Sciences, University of Barcelona, uate its benefits, and to discuss its potential implications in this speciality. Recent
Barcelona, Spain
decades have witnessed enormous changes in our profession. The arrival of new
3
Department of Odontostomatology -­
Orthodontics, Oral Health and Masticatory
and more aesthetic options in orthodontic treatment, the transition to a fully digital
System Group (IDIBELL), University of workflow, the emergence of temporary anchorage devices and new imaging methods
Barcelona, Barcelona, Spain
all provide both patients and professionals with a new focus in orthodontic care.
Correspondence Materials and methods: This review was performed following the Preferred
Nuno Gustavo Correia d’Oliveira,
Department of Odontostomatology -­
Reporting Items for Systematic reviews and Meta-­Analyses extension for Scoping
Orthodontics, Faculty of Medicine and Reviews (PRISMA-­ScR) guidelines. The electronic literature search was performed
Health Sciences, University of Barcelona,
Carrer Feixa Llarga s/n, Pavelló Govern,
through MEDLINE/PubMed, Scopus, Web of Science, Cochrane and IEEE Xplore da-
Despatx 2.7, 08907 L’Hospitalet de tabases with a 11-­year time restriction: January 2010 till March 2021. No additional
Llobregat, Barcelona, Spain.
Emails: [email protected];
manual searches were performed.
[email protected] Results: The electronic literature search initially returned 311 records, and 115 after
removing duplicate references. Finally, the application of the inclusion criteria re-
sulted in 17 eligible publications in the qualitative synthesis review.
Conclusion: The analysed studies demonstrated that Convolution Neural Networks
can be used for the automatic detection of anatomical reference points on radio-
logical images. In the growth and development research area, the Cervical Vertebral
Maturation stage can be determined using an Artificial Neural Network model and
obtain the same results as expert human observers. AI technology can also improve
the diagnostic accuracy for orthodontic treatments, thereby helping the orthodontist
work more accurately and efficiently.

KEYWORDS

artificial intelligence, machine learning, orthodontics, review

This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium,
provided the original work is properly cited.
© 2021 The Authors. Orthodontics & Craniofacial Research published by John Wiley & Sons Ltd.

Orthod Craniofac Res. 2021;00:1–10.  |

wileyonlinelibrary.com/journal/ocr     1
 |
2       MONILL-­GONZÁLEZ et al.

1 |  I NTRO D U C TI O N data. Its purpose is to build a neural network that can automati-
cally recognize patterns to improve feature detection. 2,3
The last decades have witnessed enormous changes in our profes- • Big data refers to large data sets and/or the combination of all
sion. The arrival of new and more aesthetic options in orthodontic available data points drawn from multiple sources which can be
treatment, the transition to the fully digital workflow, the emer- used to recognize patterns that inform a customized experience
gence of temporary anchorage devices and new imaging methods all for different individuals.1
work to provide both patients and professionals with a new focus in
orthodontic care.1 Orthodontic treatments are usually long procedures with an
To make the diagnostic process more accurate and efficient, the average treatment duration of nearly 29 months,4 which is why or-
use of Artificial Intelligence (AI) in orthodontics has grown signifi- thodontists must become more efficient to adapt to the needs of so-
cantly in recent years. This knowledge is fundamental for predict- ciety. The application of ML techniques can help to solve this issue.
ing treatment prognosis. However, the addition of this AI-­based Recent technological innovations in orthodontics, including cone
knowledge does not change the fact that the health professionals, beam computed tomography (CBCT) and 3D visualizations, intraoral
with their own knowledge gained through specialized education and scanners, facial scanners, instant teeth modelling software capabili-
years of experience, are the ones that ultimately have to diagnose ties and new appliance developments using robotics and 3D printing,
and determine the best treatment plan. Nevertheless, AI can be use- are changing the face of medical care and are quickly becoming in-
ful when making specific clinical decisions in a limited time. AI ap- tegrated into dentistry.5 These tools enable a better understanding
plications can guide clinicians to make better decisions and perform of the patient's anatomy and are able to create dynamic anatomical
better, because the results obtained from AI are highly accurate and reconstructions for the specific patient, and therefore accommodate
therefore, in some cases, can prevent human errors. 2 the possibility of 3D treatment planning. Convolutional neural net-
To appreciate the impact of AI on orthodontics, it is first import- works (CNNs) are increasingly applied for medical image diagnostics,
ant to discern some key terms related to AI: most frequently for the detection, segmentation or classification
of anatomical structures. Deep learning has also recently been
• AI’s main objective is to offer a machine the ability to have its own used for geometric feature learning and classification.6 Machine-­
intelligence. Put another way, AI aims for a machine to be able to learning approaches, which are algorithms trained to identify pat-
learn through data, to solve problems by itself. terns in large data sets, are ideally suited to facilitate data-­driven
• Machine learning (ML) is the main backbone of AI. It depends on decision-­making.7
algorithms to predict outcomes based on data sets and draws in- This scoping review aims to determine the applications of AI that
fluence from many research disciplines. Its purpose is to facilitate are extensively employed in the field of orthodontics, to evaluate
machines to learn from data so they can resolve issues without the benefits of AI and to discuss its potential implications in this
human input. The most commonly used techniques of ML include speciality.
the support vector machine (SVM), logistic regression (LR), naive
Bayesian classifier, decision tree (DT), random forest (RF), ex-
treme learning machine (ELM), fuzzy k-­nearest neighbour (FKNN) 2 | M ATE R I A L S A N D M E TH O DS
and convolution neural network (CNN). 2,3
• Neural networks are a set of algorithms that calculate signals 2.1 | Protocol
through artificial neurons that try to imitate the functioning of
human neurons. This review was performed following the Preferred Reporting Items
• Deep learning is an integral part of ML. It uses networks with dif- for Systematic reviews and Meta-­Analyses extension for Scoping
ferent computer layers in deep neural networks to analyse input Reviews (PRISMA-­ScR) guidelines.8 A pilot search of MEDLINE (via

TA B L E 1   Description of PICO
What is the applicability of Artificial Intelligence in the field of
elements
Study question Orthodontics?

Population Patients’ diagnostic images (orthopantomography, cephalometric

radiographs, intraoral radiographs, CBCT† , clinical images, facial
images and 3D model images).
Intervention Artificial intelligence-­based forms of diagnosis and treatment planning.
Comparison Reference standards and existing literature.
Outcome Measurable or predictive outcomes such as accuracy, sensitivity and
specificity.

Abbreviations: C, Comparison; I, Intervention; O, Outcome; P, Population.

†
Cone beam computed tomography.
MONILL-­GONZÁLEZ et al. |
      3

PubMed) was conducted to prepare the study protocol. The data ex- 3 | R E S U LT S
traction forms were constructed after the initial results of the pilot
search. 3.1 | Search and study selection
The search was based on the PICO (problem/patient/population,
intervention/indicator, comparison and outcome) elements (Table 1). The flowchart of the articles conforming to the PRISMA-­ScR and
included in this scoping review study selection is shown in Figure 1.
The electronic literature search initially returned 311 records,
2.2 | Literature search which was reduced to 115 after removing duplicate references.
After reviewing the titles and abstracts, all 115 studies were exam-
The electronic literature search was performed through MEDLINE/ ined in more detail. Two articles were excluded as their full text was
PubMed, Scopus, Web of Science, Cochrane and IEEE Xplore data- not available. Ninety records were excluded because they did not
bases between November 2020 and March 2021. meet the selection criteria, and no additional studies were found
A specific combination of words was introduced in order to com- by manual reference search. Finally, the application of the inclusion
plete a specific and reproducible search (Table 2). No additional man- criteria resulted in 17 eligible publications in the qualitative syn-
ual searches were performed. thesis review. There was a complete consensus among the evalua-
tors on the literature selection process and the classification of the
publications.
2.3 | Eligibility criteria Of the 17 studies included in this scoping review (Table 3), four
publications evaluated the use of AI in the diagnosis of surgery/
First, search engine results were evaluated for relevance based on non-­surgery decision and extraction choice. The determination of
their title and abstract. The studies whose titles or abstracts con- cervical vertebrae stages for growth and development periods with
tained different information that was not related to the study ques- ML was evaluated in two publications. Five publications evaluated
tion were excluded. An 11-­year restriction was determined, from the accuracy of the automatic detection of anatomical reference
January 2010 to March 2021, to ensure the review was based on points on lateral cephalometric images. The prediction of orthodon-
the most up-­to-­date information. Only fully available articles were tic treatment needs with an automatic orthodontic diagnosis was
considered. Articles focused on AI in the field of orthodontics were tested in two publications. The accuracy of automatic tooth seg-
included. Only those publications that used some predictive measur- mentation was assessed in two publications. One publication anal-
able outcomes such as accuracy, sensitivity and specificity, and those ysed the maxillary structure variation in unilateral canine impaction.
with adequate documentation of the data sets they employed, were Lastly, one publication quantified the 3D asymmetry of the maxilla
considered. All relevant publications and studies whose abstracts did in patients with unilateral cleft lip and palate.
not provide enough information to justify an exclusion decision were
obtained in full text to determine their eligibility. Articles wrote in any
language other than English, Spanish, Portuguese, Italian, German or 3.2 | Outcome domains of included studies
French were excluded, as well as studies related to non-­AI areas.
Considering the selected articles, a total of 472 lateral cephalometric
radiographs were used in two of the studies to analyse the accuracy
2.4 | Results extraction of using neural network ML to decide whether to use extractions to
reduce discrepancy in different orthodontic malocclusions.7,9 Jung
Table 3 depicts how we collected select information from the included et al (2016)7 reported an accuracy of 84%-­93% and Choi et al (2019)9
studies. The type of ML method, the number and type of images used noted an intra-­class correlation coefficient of 0.97-­0.99.
for testing AI software, the accuracy of the technique, and its benefits One study evaluated the use of a CNN in automatic cephalomet-
to the field of orthodontics were extracted from the articles. ric analysis. It demonstrated an accuracy of 88.43% for a total of

TA B L E 2   Electronic literature search strategy

Filters in Included
Database Keywords Time frame database Result articles

MEDLINE/ (‘Orthodontic*’) AND November 2020—­March 2021 Full text 91 17

Pubmed (‘machine learning’ OR ‘unsupervised Machine Since 2010
Scopus Learning’ 104
OR ‘supervised Machine Learning’
IEEE Xplore OR ‘artificial intelligence’ 22
Cochrane OR ‘Artificial life’ OR ‘deep learning’) 3
Web of Science 91
TA B L E 3   Summary of the included studies
|

Author
4      

Origin Major Method No. of images

Year Aim (Algorithm) for testing Type of image for testing Accuracy Benefits to Orthodontics
7
Jung et al Diagnosis of extractions using neural network 2-­LNN 156 Lateral Cephalometric 93% -­84% The AI expert system could be a
Korea machine learning. Radiograph reference for less-­experienced
2016 practitioners to evaluate the
need for extractions.
Choi et al9 Diagnosis of surgery/non-­surgery decision and 2-­LNN 316 Lateral Cephalometric ICC: 0.97-­0.99 The AI model using neural
Korea extraction determination. Radiograph network machine learning could
2019 be applied for the diagnosis of
orthognathic surgery cases.
Kim et al10 Automatic cephalometric analysis. CNN 400 Lateral Cephalometric 88.43% The AI expert system could be
Korea Radiograph used to automatically identify
2020 cephalometric landmarks with
high accuracy immediately.
Dobratulin et al11 Automatic detection of anatomical reference CNN (U-­Net) 100 Lateral Cephalometric 92% It solves the problem of detecting
Russia points on radiological images of the head profile. Radiograph anatomical reference points on a
2020 radiological image of the head in
a lateral projection.
Lee et al12 Automatic location of cephalometric landmarks BN 400 Lateral Cephalometric 90.11% It provides cephalometric
Korea with confidence regions. Radiograph landmarks and their confidence
2020 regions, which could be used as
a computer-­aided diagnosis and
educational tool.
Kim et al13 Automated landmark identification for CNN 430 CBCT images 80.4% Automated identification
Korea posteroanterior (PA) cephalometric landmarks. for CBCT-­synthesized PA
2021 cephalometric landmarks shows
better consistency than manual
identification, although does not
adequately achieve the clinically
acceptable error range of less
than 2 mm.
Kök et al14 Determination of cervical vertebrae stages for K-­NN, NB, Tree, 300 Lateral Cephalometric 77.02% AI algorithms can be used
Turkey growth and development periods. ANN, SVM, RF, LR Radiograph for diagnostic purposes in
2019 orthodontics where growth
development needs to be
determined.
Amasya et al15 Cervical vertebral maturation (CVM) analysis. CDSS 647 Lateral Cephalometric 58.3% Automatic classification of
Turkey Radiograph CVM with AI may replace
2020 conventional evaluation
methods.
MONILL-­GONZÁLEZ et al.

(Continues)
TA B L E 3   (Continued)

Author
Origin Major Method No. of images
Year Aim (Algorithm) for testing Type of image for testing Accuracy Benefits to Orthodontics

Guo et al16 To estimate human age based on a sample of CNN 10,257 Orthopantomograms 94.15% CNN models can surpass humans
China dental orthopantomographies. in age classification.
MONILL-­GONZÁLEZ et al.

2021
Li et al17 Automatic tooth root segmentation algorithm of CNN (RNN) 361 CBCT images 95.8% -­95.3% The automatic segmentation
China CBCT axial image based on deep learning. of individual tooth root has
2020 potential to improve the
segmentation efficiency and
accuracy.
Sun et al18 Automatic and accurate segmentation and CNN 100 3D digital dental casts 97% It achieves performance
China identification of individual teeth from digital improvements compared
2020 dental casts. with the state-­of-­the-­art in
both tooth segmentation and
identification tasks.
Thanathornwong Predicting the need for orthodontic treatment in BN 1000 Data sets 93% -­95% It achieved a high degree of
et al19 patients with permanent dentition. accuracy in classifying patients
Thailand into groups needing and not
2018 needing orthodontic treatment.
Murata et al20 Automatic orthodontic diagnostic imaging CNN 704 Facial photographs 64.8% It reduces doctor's assessment
Japan system. workload. It improves the
2017 accuracy in diagnosis and also
increases the number of facial
parts to be assessed.
Shin et al21 Predicting the need for orthognathic surgery of RNN 840 Lateral and frontal 95.4% A deep learning program
Korea skeletal malocclusion using cephalogram. Cephalometric Radiograph can determine the need for
2021 orthognathic surgery with
relative accuracy, helping
oral, maxillofacial surgeons,
orthodontists, and general
dentists to make standardized
decisions.
Lin et al22 To determine the cephalometric predictors of the XGBoost 56 Lateral Cephalometric 87.4% At the age of 6 years, it is possible
Korea future need for orthognathic surgery in patients Radiograph to predict the future need
2021 with repaired unilateral cleft lip and palate for surgery to correct sagittal
(UCLP). skeletal discrepancy in UCLP.
|
      5

(Continues)
 6       | MONILL-­GONZÁLEZ et al.

400 lateral cephalometric radiographs.10 Following the same line of

The deep learning-­based protocol

could automatically segment the

and alveolar crest area near the

clinical data sets to be analysed

Correlation Coefficient; K-­NN: K-­nearest neighbours; LR: Logistic Regression; NB: Naive Bayes; RF: Random Forest; RNN: Residual Neural Network; SVM: Support Vector Machine; Tree: Decision Tree;
exists in the pyriform aperture
research, 500 radiological images of the head profile were used in

of the maxilla on the cleft side

maxilla. Significant hypoplasia

Fast and efficient CBCT image

Abbreviations: 2-­LNN: 2-­layer Neural Network; ANN: Artificial Neural Network; BN: Bayesian Network; CDSS: Clinical Decision Support System; CNN: Convolutional Neural Network; ICC: Intraclass
segmentation will allow large
two articles to study the viability of automatic detection of anatom-
Benefits to Orthodontics ical reference points on radiological images using a CNN (U-­Net)11
and Bayesian network.12 The accuracies reported in these studies
were 90.11% and 92%, respectively.11,12 Kim et al (2021) performed

effectively.
the same analysis but used 430 CBCT images instead, and concluded
that automated identification was more consistent than manual

defect.
identification.13
There were two studies that evaluated the Cervical Vertebral
Maturation (CVM) analysis using AI algorithms. Kök et al (2019)14 re-
ported a mean accuracy of 77.02%, whereas the accuracy reported
ICC: 0.994
ICC >0.90
Accuracy

by Amasya et al (2020)15 was 58.3%. Another way to estimate the

-­0.999

age of a person is to focus on the dental age, which Guo et al (2021)

did in their study, with an accuracy of 94.15%,16 finding that CNN
models were able to surpass humans in age classification.
Type of image for testing

An automatic tooth root segmentation algorithm for CBCT axial

images based on deep learning was studied by Li et al (2020)17 and
Sun et al (2020)18 with an accuracy of 97%18 and 95.8%–­95.3%.17
CBCT images

CBCT images

Both studies used a CNN. Li et al (2020)17 worked with CBCT im-

ages to test the algorithm whereas Sun et al (2020)18 used 3D digital
dental casts.
Alternatively, Thanathornwong et al (2018)19 and Murata
et al (2017)20 created automated diagnostic systems for orthodon-
No. of images

tic treatment. Thanathornwong et al (2018) worked with data sets

for testing

whereas Murata et al (2017) worked with facial photographs. The

accuracy of their respective methods was reported as 93%-­95%19
60

60

and 64.8%. 20 Similarly, Shin et al (2021)21 and Lin et al (2021)22 con-

cluded, with an accuracy of 95.4%21 and 87.4%, 22 respectively, that
a deep learning program can be used to determine the need for or-
Major Method

thognathic surgery. The latter publication also determined that it is

CNN (U-­Net)
(Algorithm)

possible to predict the future need for surgery to correct sagittal

skeletal discrepancy in patients with repaired unilateral cleft lip and
RF

palate at the age of 6 years. 22For those patients with unilateral cleft
lip and palate, AI can be also useful to segment the maxilla and quan-
tify its 3D asymmetry, as was demonstrated by Wang et al (2021)23
Assessment of maxillary structure variation in
Quantify the 3D asymmetry of the maxilla in
patients with unilateral cleft lip and palate.

with an intra-­class correlation coefficient (ICC) greater than 0.90.

Using a CNN, they also determined the existence of significant max-
illary hypoplasia on the cleft side of those patients.
unilateral canine impaction.

Note: Use of machine learning in orthodontics.

4 | D I S CU S S I O N
XGBoost: Extreme Gradient Boosting.

With the aim of achieving successful orthodontic treatments, having

detailed diagnoses, accurate treatment plans and accurate outcome
predictions is crucial. The research surveyed here has demonstrated
TA B L E 3   (Continued)

that AI technology helps the orthodontist to work more efficiently

Aim

and therefore to be more adapted to the needs of society.

To decide whether extractions are necessary prior to orthodon-
tic treatment, it would be useful to have a decision-­making expert
23

Chen et al29
Wang et al

system based on an artificial neural network (ANN). Xie et al (2010)24

Author
Origin

China

Pekin
2020

used an ANN system to determine whether an extraction or non-­

2021
Year

extraction treatment was best for malocclusion patients between 11

MONILL-­GONZÁLEZ et al.       7 |
F I G U R E 1   PRISMA-­ScR flowchart
IEEE Web of
of search strategy and selection of Pubmed Cochrane Scopus
Xplore Science

Identification
articles (n= 91) (n= 3) (n= 104)
(n= 22) (n= 91)

Records idenfied in total

(n= 311)

Screening
Records aer duplicates removed
(n= 115)

Full-text arcles assessed Records excluded aer

for eligibility applying the inclusion and
(n= 113) exclusion criteria (n= 90)
Eligibility

Studies included in
Full-text arcles excluded
qualitave synthesis
(n= 6)
(n= 23)
Included

Studies included in
qualitave synthesis
(n= 17)

and 15  years old, and found the ANN worked with 80% accuracy. accuracy and demonstrated a more isotropic form of detection er-
These results were similar to the studies by Jung et al (2016)7 and rors than did SSD. Hwang et al (2020)27 concluded that AI cephalo-
9
Choi et al (2019). metric landmarks identification is as accurate as human examiners.
However, it is important to remember that there is no singularly In the same way, Kim et al (2020),10 Dobratulin et al (2020)11 and
7
correct answer for the diagnosis of extractions. Generally, most or- Lee et al (2020)12 determined, with an accuracy between 88% and
thodontists decide whether an extraction is necessary based on their 92%, that the AI expert system could be used to automatically iden-
experience and knowledge by analysing data from their patients' tify cephalometric landmarks. Guo et al (2021)16 also concluded that
clinical evaluation, photographs, dental casts and radiographs. One a deep learning technique without human interference can effec-
problem is that this often causes intra-­ and inter-­clinician variabil- tively overcome the limitations associated with manual methods of
ity in the treatment planning process. 25 By mimicking the decision-­ identification.
making of human experts, an AI expert system could be developed AI has also been used to automatically identify and classify skel-
based on various philosophies of diagnosis to assist the decision-­ etal malocclusions from 3D CBCT craniofacial images. In 2020, Kim
making process.7 Nevertheless, the final decision will always belong et al proposed a method that aimed to assist orthodontists in de-
to the clinicians. termining the best treatment path for the patient, be it orthodontic
Various studies have been conducted to demonstrate the ef- treatment, surgical treatment, or a combination of both. 28 Fast and
ficacy of AI applications in identifying cephalometric landmarks. efficient CBCT image segmentation would allow for large clinical
The diagnostic value of the analysis depends on the accuracy and data sets to be analysed effectively. 29 ML can help to determine
the reproducibility of landmark identification. In orthodontic prac- the cephalometric predictors of the future need for orthognathic
tice, lateral cephalometry has been widely used for skeletal classi- surgery, as in patients with repaired unilateral cleft lip and palate
fication and treatment planning. The incorporation of a CNN can (UCLP).30 Thus, the use of AI definitely reduces doctor assessment
12
provide an accurate and robust skeletal diagnostic system. Park workload and improves diagnostic accuracy. 20
26
et al (2019) compared two of the latest deep learning methods in The assessment of bone age and skeletal maturity and its com-
their study: You-­Only-­Look-­Once version 3 (YOLOv3) and the Single parison to chronological age is an important task for the diagnosis of
Shot Multibox Detector (SSD). YOLOv3 showed higher diagnosing paediatric endocrinology, orthodontics and orthopaedic disorders.
 |
8       MONILL-­GONZÁLEZ et al.

Because this assessment is a time-­consuming activity that may be 4.1 | Limitations

affected by inter-­ and intra-­rater variability, the use of methods
that can automate it, like ML techniques, can be of great value.31 This review presents two main limitations:
Growth and development can be determined by cervical vertebrae First, being a scoping review, the review question has to be more
stages (CVS), which can be predicted/classified using different AI al- generally defined when compared to a systematic review. Whereas
gorithms. Kök et al (2019)14 compared seven AI algorithms that are scoping reviews assess where consolidated knowledge ends and
frequently used in the field of classification: K-­nearest neighbours additional research is needed, systematic reviews clarify whether
(k-­NN), Naive Bayes (NB), decision tree (Tree), artificial neural net- existing knowledge is reliable.33 AI embraces many different fields
works (ANN), support vector machine (SVM), random forest (RF) and and applications, and therefore, it adjusts with the aim of a scoping
logistic regression (LR) algorithms. They concluded that k-­NN and review, which is to provide an overview of the evidence.
LR algorithms had the lowest accuracy values, whereas SVM-­RF-­ Second, the search was limited to the last 11 years, because the
Tree and NB algorithms had variable accuracy values, and the ANN authors agreed that it would be more useful to describe only the
would be the preferred method for determining CVS.14 Amasya H. latest applications of AI in orthodontics, rather than making an his-
15
et al (2020) developed an ANN model to determine skeletal age. torical review and thereby including obsolete technologies.
The developed ANN model performed close to, if not better than, Despite these limitations, the authors expect this to be a useful
human observers in CVM analysis. Repeatability and reproduc- overall introduction to understand the recent past of AI and the
ibility of the ANN model were in the range of human observers.15 actual present (as well as the near future) of its applications in or-
16
Guo et al (2021) concluded that deep learning techniques, with- thodontics. There is no doubt that there is still a long road ahead.
out human interference, can effectively overcome limitations of the Many of the results published in the papers used in this scoping
manual method in age classification based on panoramic images. review must be thoroughly and carefully analysed. However, those
Their CNN program focused on low-­density features around the who are already used to work with intraoral scanners and facial-­
teeth, instead of using the dental morphological traits that are typi- driven smile designs know exactly how limited and at the same
cally used by humans for age classification.16 time how useful all these new technologies are. Therefore, all the
Dental segmentation is one of the key steps in computer-­ tools available to the clinicians are of great value, and AI is one of
assisted orthodontic technology and its accuracy is closely related them.
to treatment outcome. This procedure requires precise positioning Nevertheless, the authors truly believe that despite all the future
and extraction of tooth shapes on the patient's 3D digital dental advancements in AI, it will never substitute human reasoning; how-
cast (or intraoral scan). ML using a CNN-­based model for tooth ever, it will definitely help.
segmentation and identification achieved performance improve-
ments when compared with the state-­of-­t he-­art general mesh seg-
mentation method for both tooth segmentation and identification 5 | CO N C LU S I O N S
tasks.18
Deep learning systems work in distinct areas of orthodontics. The analysed studies demonstrated that CNNs can be used for the
Orthodontists can use AI systems as an ancillary tool for increas- automatic detection of anatomical reference points on radiological
ing the accuracy of diagnosis, treatment planning and for predict- images. For growth and development areas, the CVM can be deter-
ing treatment outcomes. Automated systems can save a lot of time mined using an ANN model and obtain the same results as human
and increase the efficiency of the clinicians. For example, the use of observers. AI technology can also help improve the accuracy of di-
automated cephalometric points identification or automated teeth agnoses for orthodontic treatment, therefore, helping orthodontists
segmentation to enable a treatment preview outcome helps reduce work more efficiently. However, although the improvement of AI is
orthodontic treatment planning times.5,10-­13 Additionally, with deep definitely a great help for orthodontists and other health profession-
learning techniques it is possible to eliminate the subjectivity asso- als, the final decisions on health matters will always be the clinicians'
ciated with human decision-­making; traditional manual methods are responsibility.
likely to incorporate a relatively higher degree of intra-­ and inter-­
observer errors due to that subjectivity, which can lead to an increase AC K N OW L E D G E M E N T S
in the prediction error.32 Likewise, these systems can be used for The authors received no financial support for this work.
secondary opinions, in order to improve the accuracy of diagnosis.
Nevertheless, clinicians should always trust their clinical judgment C O N FL I C T O F I N T E R E S T
above all. The authors declared no conflict of interest.
AI could become a valuable tool to use in those procedures that
require high precision and are more time consuming, such as indi- AU T H O R C O N T R I B U T I O N S
rect bonding, precise Bolton Analysis or wire bending, in order to Monill-­González A. and Rovira-­C alatayud L performed equally with
increase the quality of the treatments we offer to our patients. the study selection, data extraction and data presentation. They also
MONILL-­GONZÁLEZ et al. |
      9

completed the initial manuscript draft and data analysis. d’Oliveira 14. Kök H, Acilar AM, İzgi MS. Usage and comparison of artificial in-
telligence algorithms for determination of growth and develop-
NG conceptualized the study and resolved disagreements in study
ment by cervical vertebrae stages in orthodontics. Prog Orthod.
selection. He analysed all data and prepared the presentation of the 2019;20(1):41. https://progr​essin​ortho​donti​c s.sprin​gerop​en.com/
final manuscript. Ustrell-­Torrent JM supervised the research activ- artic​les/10.1186/s4051​0 -­019-­0295-­8
ity and provided the necessary resources to conduct the review. All 15. Amasya H, Cesur E, Yıldırım DOK. Validation of cervical vertebral
authors contributed to critical revision of the article. All authors read maturation stages: artificial intelligence vs human observer visual
analysis. Am J Orthod Dentofac Orthop. 2020;158(6):173-­179.
and approved the final manuscript.
16. Guo Y-­C , Han M, Chi Y, et al. Accurate age classification using man-
ual method and deep convolutional neural network based on ortho-
DATA AVA I L A B I L I T Y S TAT E M E N T pantomogram images. Int J Legal Med. 2021;135(4):1589-­1597.
The data that support the findings of this study are available from 17. Li Q, Chen K, Han L, Zhuang Y, Li J, Lin J. Automatic tooth roots seg-
mentation of cone beam computed tomography image sequences
the corresponding author upon reasonable request.
using U-­net and RNN. J Xray Sci Technol. 2020;28:905-­922.
18. Sun D, Pei Y, Song G, et al. Tooth Segmentation and Labeling from
ORCID Digital Dental Casts. In: 2020 IEEE 17th International Symposium on
Anna Monill-­González  https://orcid.org/0000-0002-7182-3827 Biomedical Imaging (ISBI)IEEE; 2020:p. 669-­673
19. Thanathornwong B. Bayesian-­based decision support system for
Laia Rovira-­Calatayud  https://orcid.org/0000-0002-1764-8597
assessing the needs for orthodontic treatment. Health Inform Res.
Nuno Gustavo d’Oliveira  https://orcid.org/0000-0002-8507-4126 2018;24(1):22-­28.
Josep M. Ustrell-­Torrent  https://orcid.org/0000-0001-6435-5220 20. Murata S, Lee C, Tanikawa C, Date S. Towards a fully automated diag-
nostic system for orthodontic treatment in dentistry. Proc -­13th IEEE
Int Conf eScience, eScience. 2017;1-­8.
REFERENCES
21. Shin W, Yeom H-­G , Lee GH, et al. Deep learning based prediction
1. Allareddy V, Rengasamy Venugopalan S, Nalliah RP, Caplin JL, Lee of necessity for orthognathic surgery of skeletal malocclusion using
MK, Allareddy V. Orthodontics in the era of big data analytics. cephalogram in Korean individuals. BMC Oral Health. 2021;21(1):1-­7.
Orthod Craniofac Res. 2019;22(Suppl 1):8-­13. 22. Lin G, Kim PJ, Baek SH, Kim HG, Kim SW, Chung JH. Early pre-
2. Khanagar SB, Al-­ehaideb A, Maganur PC, et al. Developments, ap- diction of the need for orthognathic surgery in patients with re-
plication, and performance of artificial intelligence in dentistry –­ A paired unilateral cleft lip and palate using machine learning and
systematic review. J Dent Sci. 2021;16(1):508-­522. longitudinal lateral cephalometric analysis data. J Craniofac Surg.
3. Jordan MI, Mitchell TM. Machine learning: trends, perspectives, 2021;32(2):616-­620.
and prospects. Science. 2015;349:245-­260. 23. Wang X, Pastewait M, Wu T, et al. 3D morphometric quantifica-
4. Jung MH. Factors influencing treatment efficiency: a prospective tion of maxillae and defects for patients with unilateral cleft palate
cohort study. Angle Orthod. 2021;91(1):1-­8. via deep learning-­based CBCT image auto-­segmentation. Orthod
5. Nayyar N, Ojcius DM, Dugoni AA. The role of medicine and tech- Craniofac Res. 2021;00:1-­9.
nology in shaping the future of oral health. J Calif Dent Assoc. 24. Xie X, Wang LWA. Artificial neural network modeling for deciding
2020;48(3):127-­130. if extractions are necessary prior to orthodontic treatment. Angle
6. Schwendicke F, Golla T, Dreher M, Krois J. Convolutional neural Orthod. 2010;80(2):262-­266.
networks for dental image diagnostics: a scoping review. J Dent. 25. Dunbar AC, Bearn D, McIntyre G. The influence of using digital di-
2019;91:103226. agnostic information on orthodontic treatment planning -­ a pilot
7. Jung SK, Kim Ansan TW. New approach for the diagnosis of ex- study. J Health Eng. 2014;5(4):411-­428.
tractions with neural network machine learning. Am J Orthod 26. Park JH, Hwang HW, Moon JH, et al. Automated identification
Dentofac Orthop. 2016;149(1):127-­133. of cephalometric landmarks: Part 1-­Comparisons between the
8. Tricco AC, Lillie E, Zarin W, et al. PRISMA extension for scoping latest deep-­learning methods YOLOV3 and SSD. Angle Orthod.
reviews (PRISMA-­ScR): checklist and explanation. Ann Intern Med. 2019;89(6):903-­909.
2018;169:467-­473. 27. Hwang HW, Park JH, Moon JH, et al. Automated identification of
9. Choi HI, Jung SK, Baek SH, et al. Artificial intelligent model with cephalometric landmarks: Part 2-­Might it be better than human?
neural network machine learning for the diagnosis of orthognathic Angle Orthod. 2020;90(1):69-­76.
surgery. J Craniofac Surg. 2019;30(7):1986-­1989. 28. Kim I, Misra D, Rodriguez L, et al. Malocclusion classification on 3D
10. Kim H, Shim E, Park J, Kim YJ, Lee U, Kim Y. Web-­based fully au- cone-­beam CT craniofacial images using multi-­channel deep learn-
tomated cephalometric analysis by deep learning. Comput Methods ing models. Annu Int Conf IEEE Eng Med Biol Soc. 2020;1294-­1298.
Programs Biomed. 2020;194:105513. 29. Chen S, Wang L, Li G, et al. Machine learning in orthodontics: intro-
11. Dobratulin K, Gaidel A, Aupova I, Ivleva A, Kapishnikov A, Zelter P. ducing a 3D auto-­segmentation and auto-­landmark finder of CBCT
The efficiency of deep learning algorithms for detecting anatomical images to assess maxillary constriction in unilateral impacted ca-
reference points on radiological images of the head profile. arXiv. nine patients. Angle Orthod. 2020;90(1):77-­8 4.
2020;01135(18):0-­5. 3 0. Lin G, Kim PJ, Baek SH, Kim HG, Kim SW, Chung JH. Early pre-
12. Lee JH, Yu HJ, Kim MJ, Kim JW, Choi J. Automated cephalometric diction of the need for orthognathic surgery in patients with re-
landmark detection with confidence regions using Bayesian convo- paired unilateral cleft lip and palate using machine learning and
lutional neural networks. BMC Oral Health. 2020;20(1):1-­10. longitudinal lateral cephalometric analysis data. J Craniofac Surg.
13. Kim MJ, Liu Y, Oh SH, Ahn HW, Kim SH, Nelson G. Evaluation of 2021;32(2):616-­620.
a multi-­stage convolutional neural network-­based fully automated 31. Dallora AL, Anderberg P, Kvist O, Mendes E, Ruiz SD, Berglund
landmark identification system using cone-­beam computed tomog- JS. Bone age assessment with various machine learning tech-
raphysynthesized posteroanterior cephalometric images. Korean J niques: a systematic literature review and meta-­analysis. PLoS ONE.
Orthod. 2021;51(2):77-­85. 2019;14(7):1-­22.
 |
10       MONILL-­GONZÁLEZ et al.

32. Jiménez LG, Palacio LAV, De LS, Vasquez YR, Capellán MC,
Cameriere R. Validation of the third molar maturity index (I3M): How to cite this article: Monill-­González A, Rovira-­Calatayud L,
Study of a Dominican Republic sample. J Forensic Odontostomatol. d’Oliveira NG, Ustrell-­Torrent JM. Artificial intelligence in
2019;37(3):27-­33.
orthodontics: Where are we now? A scoping review. Orthod
33. Galler K, Grätz E-­M, Widbilller M, Buchallla W, Knüttel H.
Pathophysiological mechanisms of root resorption after den- Craniofac Res. 2021;00:1–­10. https://doi.org/10.1111/ocr.12517
tal trauma: a systematic scoping review. BMC Oral Health.
2021;21(1):163.