TITLE PAGE

Title

Artificial Intelligence and Orthodontics : Defining A Perspective

Authors: Anmol S Kalha 1, Shashidara R 2

Affiliation:

Anmol Kalha MDS

Professor Emeritus & Advisor Research and Policy

Coorg Institute of Dental Sciences

Maggula

Virajpet, Karnataka

India:

571218

Director Medeva Knowledge Systems, New Delhi, India

Shashidara R MDS

Associate Dean Research and AI Integration , Coorg Institute of Dental

Sciences, Virajpet, India

Correspondence:

Dr.Anmol S Kalha

Department of Orthodontics
Coorg Institute of Dental Sciences

Maggula

Virajpet, Karnataka

India: 571218

[email protected]; [email protected]

Running title: Artificial Intelligence in Orthodontics

Word Count: 3046

Number of figures, and tables: Figures – 1

Author Contributions:

(I) Conception and design: Both authors

(II) Administrative support: N/A
(III) Provision of study materials or patients: N/A
(IV) Collection and assembly of data: Both authors
(V) Data analysis and interpretation: Both authors
(VI) Manuscript writing: Both authors
(VII) Final approval of manuscript: Both authors
Abstract:

Artificial intelligence is the flavor of the season, it permeates every sphere of

human activity and thinking. Its impact seems asymmetric between fantasy and
reality. A gradual glance at published scientific papers on AI and healthcare in
2024 reveals around 25000 papers championing the cause of AI. A very
minuscule number critically analyse AI in healthcare and reflect on issues
regarding the use of AI. But it has arrived, and so will it change Orthodontics as
we know it and practice it, whether to betterment or detriment, this is a question
that only time will answer. We critically examine the justification of its
percolation into all spheres of orthodontic thinking and reasoning.

Keywords: 3~5 words

Artificial Intelligence , Machine Learning, Algorithms , Bias, Orthodontics

Introduction:

“It’s not artificial intelligence I’m worried about, it’s human stupidity.” –Neil
Jacobstein

AI as we know it in 2024 falls into three distinct categories: Narrow /weak AI,
general/strong AI and super intelligent AI. The narrow AI underlies most
common current AI systems, Internet searches, facial recognition and even the
Invisalign preferential clinician choice systems. They operate under a
predefined range or set of contexts. All current orthodontic applications of AI
come under this context. General type of AI is endowed with broad human-like
cognitive capabilities, enabling it to tackle new and unfamiliar tasks
autonomously possesses the capacity to discern, assimilate, and utilize its
intelligence to resolve any challenge without needing human guidance. Super
intelligent AI represents a future form of AI where machines could surpass
human intelligence across all fields, including creativity, general wisdom, and
problem-solving. Superintelligence is speculative and not yet realized. Current
AI systems are also classified as limited memory systems, which could help in
improved and informed decisions based on learning from past data. The theory
of mind and self-aware AI represent the future as computers develop sentience,
emotions and mimic human thought and reactions to develop beliefs and even
desires. Ethical dilemmas exist on the use of AI in healthcare. This article does
not aim to delve into the process behind AI applications or attempt to provide a
view into the multiple facets of AI use in orthodontics instead it will focus on
the most common areas of AI application in orthodontics and critique the same.
There exist serious ethical questions on the use of AI in healthcare ranging from
data safety to accessibility and questions of physician liability. However ethical
questions apart the moot question remains as to whether AI can be made all
pervasive and is there good data to support these assumptions, we will delve
into these issues here.
The Evolution of AI and its use in Healthcare and Orthodontics

We are in the age of big data, deep learning and artificial general intelligence as
AI permeates into each aspect of our lives be it in the form of Alexa ,SIRI or in
defining consumer preferences from movies to clothes and even food. The
journey however is of more than 75 years beginning with Warren McCulloch
and Walter Pits in 1943 proposing a model for artificial neurons, or the Turing
test designed by Alan Turing in 1950 to can check a machine's ability to exhibit
intelligent behaviour equivalent to human intelligence.

The first artificial neuron network (ANN) SNARC, was created in 1953 by
Marvin Minsky and Dean Edmonds utilizing 3,000 vacuum tubes to mimic a
network of 40 neurons. However, the term AI (Artificial intelligence) was
coined by the computer scientist john Mcarthy in 1956. The first chatbot ELIZA
soon followed in the year 1966 and the first intelligence robot WABOT-1 in
1972. Following two AI winters we are probably in the golden period of AI. (1)

The first AI based healthcare system was probably MYCIN in 1972 a system to
identify bacteria causing septicaemia and recommend antibacterials and dosage
for the same.(2)Dxplain, an expert system to diagnose diseases on a pattern of
symptoms was introduced in 1984.(3)However it was not until 2017 that the
FDA approved “ARTERYS “ a cloud based AI for medical imaging .

The relation between Dentistry, Orthodontics and AI is not as new as we

perceive with several documents circa 1966 speaking about AI in dentistry
though the AI as perceived then was more a buzzword for automation and use of
computers rather than AI as we perceive it today.(4)(5)

The knowledge based landmarking on cephalograms in 1986 was perhaps the

first attempt to use AI for an orthodontic diagnostic purpose albeit in the field of
radiological interpretation,(6) Age assessment needs and prediction of treatment
response have however been possible only recently.(7)
Broadly AI in healthcare can be separated into three different types based on
levels of automation and autonomy. (8)

 Assisted Intelligence: These systems assist the clinician in making

diagnosis and initiating suitable treatment, such systems do not learn or
make assumptions from the data provided.
 Augmented Intelligence: Wherein the AI augments clinical human
decision making and learns through its interaction with the human
process of clinical decision making. This type of AI is most used today
such as diagnostic and treatment algorithms.(9)
 Autonomous intelligence which represents systems which decide and act
independently and without need of human interaction.

Various iterations of these AI algorithms are used in orthodontics with claims of

performing better or as good as human experts.

Current utilization of AI in orthodontics:

1. Diagnosis:

As of May2024(10) there are 882 USFDA authorised medical devices a

majority of them (673) are in the field of radiology and only a small number of
less than 1% (5 devices) are related to dentistry (3 radiology , 1 orthodontics
(Cerec) and 1 Surgery (Xguide) whereas there exist an abundance of articles
and reviews on the utility of AI in orthodontics with a lot of promising new
areas of application. A majority of such published literature in in the field of
radiological analysis particularly of Lateral cephalograms.

Analysis of Cephalograms

A recent review of AI and machine learning in orthodontics has found that a

majority of published work in this filed in in the field of automatic landmark
detection in lateral cephalograms(11) and that more than 60% of the most cited
articles in this area are also in the field of automated landmark detection.(12)

The technique of automated analysis using AI is not new with the first such
algorithm being developed as far back as 1989.(13) (14) however in contrast to
traditional machine learning the emergence of CNN and ANN based AI analysis
of cephalograms has led to a emergence of this domain as the most utilized
domain of AI in orthodontics with several commercial applications being
available for the same.

The AI based cephalometric detection algorithms have greatly improved in

efficiency beginning at 70% accurate detection rates in 2015 to around 98% in
2022.(15) In comparison to manual tracing methods these methods have
demonstrated excellent reliability and reproducibility and Kunz et al have
demonstrated that automated tracing of cephalograms produced results no
different from 12 experts for 18 parameters in 1792 lateral cephalograms.(16)

Kim et al.(17,18), Tanikawa et al. ,Yao et al. have also demonstrated AI

algorithms which have shown comparable identification of cephalometric
landmarks with experts in the field demonstrating deviations within the ±2mm
which is considered clinically acceptable.(19) A added advantage is that most of
these algorithms are able to perform the analysis in a few seconds.(20)

The introduction of CBCT has led to the development of algorithms for

automated identification of relevant landmarks in 3D images with results similar
to the 2D cephalograms.(21,22)

These have led to the emergence of commercial applications such as CephX ,

Ceppro, AudaxCeph ,CephNinja and Web - Ceph for the purpose of automated
cephalometric analysis with acceptable results.(19)

Classification of Malocclusion
An interesting outcome of the AI based cephalometric analysis has been in the
field of forensic odontology where the landmarks have been tweaked and
support vector machines utilized to classify the type of skeletal malocclusion
with an accuracy 74.51%(23,24), whereas Yu et al. have demonstrated CNN
based models to classify skeletal malocclusion circumventing the need for
identification cephalometric landmarks with accuracy, sensitivity and specificity
of > 90% with vertical classification showing an accuracy of >96%(25) and
Kim et al. demonstrating multichannel deep learning algorithms to
automatically classify malocclusions based on CBCT images.(26) Ben
Hamadou have recently demonstrated interesting algorithms for tooth
localization, identification, and segmentation from various intraoral scanners
representing the next step to automated identification of dental malocclusion.
(27)

Determination of Skeletal age

Determination of accurate skeletal age is an important consideration in

orthodontic treatment planning. Extending the ability of AI algorithms to
identify and classify radiographic landmarks these have also been used to
determine skeletal age with various degrees of success.(28)Most of these
algorithms use the cervical vertebrae maturation stages to predict age with
accuracy ranging from 58% to 90 %(29)(28) with Aggregate Net output feature
being used to improve accuracy(30). Kook et al, in a comparison of seven
algorithms for cervical vertebral maturation have found that the ANN based
algorithms provided more accurate results.(31)

2. Personalized Treatment Planning

AI's ability to analyze vast datasets and derive actionable insights enables
personalized treatment planning tailored to each patient's unique needs thus this
would be an area of immense scope.

Determination of orthodontic treatment needs:

The decision-making for Orthodontic treatment is based on two indices i.e.

Index of Orthodontic Treatment Need (ITON) and Dental Aesthetic Index
(DAI). However these are often felt as tedious and complicated and hence the
attempt to replace AI based system. Thanathornwong utilized a Bayesian
network to create a clinical decision support system to help general practitioners
decide on the need for orthodontic treatment using data from 1000 patients, the
system provided good agreeability with the assessment of two expert
orthodontists in the study.(32) Facial photographs have been also utilized for
determining orthodontic treatment needs.(33)

Extraction vs Non-extraction Decisions:

Orthodontic extraction decisions are complex and based on many variables. The
matter gets more complex in the borderline cases as also the pattern of
extractions. Various attempts have been made to create prediction algorithms to
prove a clinical decision support system to the process of extraction decision.
The question of extraction vs non extraction based treatments is probably one of
the most perplexing questions in contemporary orthodontics and not
surprisingly this is the most researched in the domain of AI in orthodontic
treatment planning.

ANN have been developed with accuracy of 94% for extraction decision,
92.8% for anchorage pattern determination and 84% for extraction pattern
determination.(11) these algorithms have also been shown to identify important
parameters like crowding, overjet overbite etc which have a significant bearing
on the extract vs no extraction decision. (34)
Developing A Predictive Model For Probable Percentage Decisions for
Extraction / Non Extraction based on Pretreatment Variables:

The challenge of developing an ANN model for extraction prediction in a ethnic

population highlights the paucity of data that will constrain any efforts in this
direction unless a specified effort is made

Modelling Method: Binary logistic Regression

Accuracy of the Model: Overall 80%with a 90 % Extraction Prediction
Limitations: being a conceptual paper and less sample size, statistical
assumptions were violated in terms for significance and all the pre treatment
database used in the equation irrespective of significance.
Logistic regression transforms the dependent into a logit variable (natural log of
the odds of Y occurring or not occurring, which is ln(p/1-p)) and uses maximum
likelihood estimation (MLE) to estimate the coefficients.
Logistic regression estimates the probability of an event occurring, based on a
given dataset of independent variables. Since the outcome is a probability, the
dependent variable is bounded between 0 and 1
Event of interest here is Extraction and exposures are pre-treatment data
The Inclusion Criteria were :

Adult cases treated with multi banded fixed appliances

Complete pre and post treatment records including, cephalograms /photographs

Cases successfully presented and defended in the MDS exam

All craniofacial /skeletal anomalies and growth modifications excluded

Total no of cases included in the study = 200

Total No of cases treated with extractions = 150 , 75 in each ethnic group

Total No of Non Extraction cases = 50 , 25 in each ethnic group

Gender distribution = males 49%, females 51%

Frequency of Gender distribution

Non extraction: females = 64 % Males = 32%

Frequency of Extraction in Kerala sample = 63%

Frequency of Extractions in Karnataka Sample = 37 %

The cases had to meet the criteria of class 1 canine and molar occlusion, good
alignment, Overjet, overbite not exceeding 3 mm and alignment of incisors

The finished cases were reviewed by two independent orthodontists and then
the aesthetic outcomes were reviewed by two non-orthodontists for the aesthetic
and pleasing facial outcomes form the photographs at debond.

Objective: Developing a predictive model to predict the probability percentage

for extraction based on Pre Treatment Data

Variables Considered:

Dependent variable: Ground truth of Extraction and Non Extraction

Independent variables:

ANB PRE(degree)

OVERJET PRE(mm)

OVERBITE PRE(mm)

BJORK SUM PRE(degree)

U1-SN PRE(degree)

U1-NF(mm) PRE

IMPA PRE(degree)

LI-MP(mm) PRE

U1-L1 PRE(degree)
NASOLABIAL ANGLE PRE(degree)

UPPER INCISOR PROTRUSION PRE(U1 NA)-degree

LOWER INCISOR PROTRUSION PRE(L1 NB -degree)

In the Diagnosis of Extraction vs Non-Extraction the decision success rate was

an overall of 80 % with 93% in one Ethnic Group and 97% in the other
This was simple model for determining extractions only and was based on the
pretreatment measures considered relevant such as anteroposterior relationships,
vertical relationships, incisor prominence and associated soft tissue profile
parameters

Surgical Orthodontics

AI based Algorithms have also been used in orthodontic surgery largely for
computer assisted surgical simulation (CASS) and for prediction of orthodontic
surgical treatment outcomes(35) however there seem to be limited number of
studies into this domain according to recent reviews (36) , however even with
the caveat of limited number of studies these have demonstrated accuracy rates
of up to 94% using support vector machines.(11)(7) training of robots for
simulated orthodontic surgeries have also been attempted.(37)(38)

Treatment Follow-up

The Covid 19 pandemic was an important disruptor in the way health care was
delivered and has brought much attention and innovation in the field of tele-
medicine and tele dentistry.

Orthodontic treatment having an average treatment time of 20 months requires

multiple appointments for follow-up. Numerous tele-dentistry apps have been
developed for remote diagnosis and treatment advice. The Dental Monitoring
system (DMS) developed by Dental Minds & described by Caruso et al.(39)
describes an interesting union between tele-dentistry and knowledge based AI to
permit accurate semiautomated monitoring of orthodontic treatment from
pictures taken by the patient.(40) there is also a certain belief that next
generation orthodontics will be driven by virtual reality and augmented reality
providing patients with a more immersive experience.(41)

The elephant in the room: A Critique of AI in Orthodontics

AI has revolutionized the orthodontic field. Application of artificial intelligence

in the field of health care has revolutionized the accuracy in the diagnosis and
treatment planning. These systems have proven to be very efficient in
performing the tasks for which they have been designed. Orthodontics, these are
the conclusions of a majority of authors who have published work in the field of
AI in orthodontics. However, the inherent risk of the inaccuracy of these
assumptions because of the risk of various types of bias which get built into the
AI algorithms is largely regulated to a single sentence caveat. This probably is
because of the belief that AI is technology driven and data driven and hence is
free of human beliefs or errors.(42)

Bias risk refers to the potential of an AI system to produce unfair or

discriminatory results, often due to imbalances in the training data(43) and
biases do creep into the AI algorithm along three broad paths , through the data
provided (data bias) to the AI during the training phase or through errors which
occur in the design of the algorithm (algorithmic bias)(44)(45) as depicted
below
 BIAS IN AI IN
ORTHODONTI
CS

DATA ALGORITH COGNITI

BIAS MIC BIAS VE BIAS

MINORIT TRAINING
INFORMATIVN SERVING COHORT LABEL
Y BIAS SKEW BIAS BIAS
ESS BIAS

Fig 1 : Different types of Bias in AI in Orthodontics

Data biases could be a minority bias wherein

data from a section of population ex. Males
becomes the predominant data and hence might
give erroneous results in females or in a
mismatch between the training data sets of the
AI and the population in which it is finally used
again leading to erroneous results.
Algorithmic biases arise out of using flawed
training data can resulting in algorithms that
repeatedly produce errors, unfair outcomes, or
even amplify the bias inherent in the flawed data.
Programming errors can also cause algorithmic
bias.(46)

Cognitive bias can occur because of the expert or

the developer bringing in their own conscious or
unconscious beliefs into the algorithmic disease
making process.(47)

Most AI algorithms used in orthodontics do suffer

from shortcomings because of these biases, the
most researched domain of AI in orthodontics has
been in automated landmark identification in
lateral cephalograms with accuracy of unto 98% ,
however the fact remains that a lot of these
studies are based on unicentric data i.e. data
from a single centre , leading to what is called as
overfitting arising out of the model being
exposed to reduced operator variability, machine
calibration specifics etc leading to a biased
training data set and resulting in poor predictive
data when exposed to a real world scenario.(15)

These studies also suffer from diverse types of

algorithms being used and each study focussing
on relatively uniform populations. Hence the
commercially automated cephalometric analysis
algorithms have not been able to perform on par
with their training datasets in real world
scenarios resulting in poor landmark
identification/soft tissue tracing and
inconsistency of measurements, thus leading the
authors to advice caution in using the said
application in both the automated and the
semiautomated modes.(48)(49)

Age determination algorithms using CVM also

when tried in real life scenarios have also
suffered from limited number of expert readers
used to establish the gold standards and have
proven effective only in certain age ranges thus
requiring caution when interpreting AI-assisted
CVM assessment .(29)
The extraction vs non extraction algorithms have
largely ignored, important dental findings, such
as large dental fillings, periapical lesions,
periodontal damage, previous endodontic
treatment, and missing teeth, the other clinical
conditions of the patients have also been
ignored. They also suffer from a cognitive bias as
they have been created using a limited number
of experts and their treatment philosophies thus
creating an extremely narrow spectrum decision
making process.(34)(29)

Serious concerns also exist for the algorithms

developed to decide orthognathic surgery needs.
Three of the four algorithms developed for this
focus on a Korean population thus demonstrating
a data bias and the applicability of said algorithm
to other populations is questionable the
algorithms also demonstrated marked
heterogeneity in the methods used thus it would
not be possible to endorse any AI based model
for orthodontic treatment model and AI cannot
replace the clinical expertise of expert clinicians.
(34)(50)

The most obvious solution to these errors which

is recommended is to increase data set size and
heterogeneity and have a greater number of
experts assessing these data sets to setup gold
standards and eliminate bias. The need for these
datasets to be freely available to researchers and
AI development companies ensuring validation
on multicentric datasets would be a utopian
scenario.

However, the availability of such datasets would

raise serious questions about patient
confidentiality and ethics. Orthodontic records
present with an enhanced privacy risk because of
the sequential nature of record collections and
the presence of records such as photographs
would violate patient privacy and would be
viewed as invasion of privacy. There have been
suggestions for federated datasets however
these are still only conceptual.
This is the Achilles heel of AI in orthodontics and
probably the main reason why hardly any clinical
trials have been performed for validation of
algorithms developed from various studies.
Hence the superfluity of positive results
demonstrated by various studies might not be
valid or might not produce results of equal
quality in a real life scenarios and when
confronted with modified situations.

Conclusion :

As with any new technology that comes in there

seem to be unrealistic expectations from the
utilization of AI in orthodontics with proponents
often demonstrating a missionary zeal in
promoting the use of AI. However, the fact
remains that AI is not a magic bullet or panacea
to cure all the problems existing in orthodontics
or for that matter healthcare.(51)

While AI tools seem to perform exceptionally in

controlled scenarios the fact remains that many
AI powered tools are still in infancy and lack
extensive clinical validation raising serious
questions about their effectiveness in real world
settings.

The problem with AI in healthcare is in the

availability of genuine real time data, in the
absence of such data serious queries persist
regarding the use of AI in orthodontics. Thus it
would be fair to assume that AI will not be a
disruptor in replacing the clinical expertise and
professional experience of orthodontists in the
near future, however it may emerge as a tool
assist young practitioners and act as a cog
ensuring quality management in orthodontic
care.(33)

The lack of data will always dog the whole

concept of pattern recognition, building clinical
decision support systems , predicative and
prescriptive algorithms.This is where the
challenge is.

Footnote :
Funding: No sources of Funding
Acknowledgments

The Authors acknowledge Dr.Gautham Reddy and his team from the Department of

Orthodontics Coorg Institute of Dental Sciences ,Virajpet for their data regarding the

extraction vs non extraction algorithm

Conflict of Interest: Dr.Anmol S Kalha is a director of Medeva health sciences a AI and ML

solutions provider for healthcare.

Ethical Statement:

“The authors are accountable for all aspects of the work in ensuring that questions

related

to the accuracy or integrity of any part of the work are appropriately investigated and

resolved.”
References

1. History of Artificial Intelligence - Javatpoint

[Internet]. [cited 2024 May 18]. Available
from: https://www.javatpoint.com/history-of-
artificial-intelligence

2. Buchanan, Bruce G; Shortliffe EH. Rule-Based

Expert Systems: MYCIN [Internet]. Reading,
MA: Addisson Wesley. 1984 [cited 2024 May
26]. Available from:
https://people.dbmi.columbia.edu/~ehs7001/
Buchanan-Shortliffe-1984/MYCIN Book.htm

3. Hoffer EP, Feldman MJ, Kim RJ, Famiglietti KT,

Barnett GO. DXplain: Patterns of Use of a
Mature Expert System. AMIA Annu Symp Proc.
2005;321–5.

4. Salzman L. Computers and dentistry. Bull

Bronx Cty Dent Soc. 1966;21(6):343–70.

5. Ding H, Wu J, Zhao W, Matinlinna JP, Burrow

MF, Tsoi JKH. Artificial intelligence in dentistry
—A review. Front Dent Med.
2023;4(February):1–13.
6. Lévy-Mandel AD, Venetsanopoulos AN,
Tsotsos JK. Knowledge-based landmarking of
cephalograms. Comput Biomed Res.
1986;19(3):282–309.

7. Liu J, Chen Y, Li S, Zhao Z, Wu Z. Machine

learning in orthodontics: Challenges and
perspectives. Adv Clin Exp Med. 2021;30(10).

8. Schachnow J. Assisted Intelligence: How AI

Can Benefit Professionals in Everyday Work |
The AI Journal [Internet]. [cited 2024 May 26].
Available from: https://aijourn.com/assisted-
intelligence-how-ai-can-benefit-professionals-
in-everyday-work/

9. Bazoukis G, Hall J, Loscalzo J, Antman EM,

Fuster V, Armoundas AA. The inclusion of
augmented intelligence in medicine: A
framework for successful implementation.
Cell Reports Med. 2022;3(1):1–8.

10. Artificial Intelligence and Machine Learning

(AI/ML)-Enabled Medical Devices [Internet].
USFDA. 2024. Available from:
 https://www.fda.gov/medical-devices/software
-medical-device-samd/artificial-intelligence-
and-machine-learning-aiml-enabled-medical-
devices

11. Bichu YM, Hansa I, Bichu AY, Premjani P,

Flores-Mir C, Vaid NR. Applications of artificial
intelligence and machine learning in
orthodontics: a scoping review. Prog Orthod.
2021;22(1).

12. Wong KF, Lam XY, Jiang Y, Yeung AWK, Lin Y.

Artificial intelligence in orthodontics and
orthognathic surgery: a bibliometric analysis
of the 100 most-cited articles. Head Face
Med. 2023;19(1):1–11.

13. Parthasarathy S, Gregson PG, Fay DF. of

Cephalograms. Comput Biomed Res.
1989;269:248–69.

14. Benn DK. Automatic analysis of radiographic

images: I. Theoretical considerations.
Dentomaxillofacial Radiol. 1991;20(4):187–
92.
15. Nordblom NF, Büttner M, Schwendicke F.
Artificial Intelligence in Orthodontics: Critical
Review. J Dent Res. 2024;

16. Kunz F, Stellzig-Eisenhauer A, Zeman F, Boldt

J. Artificial intelligence in orthodontics:
Evaluation of a fully automated cephalometric
analysis using a customized convolutional
neural network. J Orofac Orthop.
2020;81(1):52–68.

17. Ki YH, Le C, Eun-Gyu H, Cho YJ, Ha SS. A fully

deep learning model for the automatic
identification of cephalometric landmarks.
Imaging Sci Dent. 2021;51:1–8.

18. Lee JH, Yu HJ, Kim MJ, Kim JW, Choi J.

Automated cephalometric landmark detection
with confidence regions using Bayesian
convolutional neural networks. BMC Oral
Health. 2020;20(1):1–10.

19. Kiełczykowski M, Kamiński K, Perkowski K,

Zadurska M, Czochrowska E. Application of
Artificial Intelligence (AI) in a Cephalometric
 Analysis: A Narrative Review. Diagnostics.
2023;13(16).

20. Hwang H-W, Park J-H, Moon J-H, Yu Y, Kim H,

Her S-B, et al. Automated identification of
cephalometric landmarks: Part 2-Might it be
better than human?

21. Gupta A, Kharbanda OP, Sardana V,

Balachandran R, Sardana HK. Accuracy of 3D
cephalometric measurements based on an
automatic knowledge-based landmark
detection algorithm. Int J Comput Assist
Radiol Surg. 2016;11(7):1297–309.

22. Montúfar J, Romero M, Scougall-Vilchis RJ.

Hybrid approach for automatic cephalometric
landmark annotation on cone-beam
computed tomography volumes. Am J Orthod
Dentofac Orthop. 2018;154(1):140–50.

23. Niño-Sandoval TC, Guevara Pérez S V.,

González FA, Jaque RA, Infante-Contreras C.
Use of automated learning techniques for
predicting mandibular morphology in skeletal
 class I, II and III. Forensic Sci Int. 2017 Dec
1;281:187.e1-187.e7.

24. Niño-Sandoval TC, Guevara Perez S V.,

González FA, Jaque RA, Infante-Contreras C.
An automatic method for skeletal patterns
classification using craniomaxillary variables
on a Colombian population. Forensic Sci Int.
2016 Apr 1;261:159.e1-159.e6.

25. Yu HJ, Cho SR, Kim MJ, Kim WH, Kim JW, Choi
J. Automated Skeletal Classification with
Lateral Cephalometry Based on Artificial
Intelligence. J Dent Res. 2020;99(3):249–56.

26. Kim I, Misra D, Rodriguez L, Gill M, Liberton

DK, Almpani K, et al. Malocclusion
Classification on 3D Cone-Beam CT
Craniofacial Images Using Multi-Channel Deep
Learning Models. Proc Annu Int Conf IEEE Eng
Med Biol Soc EMBS. 2020;2020-July:1294–8.

27. Ben-Hamadou A, Smaoui O, Rekik A, Pujades

S, Boyer E, Lim H, et al. 3DTeethSeg’22: 3D
Teeth Scan Segmentation and Labeling
 Challenge. 2023;1–29.

28. Dipalma G, Inchingolo AD, Inchingolo AM,

Piras F, Carpentiere V, Garofoli G, et al.
Artificial Intelligence and Its Clinical
Applications in Orthodontics: A Systematic
Review. Diagnostics. 2023;13(24):1–22.

29. Kazimierczak N, Kazimierczak W, Serafin Z,

Nowicki P, Nożewski J, Janiszewska-Olszowska
J. AI in Orthodontics: Revolutionizing
Diagnostics and Treatment Planning—A
Comprehensive Review. J Clin Med.
2024;13(2).

30. Atici SF, Ansari R, Allareddy V, Suhaym O,

Cetin AE, Elnagar MH. AggregateNet: A deep
learning model for automated classification of
cervical vertebrae maturation stages. Orthod
Craniofacial Res. 2023;26(S1):111–7.

31. Kök H, Acilar AM, İzgi MS. Usage and

comparison of artificial intelligence algorithms
for determination of growth and development
by cervical vertebrae stages in orthodontics.
 Prog Orthod. 2019;20(1).

32. Thanathornwong B. Thanathornwong, B.

(2018). Bayesian-based decision support
system for assessing the needs for
orthodontic treatment. Healthcare informatics
research, 24(1), 22-28. 2018;24(1):22–8.

33. Monill-González A, Rovira-Calatayud L,

d’Oliveira NG, Ustrell-Torrent JM. Artificial
intelligence in orthodontics: Where are we
now? A scoping review. Orthod Craniofacial
Res. 2021;24(S2):6–15.

34. Kunz F, Stellzig-Eisenhauer A, Boldt J.

Applications of Artificial Intelligence in
Orthodontics—An Overview and Perspective
Based on the Current State of the Art. Appl
Sci. 2023;13(6).

35. Kato RM, Parizotto J de OL, Oliveira PHJ de,

Gonçalves JR. Artificial Intelligence in
Orthognathic Surgery – A Narrative Review of
Surgical Digital Tools and 3D Orthognathic
Surgical Planning. J Calif Dent Assoc.
 2023;51(1).

36. Zhu J, Yang Y, Wong HM. Development and

accuracy of artificial intelligence-generated
prediction of facial changes in orthodontic
treatment: a scoping review. J Zhejiang Univ
Sci B. 2023;24(11):974–84.

37. Bouletreau P, Makaremi M, Ibrahim B,

Louvrier A, Sigaux N. Artificial Intelligence:
Applications in orthognathic surgery. J
Stomatol Oral Maxillofac Surg.
2019;120(4):347–54.

38. Siddiqui TA, Sukhia RH, Ghandhi D. Artificial

intelligence in dentistry, orthodontics and
Orthognathic surgery: A literature review. J
Pak Med Assoc. 2022;72(1):S91–6.

39. Caruso S, Caruso S, Pellegrino M, Skafi R,

Nota A, Tecco S. A Knowledge-Based
Algorithm for Automatic Monitoring of
Orthodontic Treatment: The Dental
Monitoring System. Two Cases. Sensors. 2021
Mar 7;21(5):1856.
40. Strunga M, Urban R, Surovková J, Thurzo A.
Artificial Intelligence Systems Assisting in the
Assessment of the Course and Retention of
Orthodontic Treatment. Healthc.
2023;11(5):25–6.

41. Tomášik J, Zsoldos M, Oravcová Ľ, Lifková M,

Pavleová G, Strunga M, et al. AI and Face-
Driven Orthodontics: A Scoping Review of
Digital Advances in Diagnosis and Treatment
Planning. AI. 2024;5(1):158–76.

42. Smith WM. Making Decisions. J Grad Med

Educ. 2023;15(4):509–10.

43. Albahri AS, Duhaim AM, Fadhel MA, Alnoor A,

Baqer NS, Alzubaidi L, et al. A systematic
review of trustworthy and explainable
artificial intelligence in healthcare:
Assessment of quality, bias risk, and data
fusion. Inf Fusion. 2023;96:156–91.

44. Team I data and A. Shedding light on AI bias

with real world examples [Internet]. IBM.
2023. Available from:
 https://www.ibm.com/blog/shedding-light-on-
ai-bias-with-real-world-examples/

45. Ueda D, Kakinuma T, Fujita S, Kamagata K,

Fushimi Y, Ito R, et al. Fairness of artificial
intelligence in healthcare: review and
recommendations. Jpn J Radiol. 2024;42(1):3–
15.

46. Ferrara E. Fairness and Bias in Artificial

Intelligence: A Brief Survey of Sources,
Impacts, and Mitigation Strategies. Sci.
2024;6(1).

47. Gichoya JW, Thomas K, Celi LA, Safdar N,

Banerjee I, Banja JD, et al. AI pitfalls and what
not to do: mitigating bias in AI. Br J Radiol.
2023;96(1150).

48. Yassir YA, Salman AR, Nabbat SA. The

accuracy and reliability of WebCeph for
cephalometric analysis. J Taibah Univ Med Sci.
2022;17(1):57–66.

49. Matthieu Moreno SG-C. Comparative study of

two software programs for detecting
 cephalometric landmarks by artificial
intelligence. French Orthod. 2022;93(1):41–
61.

50. Salazar D, Rossouw PE, Javed F,

Michelogiannakis D. Artificial intelligence for
treatment planning and soft tissue outcome
prediction of orthognathic treatment: A
systematic review. J Orthod. 2023;(July).

51. Hashimoto DA, Rosman G, Rus D, Meireles

OR. Artificial Intelligence in Surgery: Promises
and Perils. Vol. 268, Annals of Surgery. 2018.
p. 70–6.

Legends :

Abbreviations Used:

AI : Artificial intelligence

ANN : Artificial neuron network

CNN : Convoluted neuron network

CBCT : Cone Beam Computed Tomography

CASS : Computer Assisted Surgical Simulation

CVM : Cervical Vertebrae Maturation

DAI : Dental Aesthetic Index

DMS : Dental Monitoring system ()

ITON :Index of Orthodontic Treatment Need

ML : Machine Learning

USFDA : United States Food and Drug

Administration

Figures :

Fig 1 : Different types of Bias in AI in Orthodontics

 BIAS IN AI
IN
ORTHODON
TICS

ALGORIT COGNIT
DATA HMIC IVE
BIAS BIAS BIAS

TRAININ
MINORI G COHORT LABEL
TY BIAS INFORMATIV SERVING
NESS BIAS SKEW
BIAS BIAS