artificial intelligence in orthodontics

1. ARTIFICIAL INTELLIGENCE
Dr. Harivadhani

2. Contents
 Introduction
 AI?
 Conventional diagnostic procedures
 Neural network
 Supervised vs Unsupervised learning
 AI in orthodontics

3.  Machine learning for tooth movement
 Extraction or non extraction therapy
 Orthognathic surgery
 Segmentation
 Growth prediction
 Clefts
 TMDs
 Conclusion

4. Introduction
• When various treatment options can be applied to a problem, there is apparently a
greater decision-making difficulty.
• Besides, if any of the treatment options are novel, clinicians are apparently more
likely to choose a more obsolete form of treatment or even suggest that no
treatment is needed.
• Thus far, no tools exist to lead patients and clinicians out of the decision-making
uncertainty, in which they are trapped when they face a condition that has several
possible correct treatment options – though some better than others.

5. • AI is the behavior of non-biological entities that perceive, learn, or
react to complex environments.
• AI is not a computational tool that necessarily mimics the workings
of the human brain; rather, it is a set of tools for problem-solving,
each with its own specific rules.

6. • In AI algorithm inputs are given to the model, it estimates outputs and calculates
the difference between the estimates and the labels.
• This difference, which is the error of the model, is automatically used to correct
internal parameters to minimize future error.
• By performing this process for thousands – or millions – of different inputs, the
error decreases. When the model obtains acceptable error rates, the algorithm is
ready to be used for new, unlabeled data.

7. Artificial intelligence offers “a way to get sharper
prediction from data” by simultaneously analyzing
all the different variables present in a
malocclusion.
This capacity offers the potential to assist the
practitioner to obtain the most favorable outcome
when treating a malocclusion.

8. CONVENTIONAL DIAGNOSTIC PROCEDURE
• Studying and diagnosing a malocclusion present many challenges and bring
uncertainty to the outcome of treatment due to the large number of variables
present in the analysis.
• The orthodontist must compute mentally all the parameters to recognize patterns
based on experience and adopt the most logical or probable approach to solve the
problem presented.
• This process, called the “feedforward approach”

9. • It does not necessarily involve complex feedback mechanisms to improve on
previous diagnostics and outcome analyses.
• This explains the principle as an accepted but potentially inefficient pattern to
treat patients.
• This experience-based approach has been advocated by many clinicians as it is
simple but lacks the means to re-evaluate and learn from positive and/or
negative outcomes.

11. Neural Network?

12. • The neural networks learn to recognize patterns and assess the correct probability
of success.
• Neural networks consist of nodes loosely referred as neurons. Each node
corresponds to a random variable.
• These nodes require inputs such as amount of overbite, overjet, and crowding
which are initially randomly weighted (0 to 1%) and linked to nodes which calculate
the correlation in the input layer.

13. The connectivity between the nodes gives the power to the network. The color of the
connectors represents the “weight” or the influence of the given node in the network.
The network then establishes the output or the most probable diagnosis. This output
is dependent on the amount of data (input) and the weights allocated to the data at
the input level.

14. • The use of AI in Orthodontics is, for the moment, limited to supervised learning
such as objects or point recognition.
• The best example is cephalometric software programs such as WEBceph™ or
AudaxCeph™, Cephio™, CephX™, DentaliQ-Ortho™, EYES.OF.AI™, and FPT-
Software™ which are trained to recognized points in cephalometric radiographs.
• For clinical purposes, Cephalometric Landmark Identification Data could readily be
extended even to predict and visualize soft-tissue changes after treatment.
• The application of AI in automated cephalometric landmark identification may
lessen the burden and alleviate human errors.

16. By gathering radiographic data automatically, it also helps reduce human tasks and the
time required for both research and clinical purposes.
AI showed as accurate an identification of cephalometric landmarks as did human
examiners.

18. Another use of AI is found in the automation of case setups for indirect bracketing or
production of aligners.
These automated setups allow for the visualization of treatment outcomes using
specific algorithms again based on supervised learning.
New algorithms are being developed to identify teeth in Cone Beam CTs automatically.

20. • Unsupervised learning is referred to a higher level of complexity where the data
provided is not labeled or classified. The computer program in an unsupervised mode
will “train” algorithms to recognize patterns and suggest potential outcomes from a
dataset.
• In orthodontics, unsupervised learning is starting be used to input a large quantity of
malocclusions and let the computer predict the most appropriate treatment options.
• The before and after treatment casts of a large number of orthodontic cases are fed
into a neural network.
• The neural network will then recognize patterns and train itself to recognize and
suggest the best course of action.

22. AI IN DIAGNOSIS
• Imaging diagnoses have gradually incorporated AI to increase sensitivity and
specificity.
• There are more than 8000 identified genetic syndromes. However, despite all the
advances in genetics, including next-generation sequencing-based tests,
establishing the correct diagnosis is still a difficult task.
• Craniofacial phenotypes are extremely informative for establishing the correct
diagnosis of genetic congenital diseases because many syndromes have
recognizable facial features.

23. • One such advancement is a mobile phone application called Face2Gene (FDNA,
Boston, USA).
• The application uses the contrast of a patient’s image against thousands of images
in its databases to determine the subtle patterns that different syndromes tend to
have. The diagnostic hypothesis established by the App has already proved to be
useful and outperformed clinicians in diagnosing a number of syndromes.

25. • Another interesting recent application of AI is the prediction of extractions in
orthodontic planning.
• The teeth to be extracted (first and second premolars) and the variability of
dentofacial alterations included.
• AI has already been used to diagnose and classify osteoarthritis in the
temporomandibular joint, and it may provide future data for establishing
treatments for problems that are specific to the different severities of the
condition.

26. MACHINE LEARNING FOR TOOTH MOVEMENT
PLANNING
• AI is an excellent tool to help orthodontists to choose the best way to move teeth
from point A to point B, once the orthodontist instructs the machine where the
final position should be.
• AI today completely ignores the existence of oral diseases and possible previous
health treatments that may affect the prescription of orthodontic corrections,
either with aligners or fixed appliances.
• However, performing orthodontic movement with active disease is
contraindicated.

27. • Companies in several countries have been selling aligners to patients without proper
dental supervision. This has led to numerous reports of tooth mutilation and bone loss
in the general population.
• Another limitation of AI algorithms being implemented today is that they do not
incorporate patients’ facial analysis, their proportions, and esthetics. There is a direct
interaction between orthodontic dental movements and facial esthetics.
• Only a qualified orthodontist can perform these analyses because tooth movement in
any direction of the space is commonly connected with facial and smile esthetics

28. • AI used in contemporary planning does not consider the impact of functional
problems and the stability of the tooth position – or lack thereof – when tooth
movements are performed.
• For example, problems associated with important functional etiology, such as the
open bite malocclusion, can be treated using aligners.
• However, AI today cannot determine the etiology of the problem or predict specific
retention strategies.

29. • The natural conclusion is that the algorithms are biased by poor treatment results
and need to improve considerably before they can significantly help orthodontists
achieve excellent treatment results.
• In addition, AI algorithms do not effectively incorporate many orthodontic tools,
thereby limiting treatment tool and strategies, such as skeletal anchorage, dental
extractions, and integrated restorative procedures.
• This is at least partially associated with the mechanical limitations of aligners to
control certain tooth movements.
• However, orthodontic treatment and appliances need to be patient centered to
improve the user experience.

30. Extraction or non-extraction therapy in orthodontic treatment
• Xie et al. constructed a decision-making expert system for orthodontic treatment of
patients aged between 11 to 15 years old to decide whether tooth extraction is
needed by using back propagation ANN model
• 200 subjects were chosen, 120 receiving extraction therapy and 80 receiving non-
extraction therapy. 23 indices were selected as input data for each patient, and
extraction or non-extraction were calculated as output data.
• The constructed ANN in this study showed 80% accuracy in testing set. Moreover,
lip incompetence and IMPA(L1-MP) were the two indices that give the biggest
contribution to the output data.

32. • Jung et al. also constructed neural network model combined with back propagation
algorism. There were 156 patients that included in the study.
• Twelve cephalometric variables and 6 indexes were selected as input data.
Extraction or non-extraction and extraction pattern were set as output data.
• The treatment plans were determined by one orthodontic specialist. Different from
Xie’s study, it further divided data into training data and validation data. AIterative
learning was stopped at the minimum error point of the validation set to prevent
overfitting.
• The success rate of the models was 93% for the diagnosis of extraction or non-
extraction therapy and 84% for the selection of extraction pattern.

33. Orthognathic surgery
• Great investment has been made in research and development of digital orthodontics and 3D
simulation of orthognathic surgery.
• Knoops et al. developed a machine learning framework for automated diagnosis and
computer-assisted planning in plastic and reconstructive surgery. They presented the large-
scale clinical 3D morphable model (3DMM), a machine-learning framework including
supervised learning constructed with surface 3D scan.
• A specialist can automatically produce 3D simulation of post-surgical outcome with mean
accuracy of 1.1±0.3 mm. However, only surface scan was used in this study, so the underline
bone movement needed to be calculated according to soft tissue movement which still
remain a big task nowadays.

34. • Comparing to Knoops’s study who used 3DMM to come up with diagnosis, Choi et
al. applied ANN obtained from 12 measurement values of the lateral cephalogram
and 6 additional indexed.
• The machine learning model consisted of 2-layer neural network with one hidden
layer. The sample included 316 patients with 160 were planned with surgical
treatment and 156 were planned with non-surgical treatment.
• The success rate of the model showed 96% for whether the patient need surgical
treatment, and 91% for the detailed diagnosis of surgery type and extraction
decision. The success rate is comparable between these two studies.

35. • Niño-Sandoval et al. tried to predict mandible bone morphology based on maxilla
morphology using ANN. 299 lateral cephalograms was obtained from Colombian
patients with 19 landmarks on X and Y coordinates. The result showed high
predictability.
• Patcas et al. conducted an interesting study assessing the impact of orthognathic
treatments on facial attractiveness and estimated age by AI technologies.
• For age estimation, the convolutional neural network (CNN) model was trained with
> 0.5 million facial images with age labels.

36. • For attractiveness prediction, the CNN model was trained on data from a dating
site with > 13000 face images and >17 million ratings for attractiveness.
• Presurgical and postsurgical photos of 146 consecutive orthognathic patients were
collected for this single-center study.
• According to the algorism, 66.4% of the patients improved with the treatment
resulting in younger appearance of nearly 1 year. The study showed that AI might
be an objective way evaluating treatment outcome in terms of aesthetic
improvement.

38. Segmentation and landmark identification
• Image segmentation is the process we isolate the pixels of target organs or lesion
from medical images such as X-rays, CT, or MRI.
• Landmark identification in lateral cephalometric X-ray have been of paramount
importance in terms of diagnosis and treatment planning in orthodontic treatment
for decades

40.  Wang et al. developed a method for automated segmentation of maxilla and
mandible through CBCT.
 They applied a learning-based framework to simultaneously segment both maxilla
and mandible from CBCT based on random forest.
 Dice ratio is a popular way evaluating volumetric segmentation of medical images.
The definition is the sum of intersection voxels of the learned and ground truth
sets times two divided by the sum of the respective voxels.
 The average Dice ratio of mandible and maxilla were 0.94 and 0.91 respectively.

41. • Chen et al. used a machine learning algorism based on Wang’s technique to assess
maxillary structure variation in unilateral canine impaction.
• 20 Subjects included 30 study group patients with unilateral maxillary canine
impaction and 30 healthy control group subjects.
• Maxillary structure was auto-segmented and no significant difference in bone
volume was found between impacted side and non-impacted side in study group.
• Study group had significant smaller maxillae volume than control group. The
segmentation efficiency has been greatly improved by the automatic algorism.

42. Growth prediction
• Timing is one of the key points needed to be considered during treatment
planning, especially among growing patients. Several methods have been proposed
for growth prediction such as chronological age, menarche, change in voice and
body height, and bone age.
• The gold standard for assessing bone age was obtained by hand-wrist radiographs,
however, Lamparski reported that by reading cervical vertebrae stages, similar
accuracy could be attained and preventing additional radiation at the same time.
• Spampinato used deep learning approaches to assess bone age through hand-wrist
radiographs.

43. • The dataset contained 1391 X-ray left-hand scans of children of age up to 18 years
old with bone age values provided by two expert radiologists. The result showed an
average discrepancy between manual and automatic evaluation of about 0.8 years.
• Kok et al. compared different AI algorisms for determination of growth by cervical
vertebrae stages.
• ANN showed most stable result and was suggested the preferred method for
determining cervical vertebrae stage.

45. Cleft related studies
• Zhang collected blood samples from healthy control and non-syndromic cleft lip and
palate infants (NSCL/ P) in Han and Uyghur Chinese population to validate the
diagnostic effectiveness of 43 single nucleotide polymorphisms (SNPs) previously
detected using genome-wide association studies.
• Different machine learning algorisms was used to build predictive models with those
SNPs and evaluated their prediction performance. The result showed logistic
regression had the best performance for risk assessment.

46. • Defective variants in MTHFR and RBP4, two genes involved in folic acid and vitamin
A biosynthesis, were found to have high contributions to NSCL/P incidence based
on feature importance evaluation with logistic regression.
• This is in consistence with the impression that folic acid and vitamin A are essential
for reducing the risk of conceiving an NSCL/P baby.

47. • Patcas et al. used a CNN model previously trained on a dataset of dating site with
>13000 face images and > 17000 ratings for attractiveness to compare facial
attractiveness between treated cleft patients and controls.
• Human rated significantly higher than AI for the score of attractiveness of controls.
Attractiveness scores were comparable in treated cleft patients rated by AI and
human.
• The result suggested that AI still need to improve its interpretation of cleft features
impacting on facial attractiveness, to become a better tool evaluating aesthetics

48. TMD classification
• Shoukri et al. applied neural network to stage condylar morphology in
temporomandibular joint osteoarthritis (TMJOA). The neural network was trained
on 259 condyles to detect and classify the stage of TMJOA and compare to clinical
expert’s classification.
• Condylar morphology was classified into 6 groups by CBCT image. Predictive
analytics of the AI’s staging of TMJOA compared to the repeated clinicians’
consensus showed 73.5 and 91.2% accuracy. The results suggest that TMJOA
condylar morphology can be comprehensively classified by AI.

51. • AI have been applied to robotic surgeries in neurological, gynecological, cardiothoracic
and numerous general surgical procedures.
• It is quite promising in the near future that AI robotic technologies could be applied to
orthognathic surgery as well.
• It can reduce infection rate because only robotics have contact with the patient.
Higher precision of jaw movement can be expected at the same time.

52. • Applying AI technology, we can build a diagnostic/prognostic model based on the
‘big data’ and predicting treatment results.
• It is not only a one-way process, taking the goodness of the predictive result as a
feedback, we can further fine-tune the previous model and feature engineering
process to get a positive feedback loop.

53. • In orthodontic field, the concept of precision medicine means a more complex
diagnostic process, a more personalized treatment planning and a more
sophisticated treatment process and those might lead to a more efficient
treatment with less side effects and treatment duration.
• Hopefully, the medical quality could be raised while decreasing the medical costs
through the application and development of AI technology.

54. Conclusion
 AI is driving discoveries across all sciences.
 Its powerful pattern finding and prediction algorithms are helping researchers and
clinicians in all fields– from finding new ways to access sleep quality to classify the
presence and absence of root and crown caries.
 However, currently, there are more tangible boundaries and more immediate and
achievable goals in orthodontics.

55. References
• Jean-Marc Retrouvey ,The role of AI and machine learning in contemporary
orthodontics. APOS Trends in Orthodontics Volume 11 Issue 1 January-March
2021
• Faber J, Faber C, Faber P. Artificial intelligence in orthodontics. APOS Trends
Orthod. 2019;9(4):201-5.
• Xie X, Wang L, Wang A. Artificial neural network modeling for deciding if
extractions are necessary prior to orthodontic treatment. The Angle orthodontist.
2010 Mar;80(2):262-6.
• Patcas R, Bernini DA, Volokitin A, Agustsson E, Rothe R, Timofte R. Applying
artificial intelligence to assess the impact of orthognathic treatment on facial
attractiveness and estimated age. International journal of oral and maxillofacial
surgery. 2019 Jan 1;48(1):77-83.

56. • 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. Progress in orthodontics. 2019 Dec;20(1):1-0.
• Shoukri B, Prieto JC, Ruellas A, Yatabe M, Sugai J, Styner M, Zhu H, Huang C,
Paniagua B, Aronovich S, Ashman L. Minimally invasive approach for diagnosing
TMJ osteoarthritis. Journal of dental research. 2019 Sep;98(10):1103-11.
• Kim D, Choi E, Jeong HG, Chang J, Youm S. Expert System for Mandibular
Condyle Detection and Osteoarthritis Classification in Panoramic Imaging Using
R-CNN and CNN. Applied Sciences. 2020 Jan;10(21):7464.

57. • Patcas R, Bernini DA, Volokitin A, Agustsson E, Rothe R, Timofte R. Applying
artificial intelligence to assess the impact of orthognathic treatment on facial
attractiveness and estimated age. International journal of oral and
maxillofacial surgery. 2019 Jan 1;48(1):77-83.
• Knoops PG, Papaioannou A, Borghi A, Breakey RW, Wilson AT, Jeelani O,
Zafeiriou S, Steinbacher D, Padwa BL, Dunaway DJ, Schievano S. A machine
learning framework for automated diagnosis and computer-assisted planning
in plastic and reconstructive surgery. Scientific reports. 2019 Sep 19;9(1):1-2.

58. THANK YOU