Title	Artificial Intelligence Can Improve and Reliably Assist in Orthodontic Treatment
Clinical Question	For patients who require orthodontic treatment, will implementing artificial intelligence (AI) systems assist and even improve orthodontic treatment compared to conventional orthodontic treatment?
Clinical Bottom Line	For patients who require orthodontic treatment, implementation of AI systems results in greater accuracy in diagnosis and treatment planning compared to conventional orthodontic treatment planning methods, especially in so-called borderline extraction cases. This is supported by a systematic review and two comparative studies showing that AI is equivalent or superior to a human clinician and/or non-AI orthodontic software in parameters such as tooth position detection, extraction decision-making, and treatment monitoring.
Best Evidence
PubMed ID Author / Year Patient Group Study type (level of evidence) 37148164 Kapoor/2023 60 borderline orthodontic extraction cases Observational Analytical Study Key results The aim of this study was to create an artificial intelligence model to determine if tooth extraction should be performed in borderline orthodontic cases. In this study, 60 borderline orthodontic cases were used with 40 cases being used as a training dataset and 20 cases used as a testing dataset. The opinions of 20 orthodontists were collected regarding extraction necessity for the 40 cases in the training dataset. From this, an artificial neural network (ANN) model for extraction or non-extraction treatment plans was constructed. This model was tested with the remaining 20 borderline cases and the accuracy and F1 score were calculated. The F1 score “combines precision and recall relative to a specific positive class.” The AI model showed an accuracy of 97.97% for extraction and non-extraction decision making. The model showed precision regarding whether to extract or to not extract of 0.9 and 0.8, respectively. The recall value of the model was 0.87 for both extraction and non-extraction. The F1 score was 0.82 and 0.88 for non-extraction and extraction, respectively. These numbers are close to 1 and therefore close to perfection. This shows that the AI model showed precision in treatment planning orthodontic cases where the decision to extract teeth is considered to be borderline. 38746029 Hack/2024 45 patients aged 18-35 with Class I Angle anomalies Comparative Study Key results This study shows that using artificial intelligence in orthodontics shows an “increased level of tooth position detection and more stability in the analysis” compared to traditional orthodontic software. Two types of algorithms (AI algorithms and classical software algorithms) were used on 45 patients who had Class I Angle anomalies. The study showed that the AI algorithms resulted in better tooth position detection compared to traditional software. Each case was scored out of 10 “depending on the certainty of the detection of dental contours and the correctness of data processing.” The AI algorithms outperformed the traditional methods across all the patient groups. This study shows the accuracy of AI versus conventional methods in diagnosis and treatment planning by providing accurate guidance. 38132261 Dipalma/2023 Qualitative Analysis of 33 studies which included over 1,000 patients Systematic Review Key results This systematic review included 33 studies. These studies showed the effectiveness of artificial intelligence in improving the diagnosis, treatment planning, and assessment of orthodontic cases. The 33 studies were broken down into different uses of AI in orthodontics: diagnosis, identifying anatomic landmarks from a lateral cephalogram, assessment of vertebral maturation, treatment planning, and treatment monitoring. The systematic review highlighted that AI showed accuracy in making diagnoses. Multiple studies showed there was no statistical differences in the analysis of cephalometric radiographs between AI and human analysis. For AI-guided treatment planning, AI showed high accuracy in treatment planning, up to 93% in a study. From this systematic review, AI is shown to be precise and comparable to a clinician’s decision making.
Evidence Search	“AI” AND “orthodontics”
Comments on The Evidence	In the Kapoor et al. study, 60 patients were split into two groups, a training and testing dataset. The patients were randomly distributed into these datasets to reduce bias. The study also provided multiple quantitative values which include recall, precision, and F1 scores. Bias was also attempted to be eliminated “with the aim of constructing an ANN model that is expected to deliver an unbiased decision in predicting the treatment modality for borderline cases.” This study did have some limitations. The sample size of 60 is small. The training dataset to make the ANN model needs to be increased to improve AI’s prediction precision. The outcomes of this study are not fully generalizable but remain more suited to patients with Angle Class I anomalies. The study conducted by Hack et al. is the best evidence showing that AI systems outperform conventional orthodontic systems. The 45 patients were divided into three groups with each group being evaluated by a different orthodontist to eliminate bias. However, it should be noted that no quantitative data was collected in this study. The article states that “each case analyzed was awarded a score of 1-10 for each verification method.” The doctors belong to the Hack Clinic, which means they are familiar with the software which could lead to a certain level of subjectivity involved in this study. Because the study did not perform a statistical analysis, it is not possible to reject or fail to reject the null hypothesis. Additionally, a larger sample size greater than 45 patients will make this study’s results more believable. In the systematic review, 33 articles were selected based on PRISMA protocols to help improve the transparency of the review. The risk of bias was investigated in the review. The authors concluded that most of the studies have a high risk of confounding. The authors also stated that there is a low risk of bias with participant selection, missing data, and measurement outcomes. Bias due to postexposure interventions could not be calculated due to high heterogeneity. The authors found that “ten studies have a low risk of bias, ten studies have a high risk of bias, four have a very high risk of bias, and the remainder have a questionable risk of bias.”
Applicability	Kapoor et al. included patients aged 12-24 years, and Hack et al. included patients aged 18-35 years. These are normal ages for receiving orthodontic treatment. In the Hack et al. study, the patients that were seen had “all permanent teeth erupted on the arch except for permanent 3rd molars, presenting first class Angle, a harmonious profile, and no previous orthodontic treatment.” These are typical patients that orthodontists will see for treatment. AI has been increasingly used in dentistry for many applications such as analyzing radiographs and intraoral scanning. AI could be intriguing for the clinician to use to help simplify tasks, help save the clinician time, and help the clinician perform their tasks more precisely. However, there is a financial and time cost to implementing AI systems as well as a learning curve which could limit the applicability of AI in orthodontic offices.
Specialty	(Orthodontics)
Keywords	“artificial intelligence” “AI” “orthodontics” “treatment planning”
ID#	3559
Date of submission	06/03/2024
E-mail	Reeda4@livemail.uthscsa.edu
Author	Alexander Reed
Co-author(s)	Jacob Robinson
Co-author(s) e-mail	robinsonj1@livemail.uthscsa.edu
Faculty mentor	Kelly Lemke DDS, MS
Faculty mentor e-mail	LemkeK@uthscsa.edu
Basic Science Rationale (Mechanisms that may account for and/or explain the clinical question, i.e. is the answer to the clinical question consistent with basic biological, physical and/or behavioral science principles, laws and research?) None available
Comments and Evidence-Based Updates on the CAT (FOR PRACTICING DENTISTS', FACULTY, RESIDENTS and/or STUDENTS COMMENTS ON PUBLISHED CATs) None available