Maxillary Crowding and Spacing: Validation of an Artificial Intelligence Model vs. Digitally-Assisted Human Observer

Maxillary Crowding and Spacing: Validation of an Artificial Intelligence Model vs. Digitally-Assisted Human Observer | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Maxillary Crowding and Spacing: Validation of an Artificial Intelligence Model vs. Digitally-Assisted Human Observer Haneen Hatoum, Wael Talaat, Ahmed Kaboudan, Aasem Hamed, Engy Mahmoud, and 3 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6898322/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 7 You are reading this latest preprint version Abstract Purpose The aim of this study was to develop an artificial intelligence (AI) model capable of quantifying crowding and spacing in the upper arch and to validate its accuracy by comparing the model’s results with those of human observers. Materials and Methods This study included upper intraoral photographs and occlusal scans of orthodontic patients treated at the University of Sharjah (2022–2024). The YOLO (You Only Look Once) 8 Pose Model was generated using a training and validation dataset (832 images). The AI model performed tooth segmentation and tooth point detection on occlusal images, followed by automated quantification of tooth size arch length discrepancy (TSALD). Manual space analysis was conducted using OrthoCAD software and the data was compared with the results of the AI model using a testing dataset (300 images). TSALD was categorized based on the index of treatment complexity, outcome, and need (ICON). Qualitative data were presented as frequency and distribution, and comparisons were performed by using Fisher’s Exact test. Correlation between Manual and AI-measured TSALD was evaluated using Pearsons’s correlation coefficient. Results The model achieved an overall accuracy of 90%. The largest discrepancies were found in mild crowding ( 0.92) between manual and AI TSALD measurements indicated high reliability and potential interchangeability. Conclusions The AI model was successfully developed and validated, achieving 90% accuracy, demonstrating its potential as a reliable tool for quantifying TSALD in orthodontic diagnostics. Artificial Intelligence Orthodontics Index of Complexity Outcome and Need Tooth Size Arch Length Discrepancy OrthoCAD Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Introduction Orthodontic diagnosis begins with a thorough evaluation of the patient’s medical and dental history, clinical examination, and essential diagnostic records, including photographs, study models, and radiographs [ 1 ]. The orthodontist then analyzes this data to create a prioritized problem list, incorporating patient-specific needs to determine appropriate treatment options. However, this process is complex, time-consuming, and requires careful evaluation of multiple parameters to ensure optimal management of malocclusion [ 2 ]. To standardize treatment planning, various indices have been developed to assess orthodontic treatment needs, including the Peer Assessment Rating (PAR) index [ 3 ] and the Index of Treatment Need (IOTN) [ 4 ]. However, these indices have limitations, such as failing to describe treatment complexity and over- or underestimating treatment goals [ 5 ]. The PAR index has been criticized for overlooking poor finishes and being overly harsh on treatments with limited aims, while the IOTN does not necessarily reflect treatment complexity [ 3 , 4 ]. To address these shortcomings, the Index of Complexity, Outcome, and Need (ICON) was developed in 2000 by Daniels and Richmond [ 5 ]. Unlike previous indices, ICON integrates five components, including dental aesthetics (IOTN aesthetic component), crossbite, deep/open bite (PAR), upper arch crowding/spacing (five-point scale), and buccal segment anteroposterior relationship (PAR), providing a unified measure of treatment need, complexity, and outcome acceptability [ 5 ]. Tooth size arch length discrepancy (TSALD) is used to quantify dental crowding and spacing by measuring the difference between basal arch length and the combined mesiodistal widths of incisors, canines, and premolars. Various measurement techniques exist, including manual methods such as Lundström’s segmental analysis and Carey’s brass wire technique, visual estimation, and digital analysis. While visual estimation is convenient, it is highly subjective, with studies showing discrepancies of up to 15 mm among orthodontists [ 6 ]. Digital methods such as using OrthoCAD (Cadent, Fairview, NJ, USA) offer an effective alternative, yet digital measurements require manual identification of the tooth contact points for calculation of TSALD. This can be time-consuming, with a high risk of error due to fatigue [ 7 ]. In recent years, Artificial Intelligence (AI) has emerged as a powerful tool in orthodontics, offering potential solutions to the limitations of traditional methods. AI mimics human intelligence by processing large datasets and identifying patterns to optimize decision-making [ 8 ]. Since its introduction into healthcare in the early 2000s, AI has played a key role in enhancing accuracy and efficiency in medical diagnosis and treatment planning [ 9 ]. Within AI, Machine Learning (ML) enables computers to learn from data without explicit programming, while Deep Learning (DL) refines pattern recognition and has outperformed traditional ML techniques in tasks such as classification, segmentation, and detection [ 7 ]. The use of AI in orthodontics has expanded significantly, with numerous studies exploring its applications. Research has examined AI models for cephalometric landmark identification, bone age estimation using cervical vertebra and hand-wrist radiographs, and orthodontic tooth extraction decision-making [ 8 ]. Additionally, AI has been studied for palatal shape analysis, automated skeletal classification, and orthognathic surgery diagnosis and planning [ 8 ]. Despite these advancements, a significant gap remains in automating the quantification of crowding and spacing for evaluating orthodontic treatment need and complexity. No prior study has applied an AI model to assess these factors in alignment with the ICON index. This study takes a pioneering step toward validating an AI model against one component of the ICON index, laying the foundation for a fully automated ICON-based assessment system. The aim of this study was to develop an automated AI model capable of quantifying crowding and spacing in the upper arch and to validate its accuracy by comparing its results with those of human observers. The null hypothesis is that there would be no statistically significant differences between the AI model and human observers in quantifying crowding and spacing in the upper arch. This research aimed to bridge the gap between AI-based automation and standardized orthodontic assessment, ultimately contributing to more efficient and precise treatment planning. Materials and methods Approval for the study was obtained by the ……….; approval number (………..). Consent for the use of patient records from the university database was already obtained from an informed consent signed by the patients prior to initiating orthodontic treatment in the Department of Orthodontics at the University of Sharjah. Dataset The development of the AI model was conducted in three phases: training, validation, and testing against a human observer. A total of 831 upper intraoral images from orthodontic records were used to train and validate the model for measuring the Tooth Size–Arch Length Discrepancy (TSALD) in the upper arch. For testing, a separate dataset comprising upper intraoral images and corresponding intraoral scans from 300 patients was used to compare the model’s performance with that of a human observer. All records were collected from patients undergoing orthodontic treatment at the Orthodontic Department, College of Dental Medicine, University of Sharjah, between 2022 and 2024. Intraoral photographs were exported from the Dolphin Imaging & Management Solutions software (v.12, Chatsworth, CA, USA), and pre-treatment intraoral scans were retrieved from the iTero Element 5D scanner system (Align Technology, San Jose, CA, USA). The AI model was trained to classify teeth and measure TSALD using the intraoral images, while the human observer (….) performed the same measurements on the iTero scans. Inclusion criteria were full permanent dentition (excluding third molars), pre-treatment scans including both arches locked in occlusion. Exclusion criteria included primary or mixed dentition, retained primary teeth, scans with defects, supernumerary teeth, retained roots, and non-orthogonal intraoral images. Training of the model The training dataset consisted of 740 images. The YOLOv8 Pose model; a version of the YOLO object detection system adapted for identifying specific points, was used to detect teeth and locate their mesial and distal points. The model was trained using a multi-task loss function that included object loss, which helps the model learn whether a tooth is present, classification loss, which teaches the model to identify the type of tooth, and localization loss, which focuses on accurately predicting the position of the keypoints. Predefined anchor boxes were used to help the model detect teeth of different sizes. The AI was trained specifically on upper occlusal images and focused on two main tasks: object detection, where each tooth is recognized by its class name and a confidence score (using a threshold of 0.75), and point detection, where the mesial and distal points of each tooth are located. The dataset used for training was balanced, helping the model learn to identify different types of teeth equally and reducing bias, which improves how well the model performs on new cases ( Fig. 1 ) . After training, the model’s results were used to measure space conditions in the upper arch, including crowding (when there is not enough space), spacing (when there is extra space), and arch length (the total space available along the dental arch). Arch length was calculated by drawing a smooth curve through the key points of the teeth, while the space required was measured by adding the mesiodistal widths of all teeth ( Fig. 2 ) . To ensure accurate measurements, a calibration step was performed by estimating the average mesiodistal width of the visible teeth in the intraoral occlusal image and using it as a reference scale to convert pixel measurements into millimeters. Validation of the model A validation set of 92 cases was used. One key issue identified was the model's limited ability to accurately identify missing teeth and account for them in the space analysis. To address this, the model underwent additional training focused on recognizing and appropriately accounting for missing teeth. The model's performance metrics demonstrated significant improvements across various parameters ( Figs. 3 & 4 ) . Testing of the model The required sample size for testing the AI model was calculated using a web-based calculator ( https://wnarifin.github.io/ssc/sskappa.html ). Based on a minimum acceptable Kappa of 0.8, an expected Kappa of 0.9, an outcome proportion of 0.5, a 5% significance level, and 80% power, the minimum required sample was 283 records. Accordingly, records of 300 patients were extracted. Sample demographics (age, gender, and race) were obtained from the aXium software (Exan Group, Vancouver, BC, Canada). Measurements of crowding and spacing were conducted on OrthoCAD software (Cadent, Fairview, NJ, USA), ( Fig. 5 a and 5 b ) . For missing or impacted teeth, the width of the contralateral tooth was used; if both were absent, average mesiodistal widths from Berkovitz et al., were applied [ 10 ]. Arch length was measured from the mesial of the first molar on one side to the mesial of the first molar on the other along the arch. Arch length measurement followed the majority of teeth in the arch regardless of the presence of any blocked-out teeth. TSALD was automatically calculated by subtracting required space from available space, and results were categorized based on the ICON index [ 5 ]. These measurements served as the reference standard. The same cases were then assessed using the AI model, with an acceptable TSALD difference threshold set at 2 mm [ 11 ]. Calibration of the human observer and error measurement All measurements were performed by the primary investigator (H.H) using the OrthoCAD software after receiving proper training. To determine intra-observer reliability, twenty scans of the included sample were randomly selected to be measured again two weeks after the first assessment. To determine inter-observer reliability, the same scans were measured by another examiner. Error measurement was assessed using the intraclass correlation coefficient (ICC), showing near-perfect intraobserver (ICC = 0.998, 95% CI: 0.995–0.999) and interobserver reliability (ICC = 0.996, 95% CI: 0.993–1.000; both p = 0.0001). Statistical analysis Statistical analysis was performed with SPSS 20®, Graph Pad Prism® and Microsoft Excel 2016. Quantitative data were explored for normality by using Shapiro Wilk Normality test and Kolmogorov test and were presented as minimum, maximum, means and standard deviation (SD) values. Qualitative data were presented as frequency and distribution, and comparisons were performed by using Fishers Exact test. Correlation between Manual and AI-measured TSALD was evaluated using Pearsons’s correlation coefficient. Interobserver and Intraobserver reliability was evaluated by using ICC (Interclass correlation coefficient). Results The sample had a balanced gender distribution (53.3% male, 46.7% female). Most participants were Arab, followed by Asian, African, Caucasian, and Persian. The majority (83.3%) had no missing teeth, while 16.7% had at least one missing tooth. Regarding dental categorization, 63% had crowding, and 37% had spacing. Manual and AI measurements of TSALD were compared across crowding and spacing parameters ( Table 1 ) . Their correlations are detailed in Table 2 , showing strong alignment and suggesting high reliability and potential interchangeability. Both crowding and spacing demonstrated statistically significant correlations above 0.92. For crowding, the ICC was 0.948 between manual (-3.84 ± 3.77) and AI measurements (-4.29 ± 4.17), with a 95% CI of 0.931–0.961 (p = 0.0001). For spacing, the ICC was 0.920 between manual (3.93 ± 3.19) and AI values (3.76 ± 3.64), with a 95% CI of 0.884–0.945 (p = 0.0001). Table 1 Descriptive statistics of manual TSALD, AI TSALD, and the difference between them Minimum Maximum Median Mean Standard Deviation Crowding Manual -15.60 − .10 -2.40 -3.84 3.77 AI -18.31 1.42 -3.02 -4.29 4.17 Difference 0.00 7.42 1.04 1.28 1.30 Spacing Manual .10 13.50 3.50 3.93 3.19 AI -5.93 12.67 3.63 3.76 3.64 Difference 0.00 8.13 1.00 1.34 1.29 Shapiro Wilk Normality test and Kolmogorov test. TSALD; Tooth size arch length discrepancy, AI; artificial intelligence. Table 2 Correlation between manual TSALD and AI TSALD regarding crowding and spacing M SD Intraclass Correlation 95% Confidence Interval F Test with Lower Bound Upper Bound Value df1 df2 P value Crowding Manual -3.84 3.77 0.948 0.931 0.961 19.203 188 188 0.0001* AI -4.29 4.17 Spacing Manual 3.93 3.19 0.920 0.884 0.945 12.524 110 110 0.0001* AI 3.76 3.64 Pearsons’s correlation coefficient. *Significant correlation as P ≤0.05, TSALD; Tooth size arch length discrepancy, AI; artificial intelligence. Using a 2 mm threshold, AI achieved 90% accuracy compared to human observers, with 270 cases within and 30 cases outside the threshold. Category distributions are shown in Table 3 . Discrepancies beyond the threshold were significantly associated with the presence of missing teeth ( Table 4 ) . Table 3 Distribution & accuracy of different crowding & spacing categories among manual and AI Manual ICON AI ICON Accuracy Count Column N % Count Column N % Crowding 17mm or impacted teeth 0 0.0% 5 1.7% Spacing up to 2mm 41 13.7% 40 13.3% Spacing 2.1mm-5mm 37 12.3% 27 9.0% Spacing 5.1mm-9mm 24 8.0% 39 13.0% Spacing > 9mm 9 3.0% 6 2.0% Fischer’s Exact test. AI; artificial intelligence, ICON; Index of complexity, outcome, and need. Table 4 Association between missing teeth and the difference in TSALD between manual and AI Threshold P value Within Outside Count Column N % Count Column N % Missing Teeth No 230 85.2% 20 66.7% 0.01* Yes 40 14.8% 10 33.3% Pearsons’s correlation coefficient. *Significant correlation as P<0.05. Discussion This study aimed to evaluate the accuracy and reliability of an AI model in assessing maxillary crowding and spacing by comparing its performance to manual measurements on digital casts obtained from intraoral scans using OrthoCAD software. While manual measurement may introduce bias, it was necessary to validate the AI model against clinically accepted methods. The AI model addressed three key challenges in quantification. First, missing teeth were identified by comparing detected teeth to a reference list of permanent teeth, with substitutes determined using contralateral measurements or reference values [ 10 ]. Second, arch length was measured by drawing a smooth curve through dental landmarks. Third, scale calibration was based on the average visible tooth size in the intraoral photographs, standardizing spatial measurements across images. To reduce errors, the study excluded scans with distortions, non-orthogonal images, and cases involving deciduous, supernumerary, or retained teeth, focusing solely on permanent dentition. Future studies should aim to train the model to detect and analyze such excluded cases. Othman and Harradine [ 11 ] proposed 2 mm threshold as an appropriate benchmark for marking clinically significant TSALD. Accordingly, this study adopted the 2 mm threshold to validate the AI model, as differences within this range are unlikely to influence clinical decisions. The AI model achieved 90% accuracy, supporting its potential for clinical use in automated space analysis. These findings align with Ryu et al., who evaluated four AI models for crowding and extraction classification using intraoral photographs, with the VGG19 model achieving 92.2% accuracy [ 12 ]. Their slightly higher accuracy may be due to excluding cases with missing or unerupted teeth, which in our study contributed to larger discrepancies. Additionally, while Ryu et al. used the same images for both AI and human analysis, our study used intraoral scans for manual measurements, allowing 3D orientation in OrthoCAD for more precise landmark placement, explaining the variation in agreement. Moreover, their model did not distinguish between degrees of spacing, limiting clinical relevance. In contrast, our study applied ICON-based classifications, offering a more structured and generalizable framework. Similarly, Bardideh et al. reported higher AI accuracy than clinicians in molar (93.1%) and canine (89.1%) classification, but lower accuracy for overjet and overbite measurements [ 13 ]. This may be due to differences in image orientation; clinicians performed assessments clinically, while AI used static 2D images. Had both used 2D images, agreement may have improved, though not necessarily reliability. To address this, our study ensured that intraoral images analyzed by AI were orthogonal and distortion-free, enhancing accuracy and comparability with 3D manual analysis in OrthoCAD. The high correlation (> 0.92) between manual and AI-derived TSALD measurements shows the reliability and consistency of the AI model. Intraclass correlation coefficients (ICC) of 0.948 for crowding and 0.920 for spacing confirm strong agreement, reinforcing its clinical applicability. Narrow confidence intervals further demonstrate the precision of AI measurements, reducing uncertainty in treatment planning. The largest discrepancies occurred in mild crowding (< 2 mm) with a 7% difference, severe spacing (5.1–9 mm) with 5%, and moderate spacing (2.1–5 mm) with 3.3%. These results suggest the AI performs less accurately in borderline categories and extreme TSALD values. The narrow range of mild crowding (0.1–1.9 mm) means even small differences can shift a case between categories. Larger discrepancies in spacing may be linked to the AI's challenge in estimating cases with large spaces, particularly in cases involving missing teeth, which were more likely to exceed the error threshold. Studies on AI in orthodontics report high accuracy rates, such as 99.9% [ 14 ], 98% [ 15 ], and 99.4% [ 7 ]. Li et al. highlighted the efficiency of an AI model in classifying and archiving orthodontic records, showing that AI was 236 times faster than human experts [ 7 ]. Other studies have shown that AI models may be better in certain areas than others. Choi et al. reported 96% accuracy in diagnosing orthognathic surgery need but lower accuracy (91%) in determining specific surgical and extraction decisions [ 16 ]. Similarly, Jung et al. found 93% accuracy in extraction vs. non-extraction classification but 84% accuracy in extraction pattern identification [ 17 ]. Most AI studies focus on subjective assessments such as cervical vertebral maturation or basic objective tasks like image classification and extraction decisions based on cephalometric values. While valuable, these do not fully explore AI’s potential in complex orthodontic diagnostics. Our study distinguishes itself by challenging the AI model to perform precise quantitative tasks, including automatic tooth segmentation and millimetric assessments of crowding and spacing. It further incorporates arch length measurement, which introduces more complexity and variability than static segmentation. No previous research achieved this level of methodological advancement, making this study a significant step toward bridging AI innovation with clinical precision in orthodontics. Additionally, including a diverse patient sample and including images captured using different cameras, settings, and lighting conditions, reduces the risk of overfitting and improves the model’s performance in real-world clinical settings. Validating an AI model against a component of the ICON index is a key step toward developing a fully automated ICON-based diagnostic system. Once established, such a system could perform patient prioritization based on malocclusion complexity improving clinical organization. Integrating AI into space analysis replaces manual processes like pouring casts and taking measurements, reducing time, increasing accuracy, and minimizing human error caused by fatigue or subjectivity. Additionally, it offers educational value by helping train inexperienced practitioners and orthodontic residents to diagnose cases more accurately and consistently. Limitations Despite its strengths, this study has some limitations. The AI model was trained solely on permanent dentition, as mixed dentition requires different space analysis methods. Further training is needed for the model to identify deciduous, permanent, and supernumerary teeth, as well as retained roots, which were occasionally misclassified as erupted teeth. The model also struggled with poorly oriented or low-quality intraoral photographs, limiting its clinical applicability in varied settings. Cases with 0 mm crowding or spacing were absent, eliminating the possibility of using negative controls. Although rare, such cases would improve model validation, though the presence of many mild cases suggests the AI may detect ideal alignment. The study focused only on the maxillary arch due to its relevance in the ICON index; however, future research should include the mandibular arch to explore potential differences in accuracy. Conclusions The developed AI model achieved an overall accuracy of 90% for TSALD measurements in the maxillary arch compared to human observers. A very high correlation (> 0.92) between manual and AI TSALD measurements for crowding and spacing was observed. The largest discrepancies were observed in the mild crowding category (< 2 mm), severe spacing (5.1–9 mm), and moderate spacing (2.1–5 mm). The high correlation and accuracy of the AI model reinforce its potential as a reliable and efficient tool for orthodontic diagnostics enhancing diagnostic precision, reducing subjectivity, and simplifying clinical workflows while decreasing the time and effort required for manual analysis. Declarations Conflict of interest statement The authors have no relevant financial or non-financial interests to disclose. Funding No grants or any other funding support were received for conducting the present study. Data availability statement The data used and analyzed in this study are available from the corresponding author upon request. Human Ethics and Consent to participate Approval for the study was obtained by the Research Ethics Committee at the University of Sharjah; approval number (………). Informed consent was obtained from all individual participants included in the study approving the use of personal data and medical records including photographs and digital scans for the purpose of medical research. References Albalawi F, Alamoud KA (2022) Trends and application of artificial intelligence technology in orthodontic diagnosis and treatment planning: a review. Appl Sci 12(22):11864. https://doi.org/10.3390/app122211864 Bardideh E, Lal Alizadeh F, Amiri M et al (2024) Designing an artificial intelligence system for dental occlusion classification using intraoral photographs: a comparative analysis between artificial intelligence‑based and clinical diagnoses. Am J Orthod Dentofacial Orthop 166(2):125–137. https://doi.org/10.1016/j.ajodo.2024.03.012 Berkovitz BKB, Holland GR, Moxham BJ (2009) Oral anatomy, histology and embryology. 4th ed. Edinburgh: Mosby. https://doi.org/10.1093/EJO/CJY010 Chang AC (2020) Artificial intelligence in medicine: the future of personalized health care. In: Chang AC (ed) Intelligence-based medicine: data science, artificial intelligence, and human cognition in clinical medicine and healthcare, 1st edn. Academic Press, San Diego, pp 1–12 Choi HI, Jung SK, Baek SH et al (2019) Artificial intelligent model with neural network machine learning for the diagnosis of orthognathic surgery. J Craniofac Surg 30(7):1986–1989. https://doi.org/10.1097/SCS.0000000000005650 Daniels C, Richmond S (2000) The development of the Index of Complexity, Outcome and Need (ICON). J Orthod 27(2):149–162. https://doi.org/10.1093/ortho/27.2.149 Faber J, Faber C, Faber P (2019) Artificial intelligence in orthodontics. APOS Trends Orthod 9(4):201–205. https://doi.org/10.25259/APOS_123_2019 Jung SK, Kim TW (2016) New approach for the diagnosis of extractions with neural network machine learning. Am J Orthod Dentofacial Orthop 149(1):127–133. https://doi.org/10.1016/j.ajodo.2015.07.030 Khanagar SB, Al‑Ehaideb A, Vishwanathaiah S et al (2021) Scope and performance of artificial intelligence technology in orthodontic diagnosis, treatment planning, and clinical decision‑making: a systematic review. J Dent Sci 16(1):482–492. https://doi.org/10.1016/j.jds.2020.05.022 Li S, Guo Z, Lin J, Ying S (2022) Artificial intelligence for classifying and archiving orthodontic images. Biomed Res Int 2022:1473977. https://doi.org/10.1155/2022/1473977 Othman SA, Harradine NW (2007) Tooth size discrepancies in an orthodontic population. Angle Orthod 77(4):668–674. https://doi.org/10.2319/031406-102 Richmond S, Shaw WC, O’Brien KD et al (1992) The development of the PAR Index (Peer Assessment Rating): reliability and validity. Eur J Orthod 14(2):125–139. https://doi.org/10.1093/ejo/14.2.125 Ryu J, Kim YH, Kim TW et al (2023) Evaluation of artificial intelligence model for crowding categorization and extraction diagnosis using intraoral photographs. Sci Rep 13:5177. https://doi.org/10.1038/s41598-023-32514-7 Ryu J, Lee YS, Mo SP et al (2022) Application of deep learning artificial intelligence technique to the classification of clinical orthodontic photos. BMC Oral Health 22:454. https://doi.org/10.1186/s12903-022-02466-x Shaw WC, Richmond S, O’Brien KD, Brook P, Stephens CD (1991) Quality control in orthodontics: indices of treatment need and treatment standards. Br Dent J 170(3):107–112. https://doi.org/10.1038/sj.bdj.4807429 Talaat S, Kaboudan A, Talaat W et al (2021) The validity of an artificial intelligence application for assessment of orthodontic treatment need from clinical images. Semin Orthod 27(2):164–171. https://doi.org/10.1053/j.sodo.2021.05.012 Wallis C, McNamara C, Cunningham SJ et al (2014) How good are we at estimating crowding and how does it affect our treatment decisions? Eur J Orthod 36(4):465–470. https://doi.org/10.1093/ejo/cjt080 Additional Declarations No competing interests reported. Cite Share Download PDF Status: Under Review Version 1 posted Editorial decision: Revision requested 16 Oct, 2025 Reviews received at journal 30 Sep, 2025 Reviewers agreed at journal 26 Sep, 2025 Reviewers invited by journal 26 Sep, 2025 Editor assigned by journal 20 Jun, 2025 Submission checks completed at journal 20 Jun, 2025 First submitted to journal 15 Jun, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-6898322","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":522741657,"identity":"8073eb81-23bc-40f1-b32c-5cdd9e129443","order_by":0,"name":"Haneen Hatoum","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA8ElEQVRIiWNgGAWjYBACAwh1AEJ9bACRjI0HiNLCA1I7s4FBAkg1EK+FmResBW4pdmDOfvbgg4877sjbs/eYbrbdYVOn234YaEuNTTQuLZY9ecmGM888M+zhOWN2O/dMmoTZmUSglmNpuQ24HHYgx0yat+0wY49EDlBL22EJswNALYwNh3FrOf/G/PfftsP2YC2WIC3nHxLQciPHjJmx7XAiWAsjSMsNQrbceGMs2dt2OLnnzLGym71taZLbbgBtScDnl/M5hh9+th22bW9v3nbjZ5sNv9n59IcPPtTY4NSCAySQpnwUjIJRMApGARoAAM3eaQW1T42sAAAAAElFTkSuQmCC","orcid":"","institution":"University of Sharjah","correspondingAuthor":true,"prefix":"","firstName":"Haneen","middleName":"","lastName":"Hatoum","suffix":""},{"id":522741658,"identity":"8ff081f5-8d0d-4a22-9603-6d8968c17677","order_by":1,"name":"Wael Talaat","email":"","orcid":"","institution":"University of Sharjah","correspondingAuthor":false,"prefix":"","firstName":"Wael","middleName":"","lastName":"Talaat","suffix":""},{"id":522741659,"identity":"75904f5b-066e-4dde-b6ea-420beb567c52","order_by":2,"name":"Ahmed Kaboudan","email":"","orcid":"","institution":"El Shorouk Academy","correspondingAuthor":false,"prefix":"","firstName":"Ahmed","middleName":"","lastName":"Kaboudan","suffix":""},{"id":522741660,"identity":"bb4fce1c-3205-470a-bbd3-2d2c8a9ce465","order_by":3,"name":"Aasem Hamed","email":"","orcid":"","institution":"","correspondingAuthor":false,"prefix":"","firstName":"Aasem","middleName":"","lastName":"Hamed","suffix":""},{"id":522741661,"identity":"46be243a-c02a-4fc6-bb98-d63949d80872","order_by":4,"name":"Engy Mahmoud","email":"","orcid":"","institution":"","correspondingAuthor":false,"prefix":"","firstName":"Engy","middleName":"","lastName":"Mahmoud","suffix":""},{"id":522741662,"identity":"411e6e13-2598-4154-95b5-3b16579d1b6d","order_by":5,"name":"Sameh Talaat","email":"","orcid":"","institution":"Future University in Egypt","correspondingAuthor":false,"prefix":"","firstName":"Sameh","middleName":"","lastName":"Talaat","suffix":""},{"id":522741663,"identity":"a7052dc9-7958-4b6f-a0f5-6db3cfa60e16","order_by":6,"name":"Shishir Shetty","email":"","orcid":"","institution":"University of Sharjah","correspondingAuthor":false,"prefix":"","firstName":"Shishir","middleName":"","lastName":"Shetty","suffix":""},{"id":522741664,"identity":"674df83e-7fc1-4962-b759-0e202658e6cf","order_by":7,"name":"Mais Sadek","email":"","orcid":"","institution":"University of Sharjah","correspondingAuthor":false,"prefix":"","firstName":"Mais","middleName":"","lastName":"Sadek","suffix":""}],"badges":[],"createdAt":"2025-06-15 12:38:20","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-6898322/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6898322/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":93241162,"identity":"1e83cf6a-8134-4b8c-89e5-bf39fe648606","added_by":"auto","created_at":"2025-10-10 14:50:50","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1912151,"visible":true,"origin":"","legend":"","description":"","filename":"ManuscriptFinal.docx","url":"https://assets-eu.researchsquare.com/files/rs-6898322/v1/89776f59f90d5978c48e9481.docx"},{"id":93241157,"identity":"3f6999ad-cc2f-4c41-850e-aa4279c3daa1","added_by":"auto","created_at":"2025-10-10 14:50:50","extension":"json","order_by":1,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":9402,"visible":true,"origin":"","legend":"","description":"","filename":"c246525b564b4c2bac81c1806b9adb06.json","url":"https://assets-eu.researchsquare.com/files/rs-6898322/v1/137c5e8128e621f7379dccfb.json"},{"id":93243425,"identity":"3f379a26-2731-437f-a800-ac1427e94d9f","added_by":"auto","created_at":"2025-10-10 14:58:50","extension":"xml","order_by":2,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":77363,"visible":true,"origin":"","legend":"","description":"","filename":"c246525b564b4c2bac81c1806b9adb061enriched.xml","url":"https://assets-eu.researchsquare.com/files/rs-6898322/v1/788499f20bf5d1547cdc273c.xml"},{"id":93244400,"identity":"63072885-6906-4b2c-b6f3-1d734c95d7ad","added_by":"auto","created_at":"2025-10-10 15:06:50","extension":"png","order_by":3,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":144906,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-6898322/v1/a38a128d932ac5e8a50d28db.png"},{"id":93243423,"identity":"78a0601b-accd-4ec5-8807-fef2fcc161bd","added_by":"auto","created_at":"2025-10-10 14:58:50","extension":"jpeg","order_by":4,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":871394,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage2.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-6898322/v1/e7096d2ababd93288c1baca3.jpeg"},{"id":93241161,"identity":"55bf2b70-4cd4-4c90-88a3-9fcf5e5ceefd","added_by":"auto","created_at":"2025-10-10 14:50:50","extension":"png","order_by":5,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":55538,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-6898322/v1/ede02df6bfa8020c671f9315.png"},{"id":93243428,"identity":"4f5b74a1-1edf-42be-a4a9-b02952ca765c","added_by":"auto","created_at":"2025-10-10 14:58:50","extension":"png","order_by":6,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":299181,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-6898322/v1/c87a1ef574e6828cf89929ed.png"},{"id":93241175,"identity":"b5b36b4d-d805-4eb3-811d-f1ac98e39e00","added_by":"auto","created_at":"2025-10-10 14:50:50","extension":"png","order_by":7,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":205708,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-6898322/v1/6128a62f2d333d058e19ec57.png"},{"id":93241168,"identity":"b774e336-2357-4452-a167-016f6ead11df","added_by":"auto","created_at":"2025-10-10 14:50:50","extension":"png","order_by":8,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":279973,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage6.png","url":"https://assets-eu.researchsquare.com/files/rs-6898322/v1/d1e0a4ff89e51eb56cce9228.png"},{"id":93246556,"identity":"90a672a7-996f-4350-98bc-5f7d9906a754","added_by":"auto","created_at":"2025-10-10 15:14:50","extension":"png","order_by":9,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":16617,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-6898322/v1/e144f18030ebe33375be2eb0.png"},{"id":93241169,"identity":"0be43b1d-7e58-4ca9-9f76-c4238942c72a","added_by":"auto","created_at":"2025-10-10 14:50:50","extension":"png","order_by":10,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":87779,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-6898322/v1/4d43bdbb189492e21d264e70.png"},{"id":93246558,"identity":"8df286a3-1959-44a9-9393-0e713e0bf7fe","added_by":"auto","created_at":"2025-10-10 15:14:50","extension":"png","order_by":11,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":17982,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-6898322/v1/1d157628dfdb179861056bd4.png"},{"id":93244403,"identity":"aecea001-f559-4fe2-a2f2-0260244f61ec","added_by":"auto","created_at":"2025-10-10 15:06:50","extension":"png","order_by":12,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":57135,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-6898322/v1/0e624962ff894c4459372dc6.png"},{"id":93241173,"identity":"37312412-ad59-4e00-9f7a-1734797fff2d","added_by":"auto","created_at":"2025-10-10 14:50:50","extension":"png","order_by":13,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":40086,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-6898322/v1/e14f72ca4390b188bb085c06.png"},{"id":93247017,"identity":"15d29c41-45fe-4f77-95c7-c20ea9cfcf34","added_by":"auto","created_at":"2025-10-10 15:22:50","extension":"png","order_by":14,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":32336,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage6.png","url":"https://assets-eu.researchsquare.com/files/rs-6898322/v1/cb4d64b9ea0ad8400878159a.png"},{"id":93241177,"identity":"e4afe47b-3cf9-4348-8bd8-228a6fcf49a9","added_by":"auto","created_at":"2025-10-10 14:50:50","extension":"xml","order_by":15,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":75903,"visible":true,"origin":"","legend":"","description":"","filename":"c246525b564b4c2bac81c1806b9adb061structuring.xml","url":"https://assets-eu.researchsquare.com/files/rs-6898322/v1/e1fc0c42b72f5b2f9d2ce263.xml"},{"id":93241176,"identity":"cd4c0c39-2375-4bce-8cbe-e6fcb738cf2d","added_by":"auto","created_at":"2025-10-10 14:50:50","extension":"html","order_by":16,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":84250,"visible":true,"origin":"","legend":"","description":"","filename":"earlyproof.html","url":"https://assets-eu.researchsquare.com/files/rs-6898322/v1/0a93f9663c48058b2298d321.html"},{"id":93241155,"identity":"ba845879-8186-429a-a678-7d1f8e23a3d5","added_by":"auto","created_at":"2025-10-10 14:50:50","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":43452,"visible":true,"origin":"","legend":"\u003cp\u003eBalanced model labels (class names)\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-6898322/v1/76d89d26905e2fe7b6db9090.png"},{"id":93241160,"identity":"bf30f0ed-b5a3-4a38-b5b7-4946f03b5602","added_by":"auto","created_at":"2025-10-10 14:50:50","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":320739,"visible":true,"origin":"","legend":"\u003cp\u003eModel output images. Mesial and Distal points (left) and Arch length (right)\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-6898322/v1/b75f99732e9478b44ed16e85.png"},{"id":93241159,"identity":"423c54a9-fa4e-4450-a865-392cc4e98638","added_by":"auto","created_at":"2025-10-10 14:50:50","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":46343,"visible":true,"origin":"","legend":"\u003cp\u003eConfusion matrix normalized\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-6898322/v1/fe5c9c82deaec82225587a77.png"},{"id":93243420,"identity":"9d629708-bb80-4a5c-a7a2-4d947b2186f6","added_by":"auto","created_at":"2025-10-10 14:58:50","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":109075,"visible":true,"origin":"","legend":"\u003cp\u003eModel training/validation graphs\u003c/p\u003e","description":"","filename":"4.png","url":"https://assets-eu.researchsquare.com/files/rs-6898322/v1/722e391a8deaa7a5d2d1c958.png"},{"id":93241163,"identity":"3a101590-602c-4cdf-abd4-c14e923ee0b2","added_by":"auto","created_at":"2025-10-10 14:50:50","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":260090,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003ea \u003c/strong\u003eContra-lateral mesiodistal width measured due to missing tooth\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eb \u003c/strong\u003eArch length determination\u003c/p\u003e","description":"","filename":"5.png","url":"https://assets-eu.researchsquare.com/files/rs-6898322/v1/3966c4f0bdda1bb94c6f8345.png"},{"id":93375437,"identity":"7d2a8d7b-6a9d-47fd-83cb-794ba19dba69","added_by":"auto","created_at":"2025-10-13 08:09:29","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1715331,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6898322/v1/17e0d88e-55fa-4a4f-a63a-ceba1216ee9e.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Maxillary Crowding and Spacing: Validation of an Artificial Intelligence Model vs. Digitally-Assisted Human Observer","fulltext":[{"header":"Introduction","content":"\u003cp\u003eOrthodontic diagnosis begins with a thorough evaluation of the patient\u0026rsquo;s medical and dental history, clinical examination, and essential diagnostic records, including photographs, study models, and radiographs [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]. The orthodontist then analyzes this data to create a prioritized problem list, incorporating patient-specific needs to determine appropriate treatment options. However, this process is complex, time-consuming, and requires careful evaluation of multiple parameters to ensure optimal management of malocclusion [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eTo standardize treatment planning, various indices have been developed to assess orthodontic treatment needs, including the Peer Assessment Rating (PAR) index [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e] and the Index of Treatment Need (IOTN) [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]. However, these indices have limitations, such as failing to describe treatment complexity and over- or underestimating treatment goals [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]. The PAR index has been criticized for overlooking poor finishes and being overly harsh on treatments with limited aims, while the IOTN does not necessarily reflect treatment complexity [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e, \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]. To address these shortcomings, the Index of Complexity, Outcome, and Need (ICON) was developed in 2000 by Daniels and Richmond [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]. Unlike previous indices, ICON integrates five components, including dental aesthetics (IOTN aesthetic component), crossbite, deep/open bite (PAR), upper arch crowding/spacing (five-point scale), and buccal segment anteroposterior relationship (PAR), providing a unified measure of treatment need, complexity, and outcome acceptability [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eTooth size arch length discrepancy (TSALD) is used to quantify dental crowding and spacing by measuring the difference between basal arch length and the combined mesiodistal widths of incisors, canines, and premolars. Various measurement techniques exist, including manual methods such as Lundstr\u0026ouml;m\u0026rsquo;s segmental analysis and Carey\u0026rsquo;s brass wire technique, visual estimation, and digital analysis. While visual estimation is convenient, it is highly subjective, with studies showing discrepancies of up to 15 mm among orthodontists [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]. Digital methods such as using OrthoCAD (Cadent, Fairview, NJ, USA) offer an effective alternative, yet digital measurements require manual identification of the tooth contact points for calculation of TSALD. This can be time-consuming, with a high risk of error due to fatigue [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eIn recent years, Artificial Intelligence (AI) has emerged as a powerful tool in orthodontics, offering potential solutions to the limitations of traditional methods. AI mimics human intelligence by processing large datasets and identifying patterns to optimize decision-making [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e]. Since its introduction into healthcare in the early 2000s, AI has played a key role in enhancing accuracy and efficiency in medical diagnosis and treatment planning [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]. Within AI, Machine Learning (ML) enables computers to learn from data without explicit programming, while Deep Learning (DL) refines pattern recognition and has outperformed traditional ML techniques in tasks such as classification, segmentation, and detection [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]. The use of AI in orthodontics has expanded significantly, with numerous studies exploring its applications. Research has examined AI models for cephalometric landmark identification, bone age estimation using cervical vertebra and hand-wrist radiographs, and orthodontic tooth extraction decision-making [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e]. Additionally, AI has been studied for palatal shape analysis, automated skeletal classification, and orthognathic surgery diagnosis and planning [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eDespite these advancements, a significant gap remains in automating the quantification of crowding and spacing for evaluating orthodontic treatment need and complexity. No prior study has applied an AI model to assess these factors in alignment with the ICON index. This study takes a pioneering step toward validating an AI model against one component of the ICON index, laying the foundation for a fully automated ICON-based assessment system.\u003c/p\u003e\u003cp\u003eThe aim of this study was to develop an automated AI model capable of quantifying crowding and spacing in the upper arch and to validate its accuracy by comparing its results with those of human observers. The null hypothesis is that there would be no statistically significant differences between the AI model and human observers in quantifying crowding and spacing in the upper arch. This research aimed to bridge the gap between AI-based automation and standardized orthodontic assessment, ultimately contributing to more efficient and precise treatment planning.\u003c/p\u003e"},{"header":"Materials and methods","content":"\u003cp\u003eApproval for the study was obtained by the \u0026hellip;\u0026hellip;\u0026hellip;.; approval number (\u0026hellip;\u0026hellip;\u0026hellip;..). Consent for the use of patient records from the university database was already obtained from an informed consent signed by the patients prior to initiating orthodontic treatment in the Department of Orthodontics at the University of Sharjah.\u003c/p\u003e\n\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e\n \u003ch2\u003eDataset\u003c/h2\u003e\n \u003cp\u003eThe development of the AI model was conducted in three phases: training, validation, and testing against a human observer. A total of 831 upper intraoral images from orthodontic records were used to train and validate the model for measuring the Tooth Size\u0026ndash;Arch Length Discrepancy (TSALD) in the upper arch. For testing, a separate dataset comprising upper intraoral images and corresponding intraoral scans from 300 patients was used to compare the model\u0026rsquo;s performance with that of a human observer. All records were collected from patients undergoing orthodontic treatment at the Orthodontic Department, College of Dental Medicine, University of Sharjah, between 2022 and 2024. Intraoral photographs were exported from the Dolphin Imaging \u0026amp; Management Solutions software (v.12, Chatsworth, CA, USA), and pre-treatment intraoral scans were retrieved from the iTero Element 5D scanner system (Align Technology, San Jose, CA, USA). The AI model was trained to classify teeth and measure TSALD using the intraoral images, while the human observer (\u0026hellip;.) performed the same measurements on the iTero scans. Inclusion criteria were full permanent dentition (excluding third molars), pre-treatment scans including both arches locked in occlusion. Exclusion criteria included primary or mixed dentition, retained primary teeth, scans with defects, supernumerary teeth, retained roots, and non-orthogonal intraoral images.\u003c/p\u003e\n\u003c/div\u003e\n\u003ch3\u003eTraining of the model\u003c/h3\u003e\n\u003cp\u003eThe training dataset consisted of 740 images. The YOLOv8 Pose model; a version of the YOLO object detection system adapted for identifying specific points, was used to detect teeth and locate their mesial and distal points. The model was trained using a multi-task loss function that included object loss, which helps the model learn whether a tooth is present, classification loss, which teaches the model to identify the type of tooth, and localization loss, which focuses on accurately predicting the position of the keypoints. Predefined anchor boxes were used to help the model detect teeth of different sizes. The AI was trained specifically on upper occlusal images and focused on two main tasks: object detection, where each tooth is recognized by its class name and a confidence score (using a threshold of 0.75), and point detection, where the mesial and distal points of each tooth are located. The dataset used for training was balanced, helping the model learn to identify different types of teeth equally and reducing bias, which improves how well the model performs on new cases \u003cstrong\u003e(\u003c/strong\u003eFig. \u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003e\u003cstrong\u003e)\u003c/strong\u003e. After training, the model\u0026rsquo;s results were used to measure space conditions in the upper arch, including crowding (when there is not enough space), spacing (when there is extra space), and arch length (the total space available along the dental arch). Arch length was calculated by drawing a smooth curve through the key points of the teeth, while the space required was measured by adding the mesiodistal widths of all teeth \u003cstrong\u003e(\u003c/strong\u003eFig. \u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003e\u003cstrong\u003e)\u003c/strong\u003e. To ensure accurate measurements, a calibration step was performed by estimating the average mesiodistal width of the visible teeth in the intraoral occlusal image and using it as a reference scale to convert pixel measurements into millimeters.\u003c/p\u003e\n\u003ch3\u003eValidation of the model\u003c/h3\u003e\n\u003cp\u003eA validation set of 92 cases was used. One key issue identified was the model\u0026apos;s limited ability to accurately identify missing teeth and account for them in the space analysis. To address this, the model underwent additional training focused on recognizing and appropriately accounting for missing teeth. The model\u0026apos;s performance metrics demonstrated significant improvements across various parameters \u003cstrong\u003e(\u003c/strong\u003eFigs. \u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003e \u0026amp; \u003cspan class=\"InternalRef\"\u003e4\u003c/span\u003e\u003cstrong\u003e)\u003c/strong\u003e.\u003c/p\u003e\n\u003ch3\u003eTesting of the model\u003c/h3\u003e\n\u003cp\u003eThe required sample size for testing the AI model was calculated using a web-based calculator (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://wnarifin.github.io/ssc/sskappa.html\u003c/span\u003e\u003c/span\u003e). Based on a minimum acceptable Kappa of 0.8, an expected Kappa of 0.9, an outcome proportion of 0.5, a 5% significance level, and 80% power, the minimum required sample was 283 records. Accordingly, records of 300 patients were extracted. Sample demographics (age, gender, and race) were obtained from the aXium software (Exan Group, Vancouver, BC, Canada). Measurements of crowding and spacing were conducted on OrthoCAD software (Cadent, Fairview, NJ, USA), \u003cstrong\u003e(\u003c/strong\u003eFig. \u003cspan class=\"InternalRef\"\u003e5\u003c/span\u003ea and \u003cspan class=\"InternalRef\"\u003e5\u003c/span\u003eb\u003cstrong\u003e)\u003c/strong\u003e. For missing or impacted teeth, the width of the contralateral tooth was used; if both were absent, average mesiodistal widths from Berkovitz et al., were applied [\u003cspan class=\"CitationRef\"\u003e10\u003c/span\u003e]. Arch length was measured from the mesial of the first molar on one side to the mesial of the first molar on the other along the arch. Arch length measurement followed the majority of teeth in the arch regardless of the presence of any blocked-out teeth. TSALD was automatically calculated by subtracting required space from available space, and results were categorized based on the ICON index [\u003cspan class=\"CitationRef\"\u003e5\u003c/span\u003e]. These measurements served as the reference standard. The same cases were then assessed using the AI model, with an acceptable TSALD difference threshold set at 2 mm [\u003cspan class=\"CitationRef\"\u003e11\u003c/span\u003e].\u003c/p\u003e\n\u003ch3\u003eCalibration of the human observer and error measurement\u003c/h3\u003e\n\u003cp\u003eAll measurements were performed by the primary investigator (H.H) using the OrthoCAD software after receiving proper training. To determine intra-observer reliability, twenty scans of the included sample were randomly selected to be measured again two weeks after the first assessment. To determine inter-observer reliability, the same scans were measured by another examiner. Error measurement was assessed using the intraclass correlation coefficient (ICC), showing near-perfect intraobserver (ICC\u0026thinsp;=\u0026thinsp;0.998, 95% CI: 0.995\u0026ndash;0.999) and interobserver reliability (ICC\u0026thinsp;=\u0026thinsp;0.996, 95% CI: 0.993\u0026ndash;1.000; both p\u0026thinsp;=\u0026thinsp;0.0001).\u003c/p\u003e\n\u003cdiv id=\"Sec8\" class=\"Section2\"\u003e\n \u003ch2\u003eStatistical analysis\u003c/h2\u003e\n \u003cp\u003eStatistical analysis was performed with SPSS 20\u0026reg;, Graph Pad Prism\u0026reg; and Microsoft Excel 2016. Quantitative data were explored for normality by using Shapiro Wilk Normality test and Kolmogorov test and were presented as minimum, maximum, means and standard deviation (SD) values. Qualitative data were presented as frequency and distribution, and comparisons were performed by using Fishers Exact test. Correlation between Manual and AI-measured TSALD was evaluated using Pearsons\u0026rsquo;s correlation coefficient. Interobserver and Intraobserver reliability was evaluated by using ICC (Interclass correlation coefficient).\u003c/p\u003e\n\u003c/div\u003e"},{"header":"Results","content":"\u003cp\u003eThe sample had a balanced gender distribution (53.3% male, 46.7% female). Most participants were Arab, followed by Asian, African, Caucasian, and Persian. The majority (83.3%) had no missing teeth, while 16.7% had at least one missing tooth. Regarding dental categorization, 63% had crowding, and 37% had spacing.\u003c/p\u003e\n\u003cp\u003eManual and AI measurements of TSALD were compared across crowding and spacing parameters \u003cstrong\u003e(\u003c/strong\u003eTable \u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003e\u003cstrong\u003e)\u003c/strong\u003e. Their correlations are detailed in Table \u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003e, showing strong alignment and suggesting high reliability and potential interchangeability. Both crowding and spacing demonstrated statistically significant correlations above 0.92. For crowding, the ICC was 0.948 between manual (-3.84\u0026thinsp;\u0026plusmn;\u0026thinsp;3.77) and AI measurements (-4.29\u0026thinsp;\u0026plusmn;\u0026thinsp;4.17), with a 95% CI of 0.931\u0026ndash;0.961 (p\u0026thinsp;=\u0026thinsp;0.0001). For spacing, the ICC was 0.920 between manual (3.93\u0026thinsp;\u0026plusmn;\u0026thinsp;3.19) and AI values (3.76\u0026thinsp;\u0026plusmn;\u0026thinsp;3.64), with a 95% CI of 0.884\u0026ndash;0.945 (p\u0026thinsp;=\u0026thinsp;0.0001).\u003c/p\u003e\n\u003cdiv class=\"gridtable\"\u003e\n \u003ctable id=\"Tab1\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003eDescriptive statistics of manual TSALD, AI TSALD, and the difference between them\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\u0026nbsp;\u003c/th\u003e\n \u003cth align=\"left\"\u003e\u0026nbsp;\u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eMinimum\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eMaximum\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eMedian\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eMean\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eStandard Deviation\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" rowspan=\"3\"\u003e\n \u003cp\u003eCrowding\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eManual\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e-15.60\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u0026minus;\u0026thinsp;.10\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e-2.40\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e-3.84\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e3.77\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eAI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e-18.31\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.42\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e-3.02\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e-4.29\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e4.17\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eDifference\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e7.42\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e1.04\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e1.28\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e1.30\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" rowspan=\"3\"\u003e\n \u003cp\u003eSpacing\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eManual\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e.10\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e13.50\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e3.50\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e3.93\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e3.19\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eAI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e-5.93\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e12.67\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e3.63\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e3.76\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e3.64\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eDifference\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e8.13\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e1.00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e1.34\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e1.29\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003cp\u003eShapiro Wilk Normality test and Kolmogorov test. TSALD; Tooth size arch length discrepancy, AI; artificial intelligence.\u003c/p\u003e\n\u003cdiv class=\"gridtable\"\u003e\n \u003ctable id=\"Tab2\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003eCorrelation between manual TSALD and AI TSALD regarding crowding and spacing\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" rowspan=\"2\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\" rowspan=\"2\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\" rowspan=\"2\"\u003e\n \u003cp\u003eM\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" rowspan=\"2\"\u003e\n \u003cp\u003eSD\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" rowspan=\"2\"\u003e\n \u003cp\u003eIntraclass Correlation\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003e95% Confidence Interval\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"4\"\u003e\n \u003cp\u003eF Test with\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eLower Bound\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eUpper Bound\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eValue\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003edf1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003edf2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eP value\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" rowspan=\"2\"\u003e\n \u003cp\u003e\u003cstrong\u003eCrowding\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eManual\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e-3.84\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e3.77\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" rowspan=\"2\"\u003e\n \u003cp\u003e0.948\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" rowspan=\"2\"\u003e\n \u003cp\u003e0.931\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" rowspan=\"2\"\u003e\n \u003cp\u003e0.961\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" rowspan=\"2\"\u003e\n \u003cp\u003e19.203\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" rowspan=\"2\"\u003e\n \u003cp\u003e188\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" rowspan=\"2\"\u003e\n \u003cp\u003e188\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" rowspan=\"2\"\u003e\n \u003cp\u003e0.0001*\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eAI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e-4.29\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e4.17\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" rowspan=\"2\"\u003e\n \u003cp\u003e\u003cstrong\u003eSpacing\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eManual\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e3.93\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e3.19\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" rowspan=\"2\"\u003e\n \u003cp\u003e0.920\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" rowspan=\"2\"\u003e\n \u003cp\u003e0.884\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" rowspan=\"2\"\u003e\n \u003cp\u003e0.945\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" rowspan=\"2\"\u003e\n \u003cp\u003e12.524\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" rowspan=\"2\"\u003e\n \u003cp\u003e110\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" rowspan=\"2\"\u003e\n \u003cp\u003e110\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" rowspan=\"2\"\u003e\n \u003cp\u003e0.0001*\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eAI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e3.76\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e3.64\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003cp\u003ePearsons\u0026rsquo;s correlation coefficient. *Significant correlation as P \u0026le;0.05, TSALD; Tooth size arch length discrepancy, AI; artificial intelligence.\u003c/p\u003e\n\u003cp\u003eUsing a 2 mm threshold, AI achieved 90% accuracy compared to human observers, with 270 cases within and 30 cases outside the threshold. Category distributions are shown in Table \u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003e. Discrepancies beyond the threshold were significantly associated with the presence of missing teeth \u003cstrong\u003e(\u003c/strong\u003eTable \u003cspan class=\"InternalRef\"\u003e4\u003c/span\u003e\u003cstrong\u003e)\u003c/strong\u003e.\u003c/p\u003e\n\u003cdiv class=\"gridtable\"\u003e\n \u003ctable id=\"Tab3\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003eDistribution \u0026amp; accuracy of different crowding \u0026amp; spacing categories among manual and AI\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003eManual ICON\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003eAI ICON\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" rowspan=\"2\"\u003e\n \u003cp\u003eAccuracy\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eCount\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eColumn N %\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eCount\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eColumn N %\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cstrong\u003eCrowding\u0026thinsp;\u0026lt;\u0026thinsp;2mm\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e83\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e27.7%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e62\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e20.7%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" rowspan=\"10\"\u003e\n \u003cp\u003e90%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cstrong\u003eCrowding 2.1mm-5mm\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e52\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e17.3%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e63\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e21.0%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cstrong\u003eCrowding 5.1mm-9mm\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e28\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e9.3%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e40\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e13.3%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cstrong\u003eCrowding 9.1mm-13mm\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e20\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e6.7%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e13\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e4.3%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cstrong\u003eCrowding 13.1mm-17mm\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e2.0%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.7%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cstrong\u003eCrowding\u0026thinsp;\u0026gt;\u0026thinsp;17mm or impacted teeth\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.0%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.7%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cstrong\u003eSpacing up to 2mm\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e41\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e13.7%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e40\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e13.3%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cstrong\u003eSpacing 2.1mm-5mm\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e37\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e12.3%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e27\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e9.0%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cstrong\u003eSpacing 5.1mm-9mm\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e24\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e8.0%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e39\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e13.0%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cstrong\u003eSpacing\u0026thinsp;\u0026gt;\u0026thinsp;9mm\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e3.0%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e2.0%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003cp\u003eFischer\u0026rsquo;s Exact test. AI; artificial intelligence, ICON; Index of complexity, outcome, and need.\u003c/p\u003e\n\u003cdiv class=\"gridtable\"\u003e\n \u003ctable id=\"Tab4\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 4\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003eAssociation between missing teeth and the difference in TSALD between manual and AI\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colspan=\"2\" rowspan=\"3\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"4\"\u003e\n \u003cp\u003eThreshold\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" rowspan=\"3\"\u003e\n \u003cp\u003eP value\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003eWithin\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003eOutside\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eCount\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eColumn N %\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eCount\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eColumn N %\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" rowspan=\"2\"\u003e\n \u003cp\u003e\u003cstrong\u003eMissing Teeth\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e230\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e85.2%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e20\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e66.7%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" rowspan=\"2\"\u003e\n \u003cp\u003e0.01*\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e40\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e14.8%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e10\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e33.3%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003cp\u003ePearsons\u0026rsquo;s correlation coefficient. *Significant correlation as P\u0026lt;0.05.\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eThis study aimed to evaluate the accuracy and reliability of an AI model in assessing maxillary crowding and spacing by comparing its performance to manual measurements on digital casts obtained from intraoral scans using OrthoCAD software. While manual measurement may introduce bias, it was necessary to validate the AI model against clinically accepted methods.\u003c/p\u003e\u003cp\u003eThe AI model addressed three key challenges in quantification. First, missing teeth were identified by comparing detected teeth to a reference list of permanent teeth, with substitutes determined using contralateral measurements or reference values [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]. Second, arch length was measured by drawing a smooth curve through dental landmarks. Third, scale calibration was based on the average visible tooth size in the intraoral photographs, standardizing spatial measurements across images. To reduce errors, the study excluded scans with distortions, non-orthogonal images, and cases involving deciduous, supernumerary, or retained teeth, focusing solely on permanent dentition. Future studies should aim to train the model to detect and analyze such excluded cases.\u003c/p\u003e\u003cp\u003eOthman and Harradine [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e] proposed 2 mm threshold as an appropriate benchmark for marking clinically significant TSALD. Accordingly, this study adopted the 2 mm threshold to validate the AI model, as differences within this range are unlikely to influence clinical decisions. The AI model achieved 90% accuracy, supporting its potential for clinical use in automated space analysis. These findings align with Ryu et al., who evaluated four AI models for crowding and extraction classification using intraoral photographs, with the VGG19 model achieving 92.2% accuracy [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e]. Their slightly higher accuracy may be due to excluding cases with missing or unerupted teeth, which in our study contributed to larger discrepancies. Additionally, while Ryu et al. used the same images for both AI and human analysis, our study used intraoral scans for manual measurements, allowing 3D orientation in OrthoCAD for more precise landmark placement, explaining the variation in agreement. Moreover, their model did not distinguish between degrees of spacing, limiting clinical relevance. In contrast, our study applied ICON-based classifications, offering a more structured and generalizable framework. Similarly, Bardideh et al. reported higher AI accuracy than clinicians in molar (93.1%) and canine (89.1%) classification, but lower accuracy for overjet and overbite measurements [\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e]. This may be due to differences in image orientation; clinicians performed assessments clinically, while AI used static 2D images. Had both used 2D images, agreement may have improved, though not necessarily reliability. To address this, our study ensured that intraoral images analyzed by AI were orthogonal and distortion-free, enhancing accuracy and comparability with 3D manual analysis in OrthoCAD.\u003c/p\u003e\u003cp\u003eThe high correlation (\u0026gt;\u0026thinsp;0.92) between manual and AI-derived TSALD measurements shows the reliability and consistency of the AI model. Intraclass correlation coefficients (ICC) of 0.948 for crowding and 0.920 for spacing confirm strong agreement, reinforcing its clinical applicability. Narrow confidence intervals further demonstrate the precision of AI measurements, reducing uncertainty in treatment planning. The largest discrepancies occurred in mild crowding (\u0026lt;\u0026thinsp;2 mm) with a 7% difference, severe spacing (5.1\u0026ndash;9 mm) with 5%, and moderate spacing (2.1\u0026ndash;5 mm) with 3.3%. These results suggest the AI performs less accurately in borderline categories and extreme TSALD values. The narrow range of mild crowding (0.1\u0026ndash;1.9 mm) means even small differences can shift a case between categories. Larger discrepancies in spacing may be linked to the AI's challenge in estimating cases with large spaces, particularly in cases involving missing teeth, which were more likely to exceed the error threshold.\u003c/p\u003e\u003cp\u003eStudies on AI in orthodontics report high accuracy rates, such as 99.9% [\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e], 98% [\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e], and 99.4% [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]. Li et al. highlighted the efficiency of an AI model in classifying and archiving orthodontic records, showing that AI was 236 times faster than human experts [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]. Other studies have shown that AI models may be better in certain areas than others. Choi et al. reported 96% accuracy in diagnosing orthognathic surgery need but lower accuracy (91%) in determining specific surgical and extraction decisions [\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e]. Similarly, Jung et al. found 93% accuracy in extraction vs. non-extraction classification but 84% accuracy in extraction pattern identification [\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eMost AI studies focus on subjective assessments such as cervical vertebral maturation or basic objective tasks like image classification and extraction decisions based on cephalometric values. While valuable, these do not fully explore AI\u0026rsquo;s potential in complex orthodontic diagnostics. Our study distinguishes itself by challenging the AI model to perform precise quantitative tasks, including automatic tooth segmentation and millimetric assessments of crowding and spacing. It further incorporates arch length measurement, which introduces more complexity and variability than static segmentation. No previous research achieved this level of methodological advancement, making this study a significant step toward bridging AI innovation with clinical precision in orthodontics. Additionally, including a diverse patient sample and including images captured using different cameras, settings, and lighting conditions, reduces the risk of overfitting and improves the model\u0026rsquo;s performance in real-world clinical settings.\u003c/p\u003e\u003cp\u003eValidating an AI model against a component of the ICON index is a key step toward developing a fully automated ICON-based diagnostic system. Once established, such a system could perform patient prioritization based on malocclusion complexity improving clinical organization. Integrating AI into space analysis replaces manual processes like pouring casts and taking measurements, reducing time, increasing accuracy, and minimizing human error caused by fatigue or subjectivity. Additionally, it offers educational value by helping train inexperienced practitioners and orthodontic residents to diagnose cases more accurately and consistently.\u003c/p\u003e\u003cdiv id=\"Sec11\" class=\"Section2\"\u003e\u003ch2\u003eLimitations\u003c/h2\u003e\u003cp\u003eDespite its strengths, this study has some limitations. The AI model was trained solely on permanent dentition, as mixed dentition requires different space analysis methods. Further training is needed for the model to identify deciduous, permanent, and supernumerary teeth, as well as retained roots, which were occasionally misclassified as erupted teeth. The model also struggled with poorly oriented or low-quality intraoral photographs, limiting its clinical applicability in varied settings. Cases with 0 mm crowding or spacing were absent, eliminating the possibility of using negative controls. Although rare, such cases would improve model validation, though the presence of many mild cases suggests the AI may detect ideal alignment. The study focused only on the maxillary arch due to its relevance in the ICON index; however, future research should include the mandibular arch to explore potential differences in accuracy.\u003c/p\u003e\u003c/div\u003e"},{"header":"Conclusions","content":"\u003cp\u003e\u003cul\u003e\u003cli\u003e\u003cp\u003eThe developed AI model achieved an overall accuracy of 90% for TSALD measurements in the maxillary arch compared to human observers.\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eA very high correlation (\u0026gt;\u0026thinsp;0.92) between manual and AI TSALD measurements for crowding and spacing was observed.\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eThe largest discrepancies were observed in the mild crowding category (\u0026lt;\u0026thinsp;2 mm), severe spacing (5.1\u0026ndash;9 mm), and moderate spacing (2.1\u0026ndash;5 mm).\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eThe high correlation and accuracy of the AI model reinforce its potential as a reliable and efficient tool for orthodontic diagnostics enhancing diagnostic precision, reducing subjectivity, and simplifying clinical workflows while decreasing the time and effort required for manual analysis.\u003c/p\u003e\u003c/li\u003e\u003c/ul\u003e\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eConflict of interest statement\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors have no relevant financial or non-financial interests to disclose.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNo grants or any other funding support were received for conducting the present study.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData availability statement\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe data used and analyzed in this study are available from the corresponding author upon request. \u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eHuman Ethics and Consent to participate\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eApproval for the study was obtained by the Research Ethics Committee at the University of Sharjah; approval number (………). Informed consent was obtained from all individual participants included in the study approving the use of personal data and medical records including photographs and digital scans for the purpose of medical research.\u0026nbsp;\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eAlbalawi F, Alamoud KA (2022) Trends and application of artificial intelligence technology in orthodontic diagnosis and treatment planning: a review. \u003cem\u003eAppl Sci\u003c/em\u003e 12(22):11864. https://doi.org/10.3390/app122211864\u003c/li\u003e\n\u003cli\u003eBardideh E, Lal Alizadeh F, Amiri M et al (2024) Designing an artificial intelligence system for dental occlusion classification using intraoral photographs: a comparative analysis between artificial intelligence‑based and clinical diagnoses. \u003cem\u003eAm J Orthod Dentofacial Orthop\u003c/em\u003e 166(2):125\u0026ndash;137. https://doi.org/10.1016/j.ajodo.2024.03.012\u003c/li\u003e\n\u003cli\u003eBerkovitz BKB, Holland GR, Moxham BJ (2009) Oral anatomy, histology and embryology. 4th ed. Edinburgh: Mosby. https://doi.org/10.1093/EJO/CJY010\u003c/li\u003e\n\u003cli\u003eChang AC (2020) Artificial intelligence in medicine: the future of personalized health care. In: Chang AC (ed) Intelligence-based medicine: data science, artificial intelligence, and human cognition in clinical medicine and healthcare, 1st edn. Academic Press, San Diego, pp 1\u0026ndash;12\u003c/li\u003e\n\u003cli\u003eChoi HI, Jung SK, Baek SH et al (2019) Artificial intelligent model with neural network machine learning for the diagnosis of orthognathic surgery. \u003cem\u003eJ Craniofac Surg\u003c/em\u003e 30(7):1986\u0026ndash;1989. https://doi.org/10.1097/SCS.0000000000005650\u003c/li\u003e\n\u003cli\u003eDaniels C, Richmond S (2000) The development of the Index of Complexity, Outcome and Need (ICON). \u003cem\u003eJ Orthod\u003c/em\u003e 27(2):149\u0026ndash;162. https://doi.org/10.1093/ortho/27.2.149\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003eFaber J, Faber C, Faber P (2019)\u003c/strong\u003e Artificial intelligence in orthodontics. \u003cem\u003eAPOS Trends Orthod\u003c/em\u003e 9(4):201\u0026ndash;205. https://doi.org/10.25259/APOS_123_2019\u003c/li\u003e\n\u003cli\u003eJung SK, Kim TW (2016) New approach for the diagnosis of extractions with neural network machine learning. \u003cem\u003eAm J Orthod Dentofacial Orthop\u003c/em\u003e 149(1):127\u0026ndash;133. https://doi.org/10.1016/j.ajodo.2015.07.030\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003eKhanagar SB, Al‑Ehaideb A, Vishwanathaiah S et al (2021)\u003c/strong\u003e Scope and performance of artificial intelligence technology in orthodontic diagnosis, treatment planning, and clinical decision‑making: a systematic review. \u003cem\u003eJ Dent Sci\u003c/em\u003e 16(1):482\u0026ndash;492. https://doi.org/10.1016/j.jds.2020.05.022\u003c/li\u003e\n\u003cli\u003eLi S, Guo Z, Lin J, Ying S (2022) Artificial intelligence for classifying and archiving orthodontic images. \u003cem\u003eBiomed Res Int\u003c/em\u003e 2022:1473977. https://doi.org/10.1155/2022/1473977 \u003c/li\u003e\n\u003cli\u003eOthman SA, Harradine NW (2007) Tooth size discrepancies in an orthodontic population. \u003cem\u003eAngle Orthod\u003c/em\u003e 77(4):668\u0026ndash;674. https://doi.org/10.2319/031406-102\u003c/li\u003e\n\u003cli\u003eRichmond S, Shaw WC, O\u0026rsquo;Brien KD et al (1992) The development of the PAR Index (Peer Assessment Rating): reliability and validity. \u003cem\u003eEur J Orthod\u003c/em\u003e 14(2):125\u0026ndash;139. https://doi.org/10.1093/ejo/14.2.125\u003c/li\u003e\n\u003cli\u003eRyu J, Kim YH, Kim TW et al (2023) Evaluation of artificial intelligence model for crowding categorization and extraction diagnosis using intraoral photographs. \u003cem\u003eSci Rep\u003c/em\u003e 13:5177. https://doi.org/10.1038/s41598-023-32514-7 \u003c/li\u003e\n\u003cli\u003eRyu J, Lee YS, Mo SP et al (2022) Application of deep learning artificial intelligence technique to the classification of clinical orthodontic photos. \u003cem\u003eBMC Oral Health\u003c/em\u003e 22:454. https://doi.org/10.1186/s12903-022-02466-x\u003c/li\u003e\n\u003cli\u003eShaw WC, Richmond S, O\u0026rsquo;Brien KD, Brook P, Stephens CD (1991) Quality control in orthodontics: indices of treatment need and treatment standards. \u003cem\u003eBr Dent J\u003c/em\u003e 170(3):107\u0026ndash;112. https://doi.org/10.1038/sj.bdj.4807429\u003c/li\u003e\n\u003cli\u003eTalaat S, Kaboudan A, Talaat W et al (2021) The validity of an artificial intelligence application for assessment of orthodontic treatment need from clinical images. \u003cem\u003eSemin Orthod\u003c/em\u003e 27(2):164\u0026ndash;171. https://doi.org/10.1053/j.sodo.2021.05.012\u003c/li\u003e\n\u003cli\u003eWallis C, McNamara C, Cunningham SJ et al (2014) How good are we at estimating crowding and how does it affect our treatment decisions? \u003cem\u003eEur J Orthod\u003c/em\u003e 36(4):465\u0026ndash;470. https://doi.org/10.1093/ejo/cjt080\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"journal-of-orofacial-orthopedics-fortschritte-der-kieferorthopadie","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"","sideBox":"Learn more about [Journal of Orofacial Orthopedics / Fortschritte der Kieferorthopädie](https://link.springer.com/journal/56)","snPcode":"56","submissionUrl":"https://submission.springernature.com/new-submission/56/3","title":"Journal of Orofacial Orthopedics / Fortschritte der Kieferorthopädie","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"Artificial Intelligence, Orthodontics, Index of Complexity Outcome and Need, Tooth Size Arch Length Discrepancy, OrthoCAD","lastPublishedDoi":"10.21203/rs.3.rs-6898322/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6898322/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003e\u003cb\u003ePurpose\u003c/b\u003e\u003c/p\u003e\u003cp\u003eThe aim of this study was to develop an artificial intelligence (AI) model capable of quantifying crowding and spacing in the upper arch and to validate its accuracy by comparing the model\u0026rsquo;s results with those of human observers.\u003c/p\u003e\u003cp\u003e\u003cb\u003eMaterials and Methods\u003c/b\u003e\u003c/p\u003e\u003cp\u003eThis study included upper intraoral photographs and occlusal scans of orthodontic patients treated at the University of Sharjah (2022\u0026ndash;2024). The YOLO (You Only Look Once) 8 Pose Model was generated using a training and validation dataset (832 images). The AI model performed tooth segmentation and tooth point detection on occlusal images, followed by automated quantification of tooth size arch length discrepancy (TSALD). Manual space analysis was conducted using OrthoCAD software and the data was compared with the results of the AI model using a testing dataset (300 images). TSALD was categorized based on the index of treatment complexity, outcome, and need (ICON). Qualitative data were presented as frequency and distribution, and comparisons were performed by using Fisher\u0026rsquo;s Exact test. Correlation between Manual and AI-measured TSALD was evaluated using Pearsons\u0026rsquo;s correlation coefficient.\u003c/p\u003e\u003cp\u003e\u003cb\u003eResults\u003c/b\u003e\u003c/p\u003e\u003cp\u003eThe model achieved an overall accuracy of 90%. The largest discrepancies were found in mild crowding (\u0026lt;\u0026thinsp;2 mm, 7%), severe spacing (5.1\u0026ndash;9 mm, 5%), and moderate spacing (2.1\u0026ndash;5 mm, 3.3%). A strong correlation (\u0026gt;\u0026thinsp;0.92) between manual and AI TSALD measurements indicated high reliability and potential interchangeability.\u003c/p\u003e\u003cp\u003e\u003cb\u003eConclusions\u003c/b\u003e\u003c/p\u003e\u003cp\u003eThe AI model was successfully developed and validated, achieving 90% accuracy, demonstrating its potential as a reliable tool for quantifying TSALD in orthodontic diagnostics.\u003c/p\u003e","manuscriptTitle":"Maxillary Crowding and Spacing: Validation of an Artificial Intelligence Model vs. Digitally-Assisted Human Observer","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-10-10 14:50:45","doi":"10.21203/rs.3.rs-6898322/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2025-10-16T08:33:35+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-09-30T08:48:23+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"52428062271738504818740750105031395496","date":"2025-09-26T20:48:27+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-09-26T13:44:03+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-06-21T01:09:10+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-06-21T01:07:54+00:00","index":"","fulltext":""},{"type":"submitted","content":"Journal of Orofacial Orthopedics / Fortschritte der Kieferorthopädie","date":"2025-06-15T12:33:07+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"journal-of-orofacial-orthopedics-fortschritte-der-kieferorthopadie","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"","sideBox":"Learn more about [Journal of Orofacial Orthopedics / Fortschritte der Kieferorthopädie](https://link.springer.com/journal/56)","snPcode":"56","submissionUrl":"https://submission.springernature.com/new-submission/56/3","title":"Journal of Orofacial Orthopedics / Fortschritte der Kieferorthopädie","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"a16efc48-14fd-4479-b1fb-eb25b9c1cc22","owner":[],"postedDate":"October 10th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2025-11-17T19:53:21+00:00","versionOfRecord":[],"versionCreatedAt":"2025-10-10 14:50:45","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-6898322","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6898322","identity":"rs-6898322","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}