Skip to main content An official website of the United States government Here's how you know Here's how you know Official websites use .gov A .gov website belongs to an official government organization in the United States. Secure .gov websites use HTTPS A lock ( Lock Locked padlock icon ) or https:// means you've safely connected to the .gov website. Share sensitive information only on official, secure websites. Search Log in Dashboard Publications Account settings Log out Search… Search NCBI Primary site navigation Search Logged in as: Dashboard Publications Account settings Log in Search PMC Full-Text Archive Search in PMC Journal List User Guide PERMALINK Copy As a library, NLM provides access to scientific literature. Inclusion in an NLM database does not imply endorsement of, or agreement with, the contents by NLM or the National Institutes of Health. Learn more: PMC Disclaimer | PMC Copyright Notice Biomed Eng Online . 2026 Mar 4;25:57. doi: 10.1186/s12938-026-01543-4 Search in PMC Search in PubMed View in NLM Catalog Add to search Machine learning algorithms for predicting quality of life improvements after digital orthodontic treatment: a retrospective analysis Yuzhe Huang Yuzhe Huang 1 Faculty of Dentistry, Lincoln University College, 47301 Petaling Jaya, Selangor Darul Ehsan Malaysia Find articles by Yuzhe Huang 1 , Rasheed Abdulsalam Rasheed Abdulsalam 1 Faculty of Dentistry, Lincoln University College, 47301 Petaling Jaya, Selangor Darul Ehsan Malaysia Find articles by Rasheed Abdulsalam 1, ✉ Author information Article notes Copyright and License information 1 Faculty of Dentistry, Lincoln University College, 47301 Petaling Jaya, Selangor Darul Ehsan Malaysia ✉ Corresponding author. Received 2025 Oct 13; Accepted 2026 Feb 10; Collection date 2026. © The Author(s) 2026 Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/ . PMC Copyright notice PMCID: PMC13069775 PMID: 41781986 Abstract Background Digital orthodontic treatment has revolutionized clinical practice, yet predicting individual patient outcomes remains challenging. This retrospective study developed and validated machine learning algorithms to predict quality of life improvements following digital orthodontic treatment. Methods Clinical data from 386 patients who underwent clear aligner therapy between January 2020 and December 2023 were analyzed. The dataset included demographic information, clinical parameters, imaging data, and standardized quality of life assessments using OHIP-14, IOTN, and VAS scales. Three machine learning algorithms—random forest, support vector machine, and neural network—were trained and evaluated using 70% training, 15% validation, and 15% test sets. Results Digital orthodontic treatment demonstrated 92.3 ± 5.8% tooth movement accuracy and reduced average treatment duration to 18.5 ± 4.2 months. Quality of life assessments revealed significant improvements, with OHIP-14 scores decreasing from 24.6 ± 8.2 to 8.2 ± 4.3 (66.7% reduction, P < 0.001), and VAS aesthetic satisfaction increasing from 28.4 ± 12.3 to 85.6 ± 8.7 ( P < 0.001). The random forest algorithm achieved superior predictive performance with 87.9% accuracy, 89.5% sensitivity, 85.7% specificity, and 0.93 AUC. Feature importance analysis identified baseline OHIP-14 scores (0.142), crowding severity (0.131), treatment duration (0.121), and patient compliance (0.111) as primary predictive factors. Clinical implementation of the prediction system improved treatment understanding (92.3% vs. 78.6%, P < 0.01) and patient satisfaction (94.8% vs. 83.2%, P < 0.01) compared to conventional consultations. Conclusion This study demonstrates that machine learning can accurately predict orthodontic treatment outcomes and enhance clinical decision-making. The developed predictive system provides a valuable tool for outcome prediction and identification of factors associated with treatment success, potentially transforming orthodontic practice by enabling data-driven, patient-specific care strategies that optimize treatment outcomes and patient satisfaction. Keywords: Machine learning, Digital orthodontics, Quality of life, Clear aligners, Predictive model Background The rapid development of digital technology is profoundly transforming clinical practice patterns in orthodontics. In recent years, the application of computer-aided design and manufacturing technology, intraoral scanning systems, and three-dimensional imaging technology has shifted orthodontic treatment from traditional analog methods to digital precision diagnosis and treatment [ 1 ]. Recent advances in artificial intelligence have further expanded digital capabilities in dentistry, enabling automated image analysis and decision support across various oral and maxillofacial applications [ 2 ]. This transformation not only improves the accuracy and predictability of treatment but also significantly enhances the patient experience. Three-dimensional printing technologies combined with AI-assisted design have refined digital workflows, enabling precise appliance fabrication with improved chairside efficiency [ 3 , 4 ]. In particular, clear aligner technology has emerged as a more aesthetic and comfortable treatment option that has been widely adopted globally [ 5 ] and forms the basis of the digital orthodontic workflow examined in this study. The ultimate goal of orthodontic treatment is not only to improve patients' occlusal relationships and facial aesthetics but, more importantly, to enhance their quality of life. Oral Health-Related Quality of Life (OHRQoL) has become an important indicator for evaluating orthodontic treatment outcomes [ 6 , 7 ]. Research shows that orthodontic treatment can improve patients' overall quality of life by enhancing oral function, aesthetics, and psychosocial status [ 8 , 9 ]. However, different orthodontic techniques have varying impacts on patients' quality of life [ 10 ]. Systematic reviews show that compared to traditional fixed appliances, clear aligners provide better comfort and higher quality of life during treatment [ 11 ]. These differences are mainly reflected in pain perception, dietary restrictions, oral hygiene maintenance, and social activities. Notably, a systematic review and meta-analysis revealed substantial heterogeneity in treatment outcomes across studies (SMD = 1.29, 95% CI 0.67–1.92), indicating that the magnitude of QoL improvement varies considerably among individuals [ 12 ]. Furthermore, patients with different malocclusion classifications demonstrate distinct patterns of QoL improvement throughout treatment stages, with Class II patients benefiting most during space closure while Class I patients show greater psychological improvement during alignment [ 13 ]. This variability underscores that treatment outcomes are not uniformly predictable, and baseline psychological status significantly influences patients' perception of improvement [ 14 ]. The emergence of artificial intelligence and machine learning technology has brought revolutionary changes to dentistry [ 15 , 16 ]. In the field of orthodontics, machine learning algorithms have been successfully applied to diagnostic tasks including cephalometric analysis, skeletal age prediction, and tooth segmentation, treatment planning, and many other aspects [ 17 , 18 ]. These technologies demonstrate accuracy comparable to or even higher than human experts while greatly improving work efficiency [ 19 , 20 ]. Recent benchmarking studies have systematically evaluated the performance of large language models in oral and maxillofacial applications, revealing that while reasoning-optimized models achieve clinically meaningful accuracy gains, domain-specific validation and human oversight remain essential [ 21 ]. The advantage of machine learning lies in its ability to process large amounts of complex multidimensional data, identify patterns and associations that are difficult for humans to detect, thereby providing scientific basis for clinical decisions [ 22 ]. Machine learning shows great potential, particularly in predicting treatment outcomes and developing personalized treatment plans [ 23 ]. Xu et al. [ 24 ] pioneered the use of artificial neural networks to predict patient experience during clear aligner treatment, achieving prediction accuracies of 87.7% for pain, 93.4% for anxiety, and 92.4% for quality of life. This study demonstrated the feasibility of AI-based prediction for orthodontic patient-centered outcomes, though it focused on short-term treatment experience rather than post-treatment QoL improvement. Although existing research has made important progress in digital orthodontic technology and machine learning applications, there are still obvious limitations. Most studies focus on technical applications, such as automated cephalometric landmark identification, skeletal classification, or extraction decision prediction [ 25 ], with insufficient attention to predicting the more clinically meaningful outcome of patient quality of life improvement. While orthodontic treatment generally improves OHRQoL, the degree of improvement is neither uniform nor easily predictable prior to treatment initiation. Current clinical practice relies heavily on clinician experience to estimate expected benefits, lacking quantitative tools to support individualized prognosis. This gap is particularly significant because patients with similar baseline malocclusion severity may experience markedly different levels of satisfaction and functional improvement, influenced by factors including psychological status, compliance, and treatment duration that interact in complex, nonlinear ways. Existing quality of life assessment studies are mostly descriptive or comparative, lacking predictive model construction. Moreover, although some studies have attempted to use machine learning to predict certain aspects of orthodontic treatment, few studies have organically integrated digital orthodontic technology, quality of life assessment, and machine learning prediction. This lack of integrative research limits the development of personalized precision medicine in orthodontics. The innovation of this study lies in applying machine learning algorithms to predict the degree of quality-of-life improvement after digital orthodontic treatment. Unlike the study by Xu et al. [ 24 ], which predicted patient experience during the initial treatment phase, this study aims to predict post-treatment QoL improvement, which represents the primary outcome of interest when making treatment decisions. By integrating patients' clinical parameters, imaging data, and treatment process information, multiple machine learning models are constructed to achieve individualized prediction of post-treatment quality of life improvement. This approach not only fills the gap in existing research but also provides quantitative tools for clinicians to develop personalized treatment plans. Such predictive capability enables orthodontists to identify patients likely to achieve significant improvement, set realistic expectations for those with more modest projected gains, recognize modifiable factors that could enhance outcomes, and support shared decision-making between clinicians and patients. Unlike previous studies that only focused on technical implementation or simply compared treatment effects, this study aims to establish a predictive system that can be used for clinical decision support. This study aims to evaluate the performance of different machine learning algorithms in predicting patient quality of life improvement through retrospective analysis of data from patients who received digital orthodontic treatment, identify key factors affecting quality of life improvement, and establish an accurate and reliable predictive model. The results will provide scientific basis for orthodontists to assess patients' expected benefits before treatment, help doctors and patients jointly develop the most suitable treatment plan, and promote the application and development of personalized precision medicine in orthodontics. Results Patient baseline characteristics This study included 386 patients who received digital orthodontic treatment, including 221 females (57.3%) and 165 males (42.7%). Patient age ranged from 12 to 40 years, with a mean age of 22.8 ± 7.3 years. There were 142 adolescent patients (12–18 years, 36.8%) and 244 adult patients (> 18 years, 63.2%). Demographic characteristics and clinical baseline data of all patients are shown in Table 1 . Table 1. Demographic characteristics and clinical baseline data of the study population ( n = 386) Characteristics Value/Number (%) Demographics Gender Female 221 (57.3) Male 165 (42.7) Age (years) 22.8 ± 7.3 Adolescents (12–18 years) 142 (36.8) Adults (> 18 years) 244 (63.2) Malocclusion classification Angle classification Class I 198 (51.3) Class II division 1 124 (32.1) Class II division 2 64 (16.6) Crowding severity Mild (< 4 mm) 89 (23.1) Moderate (4-8 mm) 186 (48.2) Severe (> 8 mm) 111 (28.7) Vertical relationship Normal overbite 234 (60.6) Deep overbite 108 (28.0) Open bite 44 (11.4) Pre-treatment quality of life scores OHIP-14 total score 24.6 ± 8.2 IOTN-DHC Grade 4 (treatment required) 267 (69.2) Grade 5 (great need) 119 (30.8) IOTN-AC Patient self-assessment 7.8 ± 1.6 Clinician assessment 7.2 ± 1.4 VAS aesthetic satisfaction 28.4 ± 12.3 Open in a new tab Data presented as mean ± standard deviation or number (percentage) OHIP-14: Oral Health Impact Profile; IOTN: Index of Orthodontic Treatment Need; DHC: Dental Health Component; AC: Aesthetic Component; VAS: Visual Analogue Scale Regarding malocclusion classification, Angle Class I malocclusion was predominant (198 cases, 51.3%), followed by Angle Class II Division 1 (124 cases, 32.1%) and Angle Class II Division 2 (64 cases, 16.6%). Analysis of dental crowding severity showed 89 patients (23.1%) with mild crowding (< 4 mm), 186 patients (48.2%) with moderate crowding (4–8 mm), and 111 patients (28.7%) with severe crowding (> 8 mm). In vertical relationship assessment, 234 patients (60.6%) had normal overbite, 108 patients (28.0%) had deep overbite, and 44 patients (11.4%) had open bite. Pre-treatment quality of life assessment results showed an average OHIP-14 total score of 24.6 ± 8.2 points, indicating moderate negative impact of malocclusion on patients' oral health-related quality of life. IOTN-DHC scores showed 267 patients (69.2%) at Grade 4 (treatment required) and 119 patients (30.8%) at Grade 5 (great need). IOTN-AC patient self-assessment averaged 7.8 ± 1.6, while clinician assessment averaged 7.2 ± 1.4, showing good consistency between the two (ICC = 0.82, P < 0.001). VAS scores showed low patient satisfaction with their dental aesthetics (28.4 ± 12.3 points), which was the primary motivation for seeking orthodontic treatment. Digital orthodontic treatment significantly improves patient quality of life The precision of digital orthodontic treatment was reflected in accurate control of tooth movement. Through three-dimensional digital analysis, the average tooth movement accuracy for 386 patients reached 92.3 ± 5.8%, meaning the conformity rate between actual and expected tooth positions exceeded 90%. Among these, anterior teeth showed the highest movement accuracy (94.5 ± 4.2%), while posterior teeth were slightly lower (89.7 ± 6.5%). Control accuracy for rotational movements (88.2 ± 7.3%) was lower than translational movements (93.8 ± 4.9%), consistent with previous research findings [ 26 ]. Notably, cases using attachments ( n = 298) had higher movement accuracy than those without attachments ( n = 88) (93.1% vs. 89.5%, P < 0.01). Treatment time analysis showed an average treatment duration of 18.5 ± 4.2 months, significantly shorter than the 24–30 months reported for traditional fixed appliances. The average number of visits was 15.3 ± 3.1, approximately 30% fewer than traditional treatment. The improvement in treatment efficiency was mainly attributed to: (1) digital design optimization of tooth movement paths; (2) staged movement reducing anchorage loss; (3) good patient compliance, with average daily wear time reaching 21.2 ± 1.8 h. Correlation analysis showed a significant negative correlation between patient compliance and treatment duration ( r = − 0.42, P < 0.001), suggesting that good patient cooperation can significantly shorten treatment time. Quality of life improvement was the core indicator for evaluating treatment effectiveness. As shown in Fig. 1 , OHIP-14 total score decreased from 24.6 ± 8.2 before treatment to 8.2 ± 4.3 after treatment, a reduction of 66.7% (P < 0.001). All dimensions showed significant improvement, with psychological discomfort showing the most improvement (from 6.0 ± 2.1 to 2.5 ± 1.2, 58.3% reduction), followed by psychological disability (from 5.5 ± 1.9 to 2.0 ± 1.0, 63.6% reduction) and functional limitation (from 5.0 ± 1.8 to 2.0 ± 0.9, 60.0% reduction). VAS scores showed patient satisfaction with dental aesthetics increased from 28.4 ± 12.3 before treatment to 85.6 ± 8.7 after treatment (P < 0.001). IOTN-AC scores decreased from 7.8 ± 1.6 before treatment to 2.3 ± 0.8 after treatment (P < 0.001), indicating aesthetic improvement achieved mutual recognition from both patients and clinicians. Fig. 1. Open in a new tab Changes in OHIP-14 dimension scores before and after treatment and overall quality of life improvement Machine learning models accurately predict post-treatment quality of life improvement Machine learning models demonstrated excellent performance in predicting patient quality of life improvement. Table 2 summarizes the predictive performance of three algorithms on the test set ( n = 58). Random forest algorithm performed best with 87.9% accuracy, 89.5% sensitivity, 85.7% specificity, and AUC value of 0.93. Support vector machine followed with 84.5% accuracy and AUC value of 0.90. Neural network performance was slightly lower but still reached clinically acceptable levels (accuracy 82.8%, AUC = 0.88). ANOVA revealed significant differences among the three algorithms ( P < 0.05), with post-hoc Tukey HSD tests confirming random forest superiority over both support vector machine and neural network to provide a more complete description of the statistical approach including the post-hoc analysis. Table 2. Comparison of predictive performance of different machine learning algorithms (test set n = 58) Algorithm Accuracy (%) Sensitivity (%) Specificity (%) AUC F1 Score MAE RMSE Random forest 87.9 ± 3.2 89.5 ± 4.1 85.7 ± 3.8 0.93 ± 0.02 0.88 ± 0.03 3.2 ± 0.8 4.1 ± 1.0 Support vector machine 84.5 ± 3.8 86.2 ± 4.5 82.1 ± 4.2 0.90 ± 0.03 0.85 ± 0.04 3.8 ± 0.9 4.7 ± 1.2 Neural network 82.8 ± 4.2 84.3 ± 5.0 80.6 ± 4.6 0.88 ± 0.03 0.83 ± 0.04 4.2 ± 1.0 5.2 ± 1.3 P value 0.023 0.041 0.038 0.015 0.028 0.019 0.022 Open in a new tab Data presented as mean ± standard deviation; P values are from ANOVA results comparing three groups; Accuracy, sensitivity, specificity used for classification prediction (significant improvement vs mild improvement); MAE and RMSE used for regression prediction of quality of life improvement scores AUC: Area Under the Receiver Operating Characteristic curve; MAE: Mean Absolute Error; RMSE: Root Mean Square Error Feature importance analysis revealed key factors affecting quality of life improvement (Fig. 2 ). This feature contribution analysis method was similar to the research method by Huang et al. for evaluating feature importance in orthodontic extraction decisions [ 27 ]. The top five important features identified by the random forest model were: pre-treatment OHIP-14 score (importance 0.142), crowding severity (0.131), treatment duration (0.121), patient compliance (0.111), and age (0.101). This indicates that baseline quality of life status and malocclusion severity are the main factors predicting post-treatment improvement. Notably, patient compliance as a behavioral factor exceeded many anatomical parameters in importance, emphasizing the key role of patient cooperation in treatment success [ 28 ]. Fig. 2. Open in a new tab ROC curve comparison of three machine learning algorithms and feature variable importance ranking. A ROC curves showing discriminative ability of three algorithms, with random forest (blue line) performing best (AUC = 0.93), followed by support vector machine (orange line, AUC = 0.90) and neural network (green line, AUC = 0.88). B Feature importance analysis from random forest model, showing pre-treatment OHIP-14 score, crowding severity, and treatment duration as the three most important factors predicting quality of life improvement Model stability was validated through fivefold cross-validation and Bootstrap resampling (1000 times). The random forest model showed minimal performance fluctuation across different data subsets (AUC standard deviation 0.02), demonstrating good stability and generalization ability. Learning curve analysis showed that model performance stabilized when training sample size reached 200 cases, suggesting that the current sample size of 386 cases was sufficient to support reliable model training. Additionally, through SHAP (SHapley Additive exPlanations) value analysis, we further validated the consistency of feature contributions, enhancing model interpretability. Predictive model optimizes personalized treatment decisions The prediction system developed based on machine learning models successfully achieved personalized treatment decision support. Figure 3 shows the clinical application interface. After clinicians input patient basic information, clinical parameters, and baseline quality of life scores, the system can generate prediction results within seconds. In 58 test cases, the correlation coefficient between system-predicted quality of life improvement and actual results reached 0.86 ( P < 0.001), demonstrating its clinical utility. Fig. 3. Open in a new tab Machine learning-based personalized quality of life improvement prediction visualization interface example. The interface shows the complete prediction process: left side is the patient information input area, including demographic data, clinical parameters, and baseline quality of life scores; right side is the prediction results display area, including overall improvement prediction (80%), specific improvement in each dimension, prediction confidence (92%), and personalized treatment recommendations. Based on a machine learning model from 386 similar cases, the system can generate personalized prediction reports within seconds, effectively supporting clinical decision-making and doctor-patient communication Regarding personalized treatment plan optimization, the system can identify key factors affecting treatment outcomes for specific patients. For example, for patients with severe crowding and high baseline OHIP-14 scores, the system recommends prioritizing functional limitation and psychological discomfort issues, and recommends using attachments to improve tooth movement accuracy. By simulating different treatment plans, the system can predict the potential impact of each plan on quality-of-life improvement, helping clinicians select optimal plans. In practical application, patients using system-recommended plans ( n = 30) showed higher quality of life improvement rates compared to the conventional treatment group ( n = 28) (82.3% vs. 71.4%, P = 0.032). Visualization of expected effects greatly improved doctor-patient communication. The system not only provides overall improvement predictions but also shows specific improvement degrees for each quality of life dimension, giving patients a more intuitive understanding of treatment effects. Through visualization elements such as bar charts and progress bars, patients can clearly understand the potential benefits of treatment. Questionnaire surveys showed that patients who used the prediction system for communication had significantly higher understanding and satisfaction with treatment plans compared to the control group (understanding: 92.3% vs. 78.6%, P < 0.01; satisfaction: 94.8% vs. 83.2%, P < 0.01) [ 29 ]. Improvement in doctor-patient communication was also reflected in joint participation in treatment decisions. The prediction confidence intervals and similar case references provided by the system enhanced patient trust in prediction results. In clinical practice, 89.7% of patients reported that the prediction system helped them better understand the necessity and expected effects of treatment, and 93.1% of patients believed that this visual prediction enhanced their confidence in accepting treatment. Additionally, the personalized recommendations generated by the system provided evidence-based medicine support for clinicians, making treatment decisions more scientific and standardized [ 30 ]. Discussion Through retrospective analysis of data from 386 patients who received digital orthodontic treatment, this study successfully constructed a machine learning-based quality of life improvement prediction model, achieving 87.9% prediction accuracy. The results not only validated the significant effect of digital orthodontic technology in improving patient quality of life but, more importantly, established a personalized prediction system that can be used for clinical decision support. The clinical value of such prediction lies in addressing the substantial heterogeneity of treatment outcomes: while orthodontic treatment generally improves QoL, prior meta-analyses have demonstrated that the magnitude of improvement varies considerably among individuals [ 12 ], and patients with different malocclusion classifications exhibit distinct improvement patterns [ 13 ]. This variability makes individualized outcome prediction clinically meaningful, enabling clinicians to set realistic expectations and identify modifiable factors that could enhance outcomes. This innovative attempt provides new insights and practical basis for the development of precision medicine in orthodontics. The 92.3% tooth movement accuracy demonstrated by digital orthodontic technology in this study is significantly superior to the 70–80% accuracy reported for traditional fixed appliances. This improvement in precision is mainly attributed to high-resolution data collection from intraoral scanning technology and precise path planning from computer-aided design [ 31 , 32 ]. The systematic review by Mohammed Alassiry indicated that digital intraoral scanning accuracy has reached ± 20 μm, far exceeding traditional impression techniques[ 31 ]. The difference between 94.5% movement accuracy in the anterior region and 89.7% in the posterior region in this study is consistent with findings by Rousseau and Retrouvey, who attributed this difference to more complex root morphology and greater occlusal forces in the posterior region [ 33 ]. Notably, this study found that using attachment assistance can improve movement accuracy by 3.6 percentage points, providing quantitative basis for rational attachment application in clinical practice. Patient comfort improvement was reflected in multiple aspects. The significant reduction in OHIP-14 scores in this study (from 24.6 to 8.2) is consistent with multiple research findings and comparable to improvements reported for conventional fixed appliances [ 8 , 34 ], suggesting that QoL improvement relates primarily to malocclusion correction rather than the specific appliance type. However, compared to traditional appliances, the comfort advantages of digital clear aligners are more pronounced during the treatment period [ 9 ], which may influence patient-reported outcomes and treatment adherence. The average treatment duration of 18.5 months in this study is approximately 30% shorter than the traditional fixed appliance treatment duration (24–30 months) reported by Mavreas and Athanasiou [ 35 ]. This efficiency improvement not only reduces patients' time costs but also decreases the psychological burden of long-term treatment. The reduction in visit frequency (average 15.3 visits) further confirms the advantages of digital technology in improving treatment efficiency. Whether the predictive model developed here applies to other orthodontic modalities remains to be tested; future comparative studies across different treatment types would clarify whether modality-specific models are necessary. The predictive ability demonstrated by machine learning algorithms in this study surpasses traditional statistical methods. Compared to the early study by Jung and Kim using neural networks to predict extraction decisions (84% accuracy) [ 36 ], this study achieved higher accuracy in predicting the more complex clinical outcome of quality of life improvement. This progress is attributed partly to algorithm improvements and partly reflects the importance of feature engineering. This study screened 32 key features from 57 initial features through systematic feature selection, effectively improving model performance and interpretability. The importance of multi-institutional validation was confirmed in the research by Etemad et al., which validated the effectiveness of machine learning in orthodontic decision-making through multi-center data [ 37 ]. While the study by Xie et al. also used neural networks, they only used 14 cephalometric parameters as input features [ 30 ]. In comparison, the multidimensional data integrated in this study more comprehensively reflects the complex factors affecting treatment outcomes. Machine learning's advantages in identifying nonlinear relationships were fully demonstrated in this study. Traditional linear regression analysis has difficulty capturing complex interactions between multiple factors such as patient age, crowding severity, and compliance. Yu et al. also found similar nonlinear patterns when using machine learning to evaluate facial attractiveness [ 38 ]. In this study, the random forest algorithm successfully captured these complex relationships by constructing multiple decision trees and identified that patient compliance as a predictive factor exceeded many anatomical parameters in importance, a finding with important guiding significance for clinical practice. The importance of quality of life assessment in orthodontic treatment is increasingly prominent. The multidimensional assessment system (OHIP-14, IOTN, VAS) used in this study provides a more comprehensive treatment outcome evaluation than single indicators. Compared to traditional evaluation methods that only focus on tooth alignment, this comprehensive assessment better aligns with modern medicine's holistic health perspective. While the study by Negri et al. also used the IOTN index, it was mainly for screening rather than prognostic assessment [ 17 ]. This study innovatively combined quality of life assessment with machine learning prediction, achieving a transformation from descriptive analysis to predictive analysis. Clinical application of the personalized prediction system significantly improved doctor-patient communication. The visualization interface developed in this study can intuitively display prediction results, helping patients understand potential treatment benefits. This improvement is similar to the machine learning-assisted diagnostic system proposed by Zhou et al. [ 18 ], but this study went further to achieve quantitative prediction of treatment outcomes. Patients using the prediction system showed higher treatment understanding (92.3% vs. 78.6%) and satisfaction (94.8% vs. 83.2%). The value here is not demonstrating that communication matters, which is already well-established, but rather providing individualized, quantitative expectations that transform clinical counseling from “most patients improve” to specific projected outcomes for each patient. This difference not only reflects the value of technological progress but, more importantly, promotes the transformation to a patient-centered medical model. Despite the positive results achieved in this study, there are still some limitations that need improvement in future research. Although the sample size reached 386 cases, it still mainly came from a single medical institution, which may limit the model's generalization ability. Subramanian et al. also emphasized the importance of multi-center validation when evaluating artificial intelligence applications in cephalometric analysis [ 39 ]. While the retrospective design of this study allowed rapid collection of large amounts of data, there are inherent limitations in data quality control and standardization. Although missing CBCT data for some patients (3.7%) was handled using multiple imputation, it may still affect model accuracy. Additionally, patient compliance was assessed through self-reported wear logs, which are prone to recall bias and potential overestimation compared to objective sensor measurements. While compliance emerged as an important predictive factor, its true contribution may be affected by measurement error, and future studies should incorporate objective compliance monitoring devices. The inclusion of treatment process variables (treatment duration, compliance) also warrants consideration. These variables are not available at baseline, meaning the current model functions as a tool for identifying factors associated with QoL improvement and for retrospective outcome analysis rather than purely pre-treatment prediction. Developing a baseline-only model for true pre-treatment prognosis represents an important future direction. There is still considerable room for algorithm optimization improvement. While the traditional machine learning algorithms used in this study performed well, the rapid development of deep learning technology in recent years provides possibilities for further improving predictive performance. The review by Surendran et al. pointed out that deep learning has unique advantages in processing complex medical image data [ 40 ]. Future research could consider applying convolutional neural networks to directly analyze intraoral scans and CBCT images, potentially discovering hidden patterns difficult to capture with traditional feature engineering methods. The relatively poor performance of neural networks in this study (82.8% accuracy) may be related to network structure design and training sample size, suggesting that deep learning methods in orthodontics still need further exploration. The choice of quality of life assessment tools also warrants discussion. Although OHIP-14 is a widely used standardized tool, its original design was not specifically for orthodontic patients. The most pronounced improvement in psychological-related dimensions in this study (58.3–63.6%) may reflect limitations of this scale in capturing specific impacts of orthodontic treatment. Future research should consider developing or adopting quality of life assessment tools more specific to orthodontic treatment to more accurately reflect multidimensional improvements brought by treatment. Another limitation of this study is the lack of long-term follow-up data. While assessment one month after treatment completion can reflect immediate effects, long-term stability of orthodontic treatment and sustained quality of life improvement are equally important. Volovic et al. also emphasized the value of longitudinal data in their study predicting orthodontic treatment duration [ 41 ]. Establishing predictive models that include long-term follow-up data would help provide patients with more comprehensive treatment expectations. Future research directions should include conducting multi-center prospective validation studies. By collecting standardized data from different regions and medical institutions, existing predictive models can be validated and optimized to improve their universality. Integrating more types of data sources, such as genomic data and microbiome data, may further improve prediction accuracy and personalization. The artificial intelligence-assisted retention management system proposed by Strunga et al.[ 42 ] provides new ideas for expanding applications of this study's model. Integrating predictive models with treatment process monitoring, retention management, and other functions to construct an intelligent decision support system covering the entire orthodontic treatment cycle will be an important future development direction. Establishing real-time data updates and model adaptive learning mechanisms is also worth exploring. As clinical data continuously accumulate, predictive models should have the ability for continuous learning and self-optimization. This dynamic update mechanism not only improves prediction accuracy but also timely captures changing trends in treatment technology and patient characteristics. Additionally, integrating the prediction system developed in this study with existing digital orthodontic software platforms to achieve seamless connection from diagnosis, treatment design to prognosis assessment will greatly improve clinical work efficiency and decision quality. This study conducted beneficial exploration in applying machine learning to predict quality of life in orthodontic treatment, but as Nordblom et al. pointed out in their critical review, artificial intelligence applications in orthodontics are still in early stages [ 25 ]. Technological progress must be closely integrated with clinical needs to ensure that developed tools truly serve the goals of improving patient prognosis and enhancing medical quality. As data accumulation increases and algorithms continuously optimize, the application prospects of personalized precision medicine in orthodontics will become increasingly broad. Conclusion This study successfully developed and validated a machine learning-based predictive system for assessing quality of life improvements following digital orthodontic treatment. Through comprehensive analysis of 386 patients' clinical data, the research demonstrated that digital orthodontic treatment achieved remarkable precision with 92.3 ± 5.8% tooth movement accuracy and significantly improved patients' oral health-related quality of life, as evidenced by a 66.7% reduction in OHIP-14 scores (from 24.6 ± 8.2 to 8.2 ± 4.3, P < 0.001). The random forest algorithm emerged as the optimal predictive model, achieving 87.9% accuracy, 89.5% sensitivity, and an AUC of 0.93 in predicting treatment outcomes. Feature importance analysis identified pre-treatment OHIP-14 scores, crowding severity, treatment duration, and patient compliance as key determinants of treatment success. The clinical implementation of this predictive system enhanced doctor-patient communication, with 92.3% of patients reporting improved understanding of treatment expectations compared to 78.6% in conventional consultations ( P < 0.01). This research establishes a novel paradigm for integrating artificial intelligence into orthodontic practice, providing clinicians with evidence-based tools for outcome prediction and identification of factors associated with treatment success. The findings contribute significantly to advancing precision medicine in orthodontics by demonstrating how machine learning algorithms can transform complex multidimensional clinical data into actionable insights that optimize treatment decisions and improve patient satisfaction. Future developments should focus on multi-center validation and incorporation of additional data modalities to further enhance the model's predictive capabilities and generalizability across diverse patient populations. Materials and methods Study design and patient inclusion This study employed a retrospective cohort design, collecting and analyzing data from patients who received digital orthodontic treatment in the Department of Orthodontics at Lincoln University College from January 2020 to December 2023. Initial screening identified 526 patients treated with clear aligners. After applying strict inclusion and exclusion criteria, 386 patients were ultimately included in the study analysis. The study design referenced relevant research methods on machine learning applications in orthodontics [ 41 , 43 ], including recent multi-institutional studies on orthodontic extraction prediction [ 27 , 37 ]. By systematically reviewing patients' complete treatment records, clinical data, imaging materials, and quality of life assessment results before and after treatment were extracted to construct a dataset for machine learning model training. The inclusion criteria for the study population included: (1) patients aged 12–40 years; (2) diagnosis and treatment design using digital intraoral scanning technology; (3) completion of full orthodontic treatment using clear aligners; (4) complete clinical records and imaging data before and after treatment; (5) completion of standardized quality of life assessment questionnaires before and after treatment. Exclusion criteria included: (1) patients with cleft lip and palate or other craniofacial developmental abnormalities; (2) patients who underwent orthognathic surgery during orthodontic treatment; (3) patients with previous orthodontic treatment history; (4) patients with systemic diseases affecting oral health; (5) patients with treatment interruption or loss to follow-up; (6) patients with incomplete clinical data. Based on the above criteria, researchers conducted preliminary screening through the hospital's electronic medical record system, followed by independent review of patient data by two orthodontic specialists with more than 5 years of clinical experience to ensure that included patients met all selection criteria. For cases with disagreements, a third senior orthodontist made the final determination. Digital orthodontic treatment workflow In this study, “digital orthodontics” refers specifically to clear aligner therapy integrating intraoral scanning, computer-aided design/computer-aided manufacturing (CAD/CAM), and staged virtual setup, excluding conventional fixed appliances with direct bonding [ 3 ]. All patients in this study underwent a standardized digital orthodontic treatment workflow that integrated advanced intraoral scanning technology, computer-aided design, and 3D printing technology. The entire treatment process achieved full digitalization from data collection to aligner fabrication, significantly improving treatment accuracy and predictability (Fig. 4 ). Fig. 4. Open in a new tab Digital orthodontic treatment workflow: Complete process from intraoral scanning to treatment completion Intraoral scanning is the starting point of digital orthodontic treatment. The study used iTero Element 2 intraoral scanners (Align Technology, San Jose, CA, USA) with a reported accuracy of ± 20 μm [ 44 ] to comprehensively scan patients' dentition and soft tissues. The scanning process followed a standardized protocol: patients were seated upright, starting with the maxillary arch, scanning from the right second molar along the dental arch to the left second molar, ensuring complete capture of occlusal, buccal, and lingual surface data. The mandibular arch was then scanned in the same manner, followed by buccal occlusion scanning in centric occlusion. Scanning data was transmitted to the computer in real-time, generating high-resolution three-dimensional digital models. The software automatically performed point cloud registration and mesh optimization, generating Standard Triangulation Language (STL) format files, providing an accurate digital foundation for subsequent treatment design. Clear aligner design and fabrication were conducted using ClinCheck Pro software (Align Technology) within a fully digital environment. Orthodontists used professional software to analyze 3D models, including tooth segmentation, coordinate system establishment, and arch form analysis. Based on diagnostic results and treatment goals, personalized tooth movement plans were developed. The software decomposed overall tooth movement into a series of small steps based on biomechanical principles, with each step controlled within 0.25–0.33 mm. The system automatically generated tooth positions for each stage and designed corresponding aligner shapes. Optimized attachments were prescribed according to planned tooth movements: horizontal rectangular attachments (3–5 mm) for rotation control, vertical rectangular attachments for extrusion, and beveled attachments for root movement. After design completion, aligners (Invisalign, Align Technology) were fabricated using SmartTrack material through thermoforming technology. Each aligner was worn for 14 days, achieving progressive tooth movement through continuous light force application. Monitoring and adjustment during treatment are key to ensuring treatment effectiveness. Patients returned for follow-up every 4–6 weeks for treatment progress evaluation. During follow-ups, doctors evaluated tooth movement accuracy, aligner fit, and patient compliance through intraoral examination and necessary imaging. The study employed a digital monitoring system, quantifying treatment progress by comparing actual tooth positions with expected positions in three-dimensional deviation. Refinement was initiated when the deviation between actual and planned tooth positions exceeded 0.5 mm, or when clinically significant rotational discrepancies (> 2°) were observed. Adjustment strategies included: (1) for mild deviations, correction through adding auxiliary attachments or adjusting aligner wear time; (2) for moderate deviations, rescanning and designing supplementary aligners; (3) for severe deviations or unexpected tooth movements, reformulating the treatment plan. Recent studies have demonstrated that digital indirect bonding workflows achieve clinically acceptable transfer accuracy across various tray configurations [ 45 ]. The entire treatment process formed a closed-loop management system, ensuring achievement of predetermined treatment goals through continuous monitoring and timely adjustments. Quality of life assessment tools This study adopted a multidimensional assessment system to comprehensively evaluate patients' oral health-related quality of life. The selected assessment tools were all internationally recognized standardized scales with good reliability and validity [ 46 ]. All assessments were conducted before treatment (T0) and one month after treatment completion (T1) by the same trained researcher to ensure assessment consistency. The Oral Health Impact Profile-14 (OHIP-14) is one of the most commonly used tools for assessing oral health-related quality of life [ 47 ]. This scale is a simplified version of OHIP-49, containing 14 items covering 7 dimensions: functional limitation, physical pain, psychological discomfort, physical disability, psychological disability, social disability, and handicap. Each item uses a Likert 5-point scale (0 = never, 1 = hardly ever, 2 = occasionally, 3 = fairly often, 4 = very often), with total scores ranging from 0 to 56 points. Higher scores indicate greater negative impact of oral health on quality of life. This study used the validated Chinese version of OHIP-14, with a Cronbach's α coefficient of 0.89, demonstrating good internal consistency. Although OHIP-14 uses ordinal items, the summed total score was treated as a continuous variable for modeling, consistent with established practice in OHRQoL research where summative scores approximate interval-level measurement [ 12 ]. The Index of Orthodontic Treatment Need (IOTN) is used to objectively assess patients' orthodontic treatment needs [ 48 , 49 ]. This index was developed by Brook and Shaw [ 50 ] and includes two components: Dental Health Component (DHC) and Aesthetic Component (AC). DHC uses a 5-grade classification system based on the impact of malocclusion on oral health: Grade 1 = no need for treatment, Grade 2 = little need, Grade 3 = borderline need, Grade 4 = treatment required, Grade 5 = great need. AC uses 10 standardized photographs to assess treatment need from a dental aesthetic perspective, with scores 1–4 indicating no need or little need for treatment, 5–7 indicating borderline need, and 8–10 indicating definite need for treatment. In this study, DHC was assessed by orthodontic specialists, while AC was assessed by both patient self-evaluation and clinician evaluation. Visual Analogue Scale (VAS) was used to assess patients' subjective satisfaction with treatment outcomes [ 46 ]. A 100 mm horizontal line was used, with the left end marked “extremely dissatisfied” (0 points) and the right end marked “extremely satisfied” (100 points). Patients marked on the line according to their feelings, and the distance from the mark to the left end was measured as the score. VAS assessment covered four aspects: (1) dental aesthetic satisfaction; (2) occlusal function satisfaction; (3) treatment comfort; (4) overall treatment satisfaction. Each aspect was scored independently, facilitating identification of specific factors affecting patient satisfaction (Table 3 ). Table 3. Comparison of characteristics of three quality of life assessment tools Assessment tool Evaluation dimensions Number of items Score range Assessment method Clinical significance OHIP-14 Functional limitation, physical pain, psychological discomfort, physical disability, psychological disability, social disability, handicap 14 0–56 points Patient self-assessment Higher scores indicate poorer quality of life IOTN-DHC Dental health impact 1 Grades 1–5 Clinician assessment Grades 4–5 require treatment IOTN-AC Aesthetic impact 1 1–10 points Patient self-assessment + clinician assessment 8–10 points indicate definite need for treatment VAS Aesthetics, function, comfort, overall satisfaction 4 0–100 points Patient self-assessment Higher scores indicate greater satisfaction Open in a new tab OHIP-14: Oral Health Impact Profile; IOTN: Index of Orthodontic Treatment Need; DHC: Dental Health Component; AC: Aesthetic Component; VAS: Visual Analogue Scale To ensure assessment accuracy and reliability, all patients completed questionnaires in a quiet, independent environment. Researchers provided standardized instructions but did not guide or intervene in patients' choices. For items that were difficult to understand, researchers only provided literal explanations without any suggestive explanations. All assessment data were immediately entered into an electronic database and double-checked to ensure data accuracy. Quality of life improvement was quantified as the absolute reduction in OHIP-14 total score from baseline (T0) to post-treatment (T1): Δ OHIP - 14 = OHIP - 14 T 0 - OHIP - 14 T 1 1 Higher positive values indicate greater improvement. For the classification task, patients were categorized into “significant improvement” versus “mild improvement” groups based on a threshold of ≥ 10 points reduction, corresponding to the minimal clinically important difference (MCID) established in orthodontic populations [ 12 , 13 ]. Data collection and preprocessing This study established a systematic data collection process to ensure high-quality, standardized data for machine learning model construction (Fig. 5 ). Data collection covered three aspects: clinical parameters, imaging data, and treatment process information. All data were extracted from the electronic medical record system and digital orthodontic management platform. Fig. 5. Open in a new tab Data collection and preprocessing workflow Clinical parameter collection included three main aspects. Baseline parameters available at treatment initiation (T0) included demographic information (age, sex), malocclusion classification (Angle class, crowding severity, vertical relationship), and pre-treatment quality of life scores. Tooth movement was precisely measured by comparing digital models before and after treatment. Using professional 3D analysis software, the movement distance and rotation angle of each tooth in three dimensions (mesio-distal, bucco-lingual, vertical) were calculated in a unified coordinate system. Measurement accuracy reached 0.1 mm and 1°. Treatment process parameters recorded during treatment included treatment duration, visit frequency, and patient compliance. Compliance was assessed through patient-maintained daily wear logs recording aligner wear time. We acknowledge that self-reported compliance data are subject to recall bias and potential overestimation; this methodological limitation is further addressed in the Discussion. Imaging data extraction involved the comprehensive application of multiple imaging techniques. Three-dimensional craniofacial measurement data were extracted from cone-beam computed tomography (CBCT) images before and after treatment, including sagittal, vertical, and transverse relationship parameters of the maxilla and mandible. Cephalometric analysis used automated software to identify anatomical landmarks and calculate angular measurements such as SNA, SNB, ANB, as well as linear measurements such as Wits appraisal and facial angle. In addition to aligner fabrication, intraoral scanning data were used to quantitatively analyze morphological parameters such as arch width, length, crowding, and Bolton index. All imaging measurements were completed by the same experienced orthodontist, with 20% of samples randomly selected for repeated measurements to calculate the Intraclass Correlation Coefficient (ICC) to assess measurement reliability. Data standardization processing was key to ensuring comparability of data from different sources and dimensions (Fig. 5 ). This study processed complete data from 386 patients, including 21,924 clinical measurements and 7,720 quality of life score data points. The data preprocessing workflow included: (1) Data quality control: checking data completeness, identifying and handling outliers. Among 386 patients, 23 outliers (0.1%) were found and verified as actual measurements; (2) Missing data handling: overall missing rate was 3.7% (mainly some CBCT measurements), handled using multiple imputation method (MICE); (3) Feature engineering: creating derived variables, ultimately forming 57 feature variables; (4) Data standardization: Z-score standardization for continuous variables. The final dataset was randomly allocated in a 7:1.5:1.5 ratio: training set 270 cases (70%), validation set 58 cases (15%), and test set 58 cases (15%). All preprocessing steps including missing data imputation, feature selection, and dimensionality reduction were performed exclusively on the training set to prevent data leakage. The validation and test sets were transformed using parameters derived solely from the training data. Machine learning model construction Machine learning model construction followed standard data science workflows, ensuring model reliability and generalization ability through systematic methods [ 41 , 43 ]. The entire modeling process included four key steps: feature engineering, algorithm selection, model training, and performance evaluation (Fig. 6 ). Fig. 6. Open in a new tab Machine learning predictive model construction and validation workflow Feature selection and dimensionality reduction were important steps in improving model performance and interpretability. The initial dataset contained 57 feature variables covering demographic information, clinical measurement parameters, imaging indicators, and treatment process data. The model incorporated both baseline variables (demographics, malocclusion characteristics, pre-treatment OHIP-14 scores, cephalometric measurements) and treatment process variables (treatment duration, compliance, visit frequency). A combined strategy of multiple feature selection methods was adopted: (1) Univariate feature selection: using ANOVA F -test to evaluate the correlation between each feature and the target variable (degree of quality of life improvement), retaining features with P < 0.05; (2) Recursive Feature Elimination (RFE): combined with random forest algorithm, iteratively removing features with lowest importance until reaching the optimal feature subset; (3) L1 regularization-based feature selection: utilizing the sparsity property of Lasso regression to automatically screen important features. All feature selection procedures were conducted using only the training set through nested cross-validation. Through these methods, 32 key features were ultimately determined for model construction. To further improve computational efficiency, Principal Component Analysis (PCA) was used for dimensionality reduction, retaining principal components with cumulative variance contribution of 95%, reducing feature dimensions to 24. PCA transformation matrices were fitted on the training set and subsequently applied to validation and test sets. Algorithm selection was based on the characteristics of different machine learning methods and orthodontic data properties. This study selected three representative algorithms: Random Forest (RF), Support Vector Machine (SVM), and Multilayer Perceptron neural network (MLP) [ 43 ]. Random forest is an ensemble learning method that constructs multiple decision trees and combines their predictions, with good anti-overfitting ability and capacity to capture nonlinear relationships. Support vector machines achieve classification or regression by finding optimal hyperplanes, particularly suitable for processing high-dimensional data. Neural networks can learn complex feature representations through multiple layers of nonlinear transformations, performing excellently in medical prediction tasks [ 41 ]. The reasons for selecting these three algorithms were: (1) they represent three different machine learning paradigms: ensemble learning, kernel methods, and deep learning; (2) they have shown good performance in previous orthodontic research; (3) they can capture patterns in data from different perspectives, facilitating comparison and selection of optimal models. Gradient boosting methods (e.g., XGBoost) were not included in the primary analysis because random forest provides more stable feature importance estimates on smaller clinical datasets and our primary objective was model interpretability for clinical decision support rather than marginal performance gains [ 37 ]. Model training and validation adopted strict methodological standards to ensure result reliability and reproducibility (Fig. 6 ). For each algorithm, grid search combined with fivefold cross-validation was used for hyperparameter optimization. Random forest optimization parameters included: number of decision trees (n_estimators: 100, 200, 500), maximum depth (max_depth: 10, 20, None), minimum samples split (min_samples_split: 2, 5, 10). Support vector machine optimization parameters included: regularization parameter (C: 0.1, 1, 10, 100), kernel function type (kernel: 'rbf', 'linear', 'poly'), kernel coefficient (gamma: 'scale', 'auto', 0.001, 0.01). The neural network adopted a three-hidden-layer structure (64, 32, 16 neurons), using ReLU activation function and Adam optimizer, with learning rate set to 0.001 and early stopping strategy (patience = 10) to prevent overfitting. All model training was conducted in the same computing environment, implemented using Python 3.8 and scikit-learn 1.0 library. Performance evaluation adopted a multidimensional indicator system to comprehensively measure model predictive ability. For classification prediction of quality of life improvement degree (significant improvement vs. mild improvement, defined as ΔOHIP-14 ≥ 10 points per Eq. 1 ), main evaluation indicators included: (1) Accuracy: proportion of correctly predicted samples to total samples; (2) Sensitivity: ability to correctly identify patients with significant improvement; (3) Specificity: ability to correctly identify patients with mild improvement; (4) Area Under the Receiver Operating Characteristic curve (AUC-ROC): comprehensive evaluation of model discriminative ability; (5) F1 score: harmonic mean of precision and recall. For regression prediction of quality of life improvement scores, evaluation indicators included Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and coefficient of determination ( R 2 ). Additionally, feature importance analysis was conducted to identify clinical factors with the greatest impact on prediction results. All performance indicators were calculated on the independent test set ( n = 58), and paired t -tests were used to compare performance differences between different algorithms, with P < 0.05 considered statistically significant. Acknowledgements Not applicable. Author contributions YH and RA contributed to the design of the study and data collection, performed the data analysis and wrote the manuscript. All authors read and approve the manuscript version final. Funding This research received no specific funding. Data availability The datasets analyzed in the current study are available from the corresponding author upon reasonable request. Declarations Ethics approval and consent to participate The study was approved by the local ethics committee of the Faculty of Dentistry, Lincoln University College, all experiments were performed in accordance with relevant guidelines and regulations such as the Declaration of Helsinki and the patients signed the informed consent form and agreed to be published. Consent for publication Not applicable. Competing interests The authors declare no competing interests. Footnotes Publisher's Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. References 1. Cunha T, Barbosa IDS, Palma KK. Orthodontic digital workflow: devices and clinical applications. Dent Press J Orthod. 2021;26(6):e21spe6. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 2. Nguyen VA, Vuong TQT, Nguyen VH. Benchmarking large-language-model vision capabilities in oral and maxillofacial anatomy: a cross-sectional study. PLoS ONE. 2025;20(10):e0335775. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 3. Nguyen VA, Vuong TQT. Transfer accuracy and chairside efficiency of rigid- and flexible-printed versus double vacuum-formed lingual indirect bonding trays: a prospective cohort study. Sci Rep. 2025;15(1):29320. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 4. Hoang V, et al. AI-assisted design of 3D-printed lingual indirect bonding trays: a comparative evaluation of bracket transfer accuracy. J Clin Med. 2025;14:4303. 10.3390/jcm14124303. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 5. Wang Y, et al. Expert consensus on the clinical strategies for orthodontic treatment with clear aligners. Int J Oral Sci. 2025;17(1):19. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 6. Zhang B, et al. Effect of clear aligners on oral health-related quality of life: a systematic review. Orthod Craniofac Res. 2020;23(4):363–70. [ DOI ] [ PubMed ] [ Google Scholar ] 7. Alfawal AMH, et al. The impact of non-extraction orthodontic treatment on oral health-related quality of life: clear aligners versus fixed appliances-a randomized controlled trial. Eur J Orthod. 2022;44(6):595–602. [ DOI ] [ PubMed ] [ Google Scholar ] 8. Jaber ST, et al. The effect of treatment with clear aligners versus fixed appliances on oral health-related quality of life in patients with severe crowding: a one-year follow-up randomized controlled clinical trial. Cureus. 2022;14(5):e25472. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 9. Kaklamanos EG, Makrygiannakis MA, Athanasiou AE. Oral health-related quality of life throughout treatment with clear aligners in comparison to conventional metal fixed orthodontic appliances: a systematic review. Int J Environ Res Public Health. 2023. 10.3390/ijerph20043537. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 10. Alhafi ZM, et al. Effectiveness and efficiency of indirect bonding techniques: an umbrella review with meta-analysis of the pooled findings. Int Orthod. 2025;23(4):101036. [ DOI ] [ PubMed ] [ Google Scholar ] 11. Baneshi M, et al. Effectiveness of clear orthodontic aligners in correcting malocclusions: a systematic review and meta-analysis. J Evid Based Dent Pract. 2025;25(1):102081. [ DOI ] [ PubMed ] [ Google Scholar ] 12. Andiappan M, et al. Malocclusion, orthodontic treatment, and the oral health impact profile (OHIP-14): systematic review and meta-analysis. Angle Orthod. 2015;85(3):493–500. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 13. Zheng D-H, et al. Assessing changes in quality of life using the Oral Health Impact Profile (OHIP) in patients with different classifications of malocclusion during comprehensive orthodontic treatment. BMC Oral Health. 2015;15(1):148. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 14. Navabi N, et al. Orthodontic treatment and the oral health-related quality of life of patients. J Dent. 2012;9(3):247–54. [ PMC free article ] [ PubMed ] [ Google Scholar ] 15. Bichu YM, et al. Applications of artificial intelligence and machine learning in orthodontics: a scoping review. Prog Orthod. 2021;22(1):18. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 16. Liu J, Zhang C, Shan Z. Application of artificial intelligence in orthodontics: current state and future perspectives. Healthcare. 2023. 10.3390/healthcare11202760. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 17. Mohammad-Rahimi H, et al. Machine learning and orthodontics, current trends and the future opportunities: a scoping review. Am J Orthod Dentofacial Orthop. 2021;160(2):170-192.e4. [ DOI ] [ PubMed ] [ Google Scholar ] 18. Zhou Y, et al. Orthodontic craniofacial pattern diagnosis: cephalometric geometry and machine learning. Med Biol Eng Comput. 2023;61(12):3345–61. [ DOI ] [ PubMed ] [ Google Scholar ] 19. Khanagar SB, et al. Scope and performance of artificial intelligence technology in orthodontic diagnosis, treatment planning, and clinical decision-making—a systematic review. J Dent Sci. 2021;16(1):482–92. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 20. Liu J, et al. Machine learning in orthodontics: challenges and perspectives. Adv Clin Exp Med. 2021;30(10):1065–74. [ DOI ] [ PubMed ] [ Google Scholar ] 21. Nguyen VA, et al. Quantifying the speed-accuracy trade-off of large language models on oral and maxillofacial surgery multiple-choice questions. Sci Rep. 2025;15(1):40657. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 22. Kazimierczak N, et al. AI in orthodontics: revolutionizing diagnostics and treatment planning-a comprehensive review. J Clin Med. 2024. 10.3390/jcm13020344. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 23. Fawaz P, El Sayegh P, VandeVannet B. Artificial intelligence in revolutionizing orthodontic practice. World J Methodol. 2025. 10.5662/wjm.v15.i3.100598. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 24. Xu L, et al. Predicting patient experience of Invisalign treatment: an analysis using artificial neural network. Korean J Orthod. 2022;52(4):268–77. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 25. Nordblom NF, Büttner M, Schwendicke F. Artificial Intelligence in orthodontics: critical review. J Dent Res. 2024;103(6):577–84. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 26. Chen S, et al. Machine learning in orthodontics: introducing a 3D auto-segmentation and auto-landmark finder of CBCT images to assess maxillary constriction in unilateral impacted canine patients. Angle Orthod. 2020;90(1):77–84. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 27. Huang J, et al. Evaluation of four machine learning methods in predicting orthodontic extraction decision from clinical examination data and analysis of feature contribution. Front Bioeng Biotechnol. 2024;12:1483230. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 28. Mason T, et al. A machine learning model for orthodontic extraction/non-extraction decision in a racially and ethnically diverse patient population. Int Orthod. 2023;21(3):100759. [ DOI ] [ PubMed ] [ Google Scholar ] 29. Prasad J, et al. Machine learning predictive model as clinical decision support system in orthodontic treatment planning. Dent J. 2022. 10.3390/dj11010001. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 30. Xie X, Wang L, Wang A. Artificial neural network modeling for deciding if extractions are necessary prior to orthodontic treatment. Angle Orthod. 2010;80(2):262–6. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 31. Mohammed Alassiry A. Clinical aspects of digital three-dimensional intraoral scanning in orthodontics - a systematic review. Saudi Dent J. 2023;35(5):437–42. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 32. Negri P, et al. Index of orthodontic treatment need (IOTN) and distribution of malocclusion traits in a population of growing patients attending a public orthodontic service in Perugia (Italy). Eur J Paediatr Dent. 2021;22(4):303–8. [ DOI ] [ PubMed ] [ Google Scholar ] 33. Rousseau M, Retrouvey JM. Machine learning in orthodontics: automated facial analysis of vertical dimension for increased precision and efficiency. Am J Orthod Dentofacial Orthop. 2022;161(3):445–50. [ DOI ] [ PubMed ] [ Google Scholar ] 34. Alhafi ZM, et al. The impact of non-extraction orthodontic treatment on the oral-health-related quality of life between a modified aligner appliance with Ni-Ti springs and the traditional fixed appliances: a randomized controlled clinical trial. Medicina. 2024;60:1139. 10.3390/medicina60071139. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 35. Mavreas D, Athanasiou AE. Factors affecting the duration of orthodontic treatment: a systematic review. Eur J Orthod. 2008;30(4):386–95. [ DOI ] [ PubMed ] [ Google Scholar ] 36. Jung SK, Kim TW. New approach for the diagnosis of extractions with neural network machine learning. Am J Orthod Dentofacial Orthop. 2016;149(1):127–33. [ DOI ] [ PubMed ] [ Google Scholar ] 37. Etemad LE, et al. Effectiveness of machine learning in predicting orthodontic tooth extractions: a multi-institutional study. Bioengineering. 2024. 10.3390/bioengineering11090888. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 38. Yu X, et al. Evaluation of facial attractiveness for patients with malocclusion: a machine-learning technique employing Procrustes. Angle Orthod. 2014;84(3):410–6. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 39. Subramanian AK, et al. Cephalometric analysis in orthodontics using artificial intelligence-a comprehensive review. Biomed Res Int. 2022;2022:1880113. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 40. Surendran A, et al. The future of orthodontics: deep learning technologies. Cureus. 2024;16(6):e62045. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 41. Volovic J, et al. A novel machine learning model for predicting orthodontic treatment duration. Diagnostics. 2023. 10.3390/diagnostics13172740. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 42. Strunga M, et al. Artificial intelligence systems assisting in the assessment of the course and retention of orthodontic treatment. Healthcare. 2023. 10.3390/healthcare11050683. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 43. Etemad L, et al. Machine learning from clinical data sets of a contemporary decision for orthodontic tooth extraction. Orthod Craniofac Res. 2021;24(Suppl 2):193–200. [ DOI ] [ PubMed ] [ Google Scholar ] 44. Ender A, Zimmermann M, Mehl A. Accuracy of complete- and partial-arch impressions of actual intraoral scanning systems in vitro. Int J Comput Dent. 2019;22(1):11–9. [ PubMed ] [ Google Scholar ] 45. Nguyen VA. Comparative analysis of lingual bracket transfer accuracy using fully versus partially enclosed 3D-printed indirect bonding trays: an in vivo study. PeerJ. 2025;13:e19612. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 46. Jawad Z, Bates C, Hodge T. Can dental registrants use the Index of Orthodontic Treatment Need accurately? Part 1: knowledge of IOTN among dental registrants. Br Dent J. 2016;220(10):527–32. [ DOI ] [ PubMed ] [ Google Scholar ] 47. Slade GD. Derivation and validation of a short-form oral health impact profile. Community Dent Oral Epidemiol. 1997;25(4):284–90. [ DOI ] [ PubMed ] [ Google Scholar ] 48. de Oliveira CM. The planning, contracting and monitoring of orthodontic services, and the use of the IOTN index: a survey of consultants in dental public health in the United Kingdom. Br Dent J. 2003;195(12):704–6 ( discussion 696 ). [ DOI ] [ PubMed ] [ Google Scholar ] 49. Ireland AJ, et al. An index of orthognathic functional treatment need (IOFTN). J Orthod. 2014;41(2):77–83. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 50. Brook PH, Shaw WC. The development of an index of orthodontic treatment priority. Eur J Orthod. 1989;11(3):309–20. [ DOI ] [ PubMed ] [ Google Scholar ] Associated Data This section collects any data citations, data availability statements, or supplementary materials included in this article. Data Availability Statement The datasets analyzed in the current study are available from the corresponding author upon reasonable request. Articles from BioMedical Engineering OnLine are provided here courtesy of BMC ACTIONS View on publisher site PDF (2.5 MB) Cite Collections Permalink PERMALINK Copy RESOURCES Similar articles Cited by other articles Links to NCBI Databases Cite Copy Download .nbib .nbib Format: AMA APA MLA NLM Add to Collections Create a new collection Add to an existing collection Name your collection * Choose a collection Unable to load your collection due to an error Please try again Add Cancel Follow NCBI NCBI on X (formerly known as Twitter) NCBI on Facebook NCBI on LinkedIn NCBI on GitHub NCBI RSS feed Connect with NLM NLM on X (formerly known as Twitter) NLM on Facebook NLM on YouTube National Library of Medicine 8600 Rockville Pike Bethesda, MD 20894 Web Policies FOIA HHS Vulnerability Disclosure Help Accessibility Careers NLM NIH HHS USA.gov Back to Top