IAES Inter national J our nal of Articial Intelligence (IJ-AI) V ol. 15, No. 1, February 2026, pp. 97 ∼ 105 ISSN: 2252-8938, DOI: 10.11591/ijai.v15.i1.pp97-105 ❒ 97 Articial intelligence in orthodontics: modeling decision support systems f or tr eatment planning So wmya Lakshmi Belur Subramanya, Adv aith V ijaya Mohan, Achala V arsha V ishla v ath Pr emalatha, Manchikanti V arunsai Department of Machine Learning, B.M.S. Colle ge of Engineering, Beng aluru, India Article Inf o Article history: Recei v ed Aug 12, 2024 Re vised Oct 15, 2025 Accepted Dec 15, 2025 K eyw ords: Articial intelligence Machine learning Orthodontics Predicti v e modeling T reatment planning ABSTRA CT Orthodontic treatment planning in v olv es comple x clinical decision-making that can benet from articial intel ligence (AI). This study e v aluates machine learning and deep learning models—including random forest, AdaBoost, gradient boosting, and articial neural netw orks (ANNs)—for predicting orthodontic treatment strate gies using a dataset of 612 anon ymized patient records wit h 66 clinically v alidated features across four cate gories (e xtraction, non-e xtraction, functiona l appliance, and orthopedic case ). Preprocessing included imputation, normalization, and synthetic minority o v ersampling technique (SMO TE) for class imbalance, while performance w as assessed via 10-fold cross-v alidation. Results sho wed that ANNs achie v ed the highest balanced accurac y (0.83), F1-score (0.84), and recei v er operating characteristic - area under the curv e (R OC-A UC) (0.90), outperforming ensemble and baseline models. Shaple y additi v e e xplanations (SHAP) analysis conrmed clinically meaningful predictors such as v ertical f ace proportions and mandib ular plane angle, enhancing interpretability . Although promising, the study is limited by its single-institution dataset and lack of e xternal v alidation. Future research should incorporate multicenter , multimodal datasets and interpretable-by-design frame w orks t o enable clinically trusted AI decision-support systems in orthodontics. This is an open access article under the CC BY -SA license . Corresponding A uthor: So wmya Lakshmi Belur Subraman ya Department of Machine Learning, B.M.S Colle ge of Engineering Bull T emple Road, Basa v anagudi, Beng aluru, Karnataka, India Email: so wmyalakshmibs.mel@bmsce.ac.in 1. INTR ODUCTION Orthodontics, a branch of dentistry , subject to diagnose, pre v ent, and correct malocclusions, or misalignments, of the teeth and ja ws [1]. Be yond enhancing the aesthetic appearance of smiles, orthodontic interv entions play a vital role in impro ving o v erall oral health and functional well-being [2]. T raditional orthodontic treatments in v olv e a meticulous process of clinical e xaminations, radiographic analyses, and the use of dental impressions to create indi vidualized treatment programs based on the particular requirements of each patient. Recent adv ancements in articial intelligence (AI) ha v e spark ed a transformati v e shift in orthodontic care, signicantly enhancing precis ion and ef cienc y through sophisticated algorithms and machine learning [3]. AI can analyze v ast datasets of clinical information, unco v er patterns, and generate predicti v e insights that assist orthodontists in treatment planning, thereby augmenting clinical decision-making and e xpertise. In addition to increasing treatment prediction accurac y , this feature enables more indi vidualised treatment plans J ournal homepage: http://ijai.iaescor e .com Evaluation Warning : The document was created with Spire.PDF for Python.
98 ❒ ISSN: 2252-8938 that are catered to the requirements of each patient. The application of AI in orthodontics encompasses a spectrum of machine learning models, including ensemble techniques such as AdaBoos t and gradient boost, as well as deep learning methods lik e articial neural netw orks (ANNs). These models le v erage comple x datasets and comprehensi v e patient histories, to deli v er insights that surpass traditional analytical capabilities, optimizing treatment outcomes and appliance design [4]. Be yond data analysis, AI inte gration automates routine tasks lik e image analysis and treatment simulation, which allo ws orthodontists to allocate more time and attention to direct patient care. Real-time feedback from AI systems, dri v en by current clinical data, empo wers orthodontists with timely insights for rened treatment planning and impro v ed precision in clinical outcomes [5], [6]. This paper critically e xamines AI’ s transformati v e impact on orthodontic treatment planning. It assesses the performance of adv anced AI techniques in predicting treatment outcomes and enhanci ng care precision. Emphasizing the potential of ensemble methods and deep learning models, this study underscores AI’ s capacity to re v olutionize orthodontic practice by ef fecti v eness, enhancing clinical judgement, and ultimately enhancing patient satisf action. 2. LITERA TURE SUR VEY The inte gration of AI into orthodontics has accelerated in the past decade. It spans diagnosis, tr eatment planning, monitoring, and orthodontic sur gery . Scoping re vie ws conrm that AI-dri v en methods enhance diagnostic precision, reduce inter -e xaminer v ariability , and support e vidence-based treatment planning [7]. Diagnosis and cephalometric analysis, AI has been widely applied to cephalometric landmark detection, a traditionally labor -intensi v e task. Con v olutional neural netw orks, U-Net, and ense mble learning approaches ha v e achie v ed landmark detection accuracies e xceeding 92%, reducing manual errors and increasing reproducibility [8], [9]. Automated cephalometric systems demonstrate substantial impro v ements in diagnostic reliability and standardization across institutions [10]. T reatment planning and decision support, machine learning models, and ANNs—ha v e been used to predict orthodontic treatment strate gies. Studies report accuracies up to 87% in distinguishing e xtraction v ersus non-e xtraction cases and recommending appliances [11]–[13]. AI-po wered decision-support systems also impro v e communication by pro viding patients with treatment outcome simulations [14]. Orthognathic sur gery planning, AI models ha v e also been applied in sur gical conte xts. 3D con v olutional neural netw orks, reinforcement learning, and generati v e models predict sur gical outcomes, simulate osteotomy procedures, and impro v e pre-operati v e planning. Performance metrics such as Dice similarity (0.85–0.90) conrm clinical feasibility [15], [16]. These tools enable personalized sur gical planning, reducing operati v e times and enhancing post-sur gical stability . T reatment monitoring and progress assessment, AI has been e xtended to monitoring treatment progression. Deep learning methods including recurrent neural netw orks (RNNs) and Siamese netw orks analyze sequential progress images and wearable sensor data to detect compliance issues and early relapse. Reported precision and recall v alues e xceed 0.80, sho wing promising reliability for real-time clinical feedback [17], [18]. Such systems assist clinicians in early interv ention and indi vidualized monitoring. Challenges and g aps, despit e adv ances, se v eral g aps remain. Most datasets are single-inst itutional, limiting generalizability across populations [19]–[21]. Interpretability is often inadequate, creating barriers to clinical trust [22], [23]. Ethical concerns such as data pri v ac y and re gulatory appro v al pathw ays further constrain deplo yment [24]–[26]. Addressing these issues requires multicenter collaborations, interpretable-by-design models, and inte gration into orthodontic w orko ws. 3. METHOD The suggested AI-based orthodontic treatment planning model’ s general w orko w is sho wn in Figure 1. Ev ery step of the procedure is described in the o wchart. These steps include dataset g athering, preprocessing, model training, v alidation, and interpretability analysis. 3.1. Dataset composition and sampling The dataset w as curated from anon ymized patient case histories pro vided by the Go v ernment Dental Colle ge and Research Institute, Beng aluru, under institutional ethical appro v al. From an initial pool of 97 clinical features, 66 were identied as most rele v ant in consultation with orthodontic e xperts. These included Int J Artif Intell, V ol. 15, No. 1, February 2026: 97–105 Evaluation Warning : The document was created with Spire.PDF for Python.
Int J Artif Intell ISSN: 2252-8938 ❒ 99 Care y’ s analysis, composite analysis, v ertical f ace proportions, mandib ular plane angle, and sella-nasion-point A/sella-nasion-point B (SN A/SNB) v alues. A total of 612 patient records were a v ailable, stratied into four treatment cate gories: e xtraction (27.6%), non-e xtraction (32.8%), functional appliance (21.7%), and orthopedic case (17.9%). Stratied sampling ensured proportional representation across classes. Exclusion criteria included incomplete records and ambiguous treatment plans. Figure 1. Flo wchart of the proposed model 3.2. Data pr epr ocessing Data preprocessing w as designed to ensure consistenc y and f airness across models. Missing v al ues were imputed using the median (numerical) or mode (cate gorical). Label-encoding w as used for binary cate gorical data, whereas one-hot encoding w as used for multi-class features. Numerical features normalized via min–max scaling (0–1). T o handle class imbalance, the synthetic minority o v ersampling technique, or SMO TE, w as used to augment minority classes and reduce bias. 3.3. Model de v elopment Six models were implemented: random forest, AdaBoost, gradient boosting, and ANN, along with tw o baseline classiers (logistic re gression and decision tree) for benchmarking. – Random forest: random forest b uilds se v eral decision trees and combines their results as sho wn in (1), where B is the number of trees and f i ( x ) is the prediction of the i th tree. ˆ y R F = 1 B B X i =1 f i ( x ) (1) – AdaBoost: AdaBoost adjusts instance weights iterati v ely to impro v e classication as in (2), where D t ( i ) is the weight of instance i , α t is classier weight, h t ( x i ) is prediction, and Z t is a normalization f actor . D t +1 ( i ) = D t ( i ) exp( − α t y i h t ( x i )) Z t (2) – Gradient boosting: gradient boosting minimizes residual loss by iterati v ely adding weak learners as depicted in (3), where f n ( x ) are weak learners (decision trees) and K is the number of iterations. ˆ y GB = N X n =1 f k ( x ) (3) – ANN: ANNs consist of layers of neurons with we ighted connections. F orw ard propag ation is dened as gi v en belo w , where W ( l ) is the weight matrix, b ( l ) bias, a ( l − 1) pre vious acti v ation, and g ( l ) acti v ation function (ReLU or sigmoid). Dropout re gularization w as applied to minimize o v ertting. z ( t ) = W ( t ) a ( t − 1) + b ( t ) (4) a ( t ) = g ( t ) ( z ( t ) ) (5) – Baselines: logistic re gression and decision trees were i mplemented as baseline models to conte xtualize performance impro v ements of adv anced algorithms. Articial intellig ence in orthodontics: modeling decision ... (Sowmya Lakshmi Belur Subr amanya) Evaluation Warning : The document was created with Spire.PDF for Python.
100 ❒ ISSN: 2252-8938 3.4. V alidation strategy T o mitig ate o v ertting 10-fold cross-v alidation w as emplo yed. Statistical signicance of dif f erences between models w as assessed using paired t -tests and W ilcoxon signed-rank t ests ( p < 0 . 05 ). Ov erall accurac y for each fold w as computed as: Accur acy = T r ueP ositiv e + T r ueN eg ativ e T r ueP ositiv e + T r ueN eg ativ e + F al seP ositiv e + F al seN eg ativ e (6) 3.5. Inter pr etability analysis T o enhance clinical trust, model interpretability w as e xamined using: – Shaple y additi v e e xplanations (SHAP) v alues: identied k e y predictors inuencing model decisions. – Permutation importance: quantied performance drops when features were randomly shuf ed. – ANN visualization: decision boundaries and acti v ation patterns were inspected for clinical rele v ance. These insights were mapped to orthodontic principles, enabling clinicians to v alidate AI-dri v en recommendations. 4. RESUL TS AND DISCUSSION 4.1. Cr oss-v alidation perf ormance All models were tested for prediction performance using 10-fold cross-v alidation. T able 1 s ummarizes the performance results. ANN achie v ed the highest balanced accurac y (0.83) and macro-F1 (0.84), signicantly outperforming the ensemble and baseline model s ( p < 0 . 05 ). Gradient boosting and random forest sho wed competiti v e results, b ut struggled with minority class recall. T o further assess generalization, training and v alidation accurac y curv es are presented in Figure 2. ANN demonstrated smooth con v er gence with minimal o v ertting com pared to the ensemble methods, which sho wed lar ger g aps between the v alidation and training scores. ANN achie v ed the highest generalization ability with minimal o v ertting. T able 1. Cross-v alidation results (mean ± SD) Model Accurac y (%) Balanced acc Precision Recall F1-score R OC-A UC Logistic re gression 72.4 ± 2.1 0.69 ± 0.02 0.71 ± 0.03 0.70 ± 0.02 0.70 ± 0.03 0.75 ± 0.02 Decision tree 74.1 ± 2.8 0.71 ± 0.03 0.72 ± 0.03 0.71 ± 0.03 0.71 ± 0.03 0.77 ± 0.03 Random forest 79.8 ± 1.9 0.76 ± 0.02 0.78 ± 0.02 0.77 ± 0.02 0.77 ± 0.02 0.83 ± 0.02 Gradient boosting 81.2 ± 2.3 0.78 ± 0.02 0.80 ± 0.02 0.79 ± 0.03 0.79 ± 0.02 0.85 ± 0.02 AdaBoost 77.5 ± 2.6 0.74 ± 0.03 0.75 ± 0.03 0.74 ± 0.03 0.74 ± 0.03 0.81 ± 0.03 ANN 86.9 ± 1.5 0.83 ± 0.01 0.85 ± 0.02 0.84 ± 0.02 0.84 ± 0.02 0.90 ± 0.01 Figure 2. V alidation v ersus training curv es of accurac y for the models Int J Artif Intell, V ol. 15, No. 1, February 2026: 97–105 Evaluation Warning : The document was created with Spire.PDF for Python.
Int J Artif Intell ISSN: 2252-8938 ❒ 101 4.2. Confusion matrix analysis While cross-v alidation metrics highlight a v erage performance, confusion matrices pro vide insi ght into class-specic errors as sho wn in Figure 3. Logistic re gression and decision trees e xhibited high misclassication rates for functional appliance and orthopedic case cate gories. Ensemble models reduced these errors, b ut ANN demonstrated the most balanced performance across all four treatment cate gories. ANN sho ws fe wer misclassications, particularly in minority classes. This nding is clinically signicant because errors in minority classes may directly af fect treatment prescriptions, s uch as o v erlooking orthopedic interv entions or prescribing incorrect appliances. Figure 3. Confusion matrices of the models across four treatment cate gories 4.3. Inter pr etability analysis T o address interpretability , SHAP v alues were computed to e xamine feature importance. V ert ical f ace proportions, mandib ular plane angle, and SN A/SNB v alues consistently emer ged as k e y predictors inuencing treatment decisions. These align with established orthodontic principles, conrming that the models rely on clinically meaningful features rather than spurious correlations. 4.4. Comparati v e discussion T able 2 compares ANN, ensemble/baseline models, and logistic re gression/decision trees acros s performance, interpretability , ef cienc y , and clinical rele v ance. ANN sho wed the highest accurac y and clinical alignment b ut required more resources, while ensemble and baseline models performed moderately . Logistic re gression and decision trees were f astest b ut less accurate and less suited for comple x cases. Articial intellig ence in orthodontics: modeling decision ... (Sowmya Lakshmi Belur Subr amanya) Evaluation Warning : The document was created with Spire.PDF for Python.
102 ❒ ISSN: 2252-8938 T able 2. Comparison of ANN, ensemble/baseline models, and logistic re gression/decision tree Aspect ANN (deep learning) Ensemble/baseline models Logistic re gression/decision tree Predicti v e performance Achie v ed the highest accurac y with notable impro v ements in minority class recall Demonstrated moderate performance, particularly weak er in minority class prediction Produced comparati v ely lo wer accurac y across classes Interpretability SHAP analysis highlighted clinically meaningful predictors recognized by orthodontists Of fered limited insights into feature contrib utions Pro vided basic interpretability b ut lack ed clinical depth Computational ef cienc y Required greater comput ational resources and training time Sho wed moderate computational demand Deli v ered the f astest results b ut with reduced predicti v e po wer Clinical rele v ance Ef fecti v ely bridged AI outputs with e xpert orthodontic judgment Limited applicability in clinical decision-making Suitable for quick assessments b ut insuf cient for comple x cases 5. DISCUSSION AND LIMIT A TIONS 5.1. Comparati v e analysis of models The comparati v e results demonstrate that ANNs achie v ed the best o v erall performance across accurac y , balanced accurac y , F1-score, and R OC-A UC. Ensemble models such as gradient boosting and random forest performed competiti v ely b ut were less ef fecti v e at handling minority classes, which are clinically important. Baseline models (logistic re gression and decision tree) pro vided useful references b ut underperformed, conrming the adv antage of adv anced AI methods in orthodontic decision support. 5.2. Generalizability A k e y limitation of this study is the reliance on a single-institution dataset. Although s tratied sampling and SMO TE were applied to impro v e class balance, e xternal v alidation across mult iple clinics, di v erse populations, and v arying imaging equipment w as not performed. W ithout multicenter testing, generalizability remains uncertain. Future research should prioritize e xternal v alidation and domain adaptation techniques to ensure rob ustness under real-w orld clinical v ariability . 5.3. Inter pr etability and clinical trust While deep learning models deli v ered the highest predicti v e performance, their black-box nature poses challenges for clinical adoption. SHAP analysis w as emplo yed to highlight inuential features such as v ertical f ace proportions, mandib ular plane angle, and SN A/SNB v alues, which align with orthodontic diagnostic principles. Ho we v er , further ef forts to w ard interpretable-by-design models or visualizati o n of decision boundaries are needed to enhance transparenc y and clinician trust. 5.4. Clinical translation barriers Despite promising results, se v eral issues must be handled before AI tools can be deplo yed in orthodontics. Important obstacles in clinical deplo yment include achie ving re gulatory compliance (such as food and drug administration (FD A) and European Conformity (CE) certication) and safe guarding patient pri v ac y through strong anon ymisation and safe data-sharing procedures. The seamless inte gration of AI technologies into current orthodontic operations and the reduction of biases resulting from unrepresentati v e or unbalanced datasets are equally crucial. 5.5. Futur e dir ections Building on the ndings of this study , se v eral directions are recommended. T o impro v e model generalisability , future studies should concentrate on gro wing datasets to encompass multicente r and multiethnic groups. Comprehensi v e treatment planning can be further supported by incorporating multimodal data, such as radiographs, photos , and clinical records. Furthermore, in v estig ating domain adaptation and transfer learning strate gies might lessen performance de gradation in dif ferent data scenarios. Clinical trust depends on the creation of interpretable AI frame w orks that strik e a compromise between predicted accurac y and transparenc y . Lastly , to enable clinical trials and of cial certication of AI-based solutions, cooperation with orthodontic associations and re gulatory bodies w ould be essential. Int J Artif Intell, V ol. 15, No. 1, February 2026: 97–105 Evaluation Warning : The document was created with Spire.PDF for Python.
Int J Artif Intell ISSN: 2252-8938 ❒ 103 6. CONCLUSION The proposed w ork illustrates ho w AI can be used to optimize orthodontic treatments by systematically e v aluating ensemble models and deep learning architectures ag ainst baseline classiers. Among the models, ANNs achie v ed the highest balanced accurac y and F1-scores, particularly e xcelling in correctly identifying minority treatment cate gories. Ensemble methods such as gradient boosting and random forest pro vided competiti v e alternati v es with lo wer computational cost, while baseline models conrmed the relati v e g ains achie v ed by adv anced AI methods. Importantly , interpretability analysis using SHAP re v ealed that clinically meaningful features—including v ertical f ace proportions, mandib ular plane angle, and SN A/SNB v alues—were consistently prioritized by the models. This re inforces the clinical v alidity of AI-assi sted decision-making and pro vides transparenc y to bridge the g ap between algorithmic outputs and orthodontic e xpertise. Nonetheless, se v eral limitations must be ackno wledged. The reliance on a single-institution dataset constrains generalizability , and the absence of e xternal v alidation restricts broader applicability . Ethical concerns, re gulatory appro v al, and seamless w orko w inte gration also remain challenges for real-w orld adoption. Looking ahead, future w ork should e xpand datasets across multiple populations and clinical settings, inte grate multimodal inputs such as radiographs and photographs, and de v elop interpretable-by-design AI frame w orks. By addressing these challenges, AI-dri v en systems can e v olv e from e xperimental tools into clinically trusted decision-support systems that enhance orthodontic ef cienc y , reliability , and patient care. A CKNO WLEDGMENTS The management of B.M.S. Colle ge of Engineering, Beng aluru, is deeply appreciated by the authors for their unw a v ering support and funding under F aculty Research Promotion Scheme, which enabled this research. The y also wish to e xpress deep appreciation to the Go v ernment Dental Colle ge and Research Institute, Beng aluru for pro viding in v aluable data and their e xpertise in t he eld of orthodontics. Further thanks are due to all the f aculty members, their support and dedication ha v e been pi v otal in adv ancing this research. FUNDING INFORMA TION This research w as supported by B.M.S. Colle ge of Engineering, Beng aluru. The authors appreci ate the nancial and infrastructural support pro vided by the institution to carry out this study . A UTHOR CONTRIB UTIONS ST A TEMENT This journal uses the Contrib utor Roles T axonomy (CRediT) to recognize indi vidual author contrib utions, reduce authorship disputes, and f acilitate collaboration. Name of A uthor C M So V a F o I R D O E V i Su P Fu So wmya Lakshmi Belur ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ Subraman ya Adv aith V ijaya Mohan ✓ ✓ ✓ ✓ ✓ ✓ ✓ Achala V arsha V ishla v ath ✓ ✓ ✓ ✓ ✓ ✓ ✓ Premalatha Manchikanti V arunsai ✓ ✓ ✓ ✓ ✓ ✓ ✓ C : C onceptualization I : I n v estig ation V i : V i sualization M : M ethodology R : R esources Su : Su pervision So : So ftw are D : D ata Curation P : P roject Administration V a : V a lidation O : Writing - O riginal Draft Fu : Fu nding Acquisition F o : F o rmal Analysis E : Writing - Re vie w & E diting CONFLICT OF INTEREST ST A TEMENT Re g arding the publishing of this paper , the authors state that the y ha v e no conicts of interest. Articial intellig ence in orthodontics: modeling decision ... (Sowmya Lakshmi Belur Subr amanya) Evaluation Warning : The document was created with Spire.PDF for Python.
104 ❒ ISSN: 2252-8938 D A T A A V AILABILITY The corpus is not publicly accessible due to condentiality and pri v ac y agreements. Ho we v er , upon request from the corresponding author and institutional appro v al, we may mak e it a v ailable. REFERENCES [1] Y . M. Bic hu, I. Hans a, A. Y . Bic hu, P . P remjani, C. F .-Mir , and N. R. V aid, “ Applications of articial intelligence and machine learning in orthodontics: a scoping re vie w , ” Pr o gr ess in Orthodontics , v ol. 22, no. 1, Dec. 2021, doi: 10.1186/s40510-021-00361-9. [2] W . J. P ark and J.-B. P ark, “History and application of articial neural netw orks in dentistry , ” Eur opean J ournal of Dentist ry , v ol. 12, no. 4, pp. 594–601, Oct. 2018, doi: 10.4103/ejd.ejd 325 18. [3] N. Ahmed et al. , “ Articial intelligence techniques: analysis, application, and outcome in dentistry - a systematic re vie w , ” BioMed Resear c h International , v ol. 2021, no. 1, Jan. 2021, doi: 10.1155/2021/9751564. [4] A. M.-Gonz ´ alez, L. R.-Calatayud, N. G. d’Oli v eira, and J. M. U.-T orrent, “ Articial intelligence in orthodontics: where are we no w? a scoping re vie w , ” Orthodontics and Cr aniofacial Resear c h , v ol. 24, no. S2, pp. 6–15, Dec. 2021, doi: 10.1111/ocr .12517. [5] M . Strung a, R. Urban, J. Suro vk o v ´ a, and A. Thurzo, “ Articial intelligence systems assist ing in the assess ment of the course and retention of orthodontic treatment, ” Healthcar e , v ol. 11, no. 5, Feb . 2023, doi: 10.3390/healthcare11050683. [6] C. C. K o et al. , “Machine learning in orthodontics: application re vie w , ” in Embr acing No vel T ec hnolo gies in Dental Orthodontics, pr oceedings of the F orty-sixth Annual Moyer s Symposium and the F orty-fourth Annual International Confer ence on Cr aniofacial Resear c h , Ann Arbor , Michig an, v ol. 56, pp. 117–135, 2019. [7] Z. J. Akw aid and E. Almousa, “Shaping the future of ort hodontics with articial intelligence: an o v ervie w of inno v ations today , insights for tomorro w , ” A ustr alasian Orthodontic J ournal , v ol. 41, no. 1, pp. 88–99, Jan. 2025, doi: 10.2478/aoj-2025-0007. [8] R . Neeraja and L. J. Anbarasi, “ A critical re vie w of articial intelligence based techniques for aut omatic prediction of cephalometric landmarks, ” Articial Intellig ence Re vie w , v ol. 58, no. 5, Feb . 2025, doi: 10.1007/s10462-025-11135-8. [9] A . K. Subramanian, Y . Chen, A. Almalki, G. Si v amurth y , and D. Kae, “Cephalomet ric analysis in orthodontics using articial intelligence - a comprehensi v e re vie w , ” BioMed Resear c h International , v ol. 2022, no. 1, Jan. 2022, doi: 10.1155/2022/1880113. [10] J . T om ´ a ˇ sik et al. , “ AI and f ace-dri v en orthodontics: a scoping re vie w of digital adv ances in diagnosis and treatment planning, ” AI , v ol. 5, no. 1, pp. 158–176, Jan. 2024, doi: 10.3390/ai5010009. [11] L. Bijo y , H. Gopal, S. R, and K. K, “ Articial intelligence in orthodontics- a narrati v e re vie w , ” International J ournal of En vir onmental Sciences , v ol. 11, no. 12s, pp. 172–182, Jun. 2025, doi: 10.64252/0m2frt78. [12] M. Jihed, I. Dallel, S. T obji, and A. B. Amor , “The impact of articial intelligence on contemporary orthodontic treatment planning - a systematic re vie w and met a-analysis, ” Sc holar s J ournal of Dental Sciences , v ol. 9, no. 5, pp. 70–87, Jun. 2022, doi: 10.36347/sjds.2022.v09i05.001. [13] A. D. Real, O. D. Real, S. Sardina, and R. Oyonarte, “Use of automated articial intelligence to predict the need for orthodontic e xtractions, ” K or ean J ournal of Orthodontics , v ol. 52, no. 2, pp. 102–111, Mar . 2022, doi: 10.4041/kjod.2022.52.2.102. [14] A. Y ounus A, N. Ahmed, U. A ymen, and N. A. Rahul, “ Articial intelligence in orthodontics: a w ay to w ards modernization, ” IP Indian J ournal of Orthodontics and Dentofacial Resear c h , v ol. 9, no. 1, pp. 3–7, Mar . 2023, doi: 10.18231/j.ijodr .2023.002. [15] S. B. Khanag ar , K. Alfouzan, M. A w a wdeh, L. Alkadi, F . Albala wi, and M. A. Alghilan, “Performance of articial intelligence models designed for diagnosis, treatment planning and predicting prognosis of orthognathic sur gery (OGS)—a scoping re vie w , ” Applied Sciences , v ol. 12, no. 11, May 2022, doi: 10.3390/app12115581. [16] R. M. Kato, J. de O. L. P arizotto, P . H. J. de Ol i v eira, and J. R. Gonc ¸ alv es, “ Articial intelligence in orthognathic sur gery–a narrati v e re vie w of sur gical digital tools and 3D orthognathic sur gical planning, ” J ournal of the California Dental Association , v ol. 51, no. 1, Dec. 2023, doi: 10.1080/19424396.2023.2202444. [17] X. Guo and Y . Shao, “ AI-dri v en dynamic orthodontic treatment management: personalized progress tracking and adjustments—a narrati v e re vie w , ” F r ontier s in Dental Medicine , v ol. 6, Aug. 2025, doi: 10.3389/fdmed.2025.1612441. [18] F . Miranda et al. , “ Articial intelligence applications in orthodontics, ” J ournal of the California Dental Association , v ol. 51, no. 1, Dec. 2023, doi: 10.1080/19424396.2023.2195585. [19] R. Leonardi and N. V aiid, “ Articial intelligence in orthodontics: concerns, conjectures, and ethical dilemmas, ” International Dental J ournal , v ol. 75, no. 1, pp. 20–22, Feb . 2025, doi: 10.1016/j.identj.2024.11.002. [20] M. A. Hussain et al. , “ Articial intelligence in orthodontics, ” International journal of health sciences , pp. 9378–9383, May 2022, doi: 10.53730/ijhs.v6ns2.7445. [21] A. Moeini and S. T orabi, “The role of articial intelligence i n dental diagnosis and treatment planning, ” J ournal of Or al and Dental Health Ne xus , v ol. 2, no. 1, pp. 14–26, 2025, doi: 10.61838/kman.jodhn.2.1.2. [22] H. J. J. Lee, N. S. M. Zamani, A. Ashari, and R. A. W . M. Ho w , “T ransforming orthodontics: e xploring e v olution and applications of articial intelligence, ” International J ournal on Advanced Science , Engineering and Information T ec hnolo gy , v ol. 15, no. 2, pp. 395–401, Apr . 2025, doi: 10.18517/ijaseit.15.2.20157. [23] J. Lee, H. Hw ang, and S. Kim, “Deep learning in orthodontics: applications in cephalometric analysis and treatment planning, ” K or ean J ournal of Orthodontics , v ol. 53, no. 2, pp. 85–94, 2023, doi: 10.4041/kjo.2023.53.2.85. [24] F . Schwendick e, W . Samek, and J. Krois, “ Articial intelligence in dentistry: chances and challenges, ” J ournal of Dental Resear c h , v ol. 99, no. 7, pp. 769–774, Jul. 2020, doi: 10.1177/0022034520915714. [25] Y . Ren, C. F .-Mir , and N. R. V aid, “Future of articial intelligence in orthodontics: clinical translation and re gulatory challenges, ” Seminar s in Orthodontics , v ol. 28, no. 4, pp. 318–328, 2022, doi: 10.1053/j.sodo.2022.10.004. [26] P . L. T onellato, A. K. Subramanian, and N. Ahmed, “Explainable AI in dentistry: b uilding clinical trust through interpretability , ” F r ontier s in Dental Medicine , v ol. 4, 2023, doi: 10.3389/fdmed.2023.115678. Int J Artif Intell, V ol. 15, No. 1, February 2026: 97–105 Evaluation Warning : The document was created with Spire.PDF for Python.
Int J Artif Intell ISSN: 2252-8938 ❒ 105 BIOGRAPHIES OF A UTHORS Dr . So wmya Lakshmi Belur Subramanya an assistant professor in the Department of Machine Learning at BMS Colle ge of Engineering, has research interests in deep learning, machine learning, natural language processing, and inf ormation retrie v al. She graduated from Adichunchangiri Institute of T echnology , Chikmag alur in 2011 and completed her post-graduation at Dayanand Sag ar Colle ge of Engineering in 2013. She recei v ed her Ph.D. from V isv esv araya T echnological Uni v ersity for her thesis titled ‘cross-language information retrie v al for code-mix ed Kannada-English queries’ in 2021. She joined B.M.S. Colle ge of Engineering as an assistant professor in October 2021. She has 20 publications to her credit in international journals and conferences in the domains of natural language proces sing and information retrie v al. She can be contacted at email: so wmyalakshmibs.mel@bmsce.ac.in. Adv aith V ijaya Mohan recei v ed his Bachelor of Engineering de gree in Articial Intelligence and Machine Learning from B.M.S. Colle ge of Engineering, Beng aluru, India, in 2024. He is currently associated with Cisco Syste ms as a b usiness analyst, where he contrib utes to the de v elopment and implementation of AI-dri v en solutions across di v erse b usiness domains. His research interests include natural language processing, data science, and the application of articial intelligence in supply chain and healthcare systems. He has demonstrated strong procienc y in combining technical inno v ation with practical problem-solving. His w ork reects a continued dedication to adv ancing applied AI research and intell igent system de v elopment. He can be contacted at email: adv aithvij@gmail.com. Achala V arsha V ishla v ath Pr emalatha is currently w orking as a data scientist at A T&T . She has comple ted her Bachelor of T echnology in Articial Intelligence and Machine Learning from BMS Colle ge of Engineering. Her academic background has pro vided a strong foundation in computational intelligence and data-dri v en technologies. Her k e y research interests include machine learning, deep learning, natural language processing, and generati v e AI, with a focus on applying these techniques to solv e real-w orld challenges. In the current paper , she contrib uted to conducting a detailed literature re vie w , data collection, and model de v elopment. She played a k e y role in inte grating articial intelligence methods into the study frame w ork and ensuring the model’ s accurac y and ef cienc y . Her contrib utions reect a strong interest in adv ancing the practical applications of AI in emer ging interdisciplinary domains such as healthcare and automat ion. She can be contacted at email: achala vp@gmail.com. Manchikanti V arunsai holds a Bachelor of Engineering (B.E.) de gree in Articial Intelligence and Machine Learning from BMS Colle ge of Engineering, equipping him with a rob ust understanding of AI, data science, and natural language processing (NLP). Currently , he serv es as a b usiness intelligence analyst at Dell T echnologies while acti v ely eng aging in research focused on general articial intelligence and related elds. W ithin this study , he played a vital role in establishing the research methodology and pro viding essential resources. He can be contacted at email: v arunsaimanchikanti@gmail.com. Articial intellig ence in orthodontics: modeling decision ... (Sowmya Lakshmi Belur Subr amanya) Evaluation Warning : The document was created with Spire.PDF for Python.