Digital Convergence in Dental Informatics: A Structured Narrative Review of Artificial Intelligence, Internet of Things, Digital Twins, and Large Language Models with Security, Privacy, and Ethical Perspectives

Abstract

Background: Dentistry is undergoing a digital transformation driven by emerging technologies such as Artificial Intelligence (AI), Internet of Things (IoT), Digital Twins (DTs), and Large Language Models (LLMs). These advancements offer new paradigms in clinical diagnostics, patient monitoring, treatment planning, and medical education. However, integrating these technologies also raises critical questions around security, privacy, ethics, and trust. Objective: This review aims to provide a structured synthesis of the recent literature exploring AI, IoT, DTs, and LLMs in dentistry, with a specific focus on their application domains and the associated ethical, privacy, and security concerns. Methods: A comprehensive literature search was conducted across PubMed, IEEE Xplore, and SpringerLink using a custom Boolean query string targeting publications from 2020 to 2025. Articles were screened based on defined inclusion and exclusion criteria. In total, 146 peer-reviewed articles and 18 technology platforms were selected. Each article was critically evaluated and categorized by technology domain, application type, evaluation metrics, and ethical considerations. Results: AI-based diagnostic systems and LLM-driven patient support tools were the most prominent technologies, primarily applied in image analysis, decision-making, and health communication. While numerous studies reported high performance, significant methodological gaps exist in evaluation design, sample size, and real-world validation. Ethical and privacy concerns were mentioned frequently, but were substantively addressed in only a few works. Notably, IoT and Digital Twin implementations remained largely conceptual or in pilot stages, highlighting a technology gap in dental deployment. Conclusions: The review identifies significant potential for converged intelligent dental systems but also reveals gaps in integration, security, ethical frameworks, and clinical validation. Future work must prioritize cross-disciplinary development, transparency, and regulatory alignment to realize responsible and patient-centered digital transformation in dentistry.

1. Introduction

Over the past half-decade, dentistry has entered an era of digital convergence in which once-disparate technologies––Artificial Intelligence (AI), Internet of Things (IoT), sensor networks, high-fidelity Digital Twin (DT), and Large Language Model (LLM)––are increasingly interwoven into the clinical, research, and educational fabric of oral healthcare. Bibliometric mapping of 2020–2024 literature (Figure 1) reveals a steep year-on-year rise in publications that combine at least one of these technologies with dentistry-specific terms, underscoring a vibrant and rapidly expanding evidence base.

Several prior reviews have provided valuable insights into the adoption of emerging technologies in dentistry. For instance, Luke and Rezallah conducted a systematic review and meta-analysis assessing the diagnostic accuracy of AI models for caries detection, establishing strong evidence for Machine Learning (ML) applications in dental imaging [1]. Thurzo et al. presented one of the most comprehensive overviews of AI applications across various dental subfields, identifying key trends and use cases [2]. Javaid et al. introduced the Dentistry 4.0 framework, documenting the adoption of digital technologies like tele-dentistry during the COVID-19 pandemic [3]. Maddahi and Chen explored the role of IoT-driven digital twins in personalized and simulation-based dental care [4], while Bastani et al. performed a bibliometric analysis of digital health intervention in pediatric oral care [5]. These studies collectively shaped our approach, motivating a multi-technology synthesis that also emphasizes the ethical, privacy, and regulatory considerations that accompany the integration of AI, IoT, DTs, and LLMs in modern dental informatics. Accordingly, this structured narrative review synthesizes 146 primary studies and 18 commercial products published between 2020 and mid-2025 to (i) chart the state-of-the-art across AI/ML, LLMs, IoT, and DTs in dentistry; (ii) interrogate their reported evaluation metrics, deployment readiness, and real-world impact; and (iii) critically appraise the accompanying security, privacy, and ethical discourse. In total, the final selection comprised 94 AI-related articles, followed by 54 related to LLMs, 43 pertaining to IoT, and 32 focused on DTs. The predominance of AI-focused literature underscores a clear research emphasis on AI-driven applications within dentistry, with many studies across IoT and DT domains incorporating AI elements such as image classification, anomaly detection, or predictive modeling.

By weaving together these dimensions, we aim to provide clinicians, researchers, and policymakers with an integrated roadmap for responsible digital transformation in dental informatics.

2. Methodology

This structured narrative review aims to examine the convergence of emerging technologies, namely AI, IoT, DT, and LLMs, within the domain of dental informatics. Em- phasis is placed on their applications in clinical practice, diagnostics, patient interaction, education, and on ethical, privacy, and security implications. The methodology adopted adheres to structured review principles with PRISMA-inspired transparency, tailored to suit the technological and interdisciplinary nature of the topic. Figure 2 illustrates the taxonomy and methodological flow of the structured narrative review conducted in the paper. In addition, to enhance clarity and avoid redundancy, included studies were organized based on the type and depth of contribution. Studies with well-defined evaluation metrics, datasets, and model details were tabulated to support comparative analysis (e.g., accuracy, image types, model types). These primarily involved AI/ML or computer vision applications with benchmarkable results. In contrast, articles that provided conceptual innovations, domain-specific adaptations, or emerging use-cases, without detailed quantitative evaluation, were described narratively or presented as illustrative case studies. This structural separation enables readers to differentiate between empirically validated approaches and exploratory or conceptual contributions.

2.1. Search Strategy

A comprehensive literature search was conducted across multiple academic databases: PubMed, SpringerLink, IEEE Xplore, ScienceDirect, Web of Science, Scopus, and Google Scholar. The search focused on articles published between January 2020 and May 2025. The following Boolean search string was used with adaptations per database:

(“Artificial Intelligence” OR “Machine Learning” OR “Deep Learning” OR “Neural Net- works” OR “Natural Language Processing” OR “LLMs” OR “Large Language Models” OR “Internet of Things” OR IoT OR “Connected Devices” OR “Smart Devices” OR “Digital Twin” OR “Virtual Patient” OR “3D Modeling” OR “Simulation”) AND (dentistry OR dental OR “oral health” OR “maxillofacial” OR “orthodontics” OR “periodontics” OR “prosthodontics” OR “endodontics” OR “oral surgery”) AND (security OR cybersecurity OR privacy OR “data protection” OR “data breach” OR ethics OR ethical OR bias OR fairness OR explainability OR interpretability OR transparency OR “informed consent” OR “patient autonomy” OR “accountability” OR misinformation OR vulnerability OR risk OR compliance OR regulation).

The search retrieved 94 results from PubMed, 1450 from SpringerLink, and 278 from IEEE Xplore.

2.2. Screening and Selection Criteria

All search results were first deduplicated and then screened based on their titles and abstracts. Articles were included if they were published in English between 2020 and 2025; focused on dentistry or oral health-related domains; discussed the use or implications of AI, IoT, LLMs, or DTs; contained original experimental or clinical work (i.e., not review articles); included evaluation metrics or quantitative analysis; and addressed or intersected with themes such as ethics, security, privacy, or regulatory considerations. The selection of articles from 2020 onward was based on the need to capture the most recent advancements in AI-integrated dental technologies, which have accelerated significantly in the post- pandemic era due to the global shift toward digital healthcare solutions

Articles were excluded if they were unrelated to dental or oral health contexts, did not involve any of the targeted technologies, were solely reviews, editorials, or commentaries, lacked quantitative evaluation or validation, or focused purely on product promotion without sufficient methodological details.

After applying the inclusion and exclusion criteria, a final set of 146 peer-reviewed research articles and 18 relevant product/system descriptions were included for synthesis and analysis.

2.3. Data Extraction and Classification

Each included article was read in full. A structured spreadsheet was developed to capture metadata (title, authors, DOI), publication details, abstract-derived classification (original/review), relevance to dentistry, core technology (AI, IoT, LLM, DT), application domain (e.g., diagnostics, education, monitoring), sample size and data type (e.g., number of images), evaluation metrics used, and any ethical or regulatory themes.

Additionally, each article was mapped to a thematic section of the review to facilitate coherent synthesis. Articles were also analyzed for trends in publication year, keyword distribution, and methodological approach (e.g., computer vision, chatbot simulation, historical prediction). Figure 3 presents a bar chart illustrating the frequency distribution of consolidated keywords extracted from article titles. All BibTeX entries were parsed, and titles were tokenized and mapped to predefined keyword categories (e.g., “Deep Learning,” “Dentistry,” “AI”) using pattern matching and keyword normalization. The keyword frequencies were then aggregated, sorted, and visualized. The analysis indicates that AI-related terms occur most frequently, reflecting the centrality of AI in current research. Its integration with adjacent domains such as IoT, Digital Twins, and LLMs further emphasizes its pervasive role across technological disciplines.

3. Digital Advancements in Dentistry: Applications, Challenges, and Future Directions

This section explores each key digital advancement in dentistry, detailing its applications; the associated security, privacy, and ethical challenges (illustrated with case studies); and potential future directions as solutions.

3.1. Artificial Intelligence (AI) and Machine Learning (ML)

AI/ML have become foundational technologies in modern dentistry, enabling intelligent diagnostics, predictive treatment planning, and personalized patient care [2,6]. These technologies rely on continuous data acquisition from clinical records and imaging tools, followed by sophisticated processing and model training pipelines that generate actionable insights for dental professionals. The ecosystem also includes tools for clinician feedback, iterative retraining, and real-time visualizations, ensuring dynamic improvements in care delivery [7].

Figure 4 illustrates a complete AI/ML-enabled dental decision support architecture. The system starts with the ingestion of structured data from external electronic medical records (EMRs) and routes it through preprocessing pipelines. Insights are served via a web portal, while performance logs and clinician feedback loop into a retraining system. The architecture demonstrates how AI is deeply embedded in a continuous learning loop, from data curation to clinician-interfaced insight delivery, ultimately driving smarter and safer dental practices.

3.1.1. Applications of AI/ML in Dentistry

AI and ML applications in dentistry can be observed under various stages, such as diagnostics, treatment planning, practice management, and research.

In diagnostics, U-Net convolutional networks have achieved high accuracy (96.2%) in automatic tooth segmentation from periapical X-rays [8]. Other models such as YOLOv4 and EfficientNetV2 have been employed for cephalometric landmark detection and positioning on radiographs, while MobileNetV2 combined with Explainable Artificial Intelligence (XAI) aids in diagnosing dental diseases with improved transparency [9,10]. Furthermore, recurrent probabilistic neural networks (RPNN) have demonstrated exceptional accuracy (99.08%) in toothbrush posture recognition, outperforming conventional Convolutional Neural Network (CNN) and LSTM architectures while enabling low-computational real-time feedback on smartphones [11]. Complementarily, AI-assisted diagnostic tools, such as Planmeca Romexis, have achieved diagnostic consistency comparable to that of senior dental practitioners in detecting alveolar bone loss and periodontal pockets, showcasing their value in clinical periodontal assessments [12,13].

In dental treatment planning, AI and ML techniques are increasingly assisting clinicians with complex decision-making tasks. Traditional CNN-based models have been widely used to evaluate periodontal conditions and simulate implant positions in Cone–Beam Computed Tomography (CBCT) scans with high spatial precision, enhancing the fidelity of treatment planning systems [14,15,16]. Complementing these efforts, Shojaei and Augusto proposed a comprehensive comparative framework using multiple ML models, such as Logistic Regression, SVMs, Random Forests, and Neural Networks for orthodontic treatment decisions including tooth extraction, extraction patterns, and anchorage strategies. Their neural network model achieved a notable 93% accuracy for extraction decisions and 89% for extraction patterns, outperforming other classifiers in multi-output scenarios [17]. Furthermore, Gabbar et al. introduced Tooth.AI, a semantic network–driven AI system that leverages patient history, X-ray data, and incremental learning to personalize treatment suggestions. The system supports diagnosis of both dental and skeletal abnormalities and integrates cephalometric landmark analysis to guide treatment pathways [18]. Recent developments also include the integration of generative AI (GenAI) into diagnostic work-flows, such as Tooth GenAI, which employs large language models to interpret clinical scenarios and assist with real-time treatment planning [19,20]. These approaches signal a shift from heuristic-based to evidence-based planning, bringing AI closer to replicating expert-level reasoning in personalized dental care.

Practice management systems increasingly utilize Natural Language Processing (NLP) and chatbot models to streamline operations such as patient scheduling and query resolution, thereby significantly enhancing clinic efficiency [21,22,23]. These AI-driven chatbots, which merge AI with NLP, are capable of understanding and generating responses through complex algorithms and ML, fostering real-time, interactive experiences and even supplementing faculty support in educational settings [23]. Concurrently, interactive AI-powered visualizations, such as the Tooth GenAI system, analyze dental images and patient records to detect conditions like bone growth, bone loss, and tooth cavities, providing timely and precise diagnoses and treatment recommendations [19,24,25]. This minimizes subjective interpretation and manual analysis, improving both informed consent and overall patient engagement by simulating likely outcomes [19]. In research, Federated Learning (FL) frameworks are being applied to train anomaly detection models across various clinics while preserving crucial data privacy, allowing for the analysis of large quantities of text to better understand diseases and support more precise and personalized care in dentistry [22,26].

As observed in

Table 1. Comparative Analysis of Computer Vision based AI and ML Techniques Applied to Dental and Related Diagnostic Tasks.

Study	Focus Area	Dataset Size	Image Type	ML/DL Models	Best Accuracy Achieved
Caries Detection using Gabor + ML (Support Vector Machine (SVM), KNN) [27]	Dental Caries Classification	347 (augmented to 1041)	X-ray	Cubic SVM, Medium Gaussian SVM, Fine KNN, Weighted KNN	Cubic SVM (90.5%)
Dental X-Ray Image Analysis using CNN [28]	Dental Caries Classification	800 (real-time performance)	Dental X-rays	CNN compared with ResNet, DenseNet, U-Net	98.60% (CNN)
Interpretable DL for Dental Caries Forecasting [29]	Caries Detection (Panoramic)	Not specified	Panoramic Radiographs	CNN with Grad-CAM and LIME for interpretability	91.25%
Entangled CNN for Periodontitis Grading [30]	Periodontitis Staging	Unspecified (multiple periapical images)	Periapical Radiographs	End-to-end CNN with segmentation-classification linkage	Area Under Curve (AUC): 0.97, Dice Similarity Coefficient (DSC): 0.96 (Stage I), 0.94 (Stage II)
Gingivitis Detection using Hue-Saturation-Value (HSV) + CNN [31]	Early Gingivitis Detection	417 intraoral images	Intraoral Photos	CNN with custom HSV preprocessing	CNN + HSV (81.68%)
CNN vs Graph Neural Network (GNN) for Dental Disease Prediction [32]	Dental Disease Prediction	204 (Panoramic images)	Panoramic Radiographs	GNN and CNN models trained separately	GNN: 95%, CNN: 89.2%
Supernumerary Teeth Detection using CNN [33]	Supernumerary Teeth Identification	1200 annotated occlusal images	Occlusal Radiographs	CNN Model trained from scratch	CNN Model (94%)
ML-based Support for Orthognathic Surgery [34]	Dento-maxillofacial Diagnosis and Surgical Planning	1500 images	Spiral CT	BR-XGBoost, SVM, Naive Bayes, Neural Network, Discriminant Analysis	>90% (BR-XGBoost)
DL for Fractured Endodontic Instrument Detection [35]	Fractured Instrument Detection	5110 images (multiple subsets: 166, 269, 800, 859)	Periapical Radiographs	DenseNet201, EfficientNet B0, ResNet-18, VGG-19, MaxVit-T	AUC: 0.90, Matthews Correlation Coefficient (MCC): 0.81 (DenseNet201)
Early Caries Detection using CNN [36]	Caries Classification (early stage)	7670 images, 481 teeth	Occlusal Surface Images	Convolutional Neural Network (CNN)	Accuracy: 89.2%
NPC Segmentation on CBCT [37]	Nasopalatine Canal Segmentation and Furcation Detection	200 patients, 5958 images	CBCT	You Only Look Once (YOLO)	AUC: 0.95
CNN-based Bone Loss Detection [38]	Bone Loss Detection	1500 images (subsets: 435, 349, 43)	Periapical Radiographs	CNN, ResNet50, VGG16	Accuracy: 92.8%
Deep Caries Detection [39]	Deep Caries Classification	200 images	Bitewing Radiographs	ResNet18, CNN	ResNet18: 95.5%
Dental Pulp Stone Detection [40]	Pulp Stone Identification	381 teeth	Panoramic Radiographs	YOLOv5, Faster R-CNN	mean Average Precision (mAP): 0.93 (YOLOv5)

, recent research in dental diagnostics increasingly leverages CNNs and their variants across diverse imaging modalities such as panoramic radiographs, bitewings, intraoral photos, and CBCT scans. A strong emphasis is placed on early-stage disease detection including caries, gingivitis, and bone loss with multiple studies reporting accuracies above 90%. Notably, newer approaches are integrating interpretability tools (e.g., Grad-CAM, LIME) and comparing traditional CNNs with more advanced architectures like GNNs, YOLO, and DenseNet for specific tasks such as fracture detection and anatomical segmentation. This trend indicates a growing prioritization of both diagnostic precision and clinical applicability.

Table 2. Comparative Analysis of Top AI/ML Models in Dentistry.

Model	Application	Acc.	Prec.	Recall	F1-Score	Dataset Size
DenseNet121 [41]	Oral lesion classification	99%	99%	100%	99%	700 images
RPNN (Smart Toothbrush) [42]	Toothbrush posture	99.08%	-	-	-	Proprietary inertial dataset
HRNet [43]	Smartphone oral cancer detection	96.6%	84.3%	83.0%	83.6%	455 images
iCNN (Experimental Analysis) [44]	Dental caries prediction	98.34%	-	-	-	>500 images
Faster R-CNN–GoogleNet [45]	3D X-ray dental detection	94.18%	95.39%	90.7%	≈92.9%	Custom 3D dataset
Faster R-CNN (Panoramic) [46]	Tooth region detection and classification	91.7%	-	98.9%	≈90.2%	160 images

presents a comparative analysis of several prominent AI/ML models ap- plied in various dental domains. The models, including DenseNet121 [41], RPNN (Smart Toothbrush) [42], HRNet [43], iCNN (Experimental Analysis) [44], Faster R-CNN– GoogleNet [45], and Faster R-CNN (Panoramic) [46], demonstrate diverse applications ranging from oral lesion classification and toothbrush posture detection to oral cancer detection, dental caries prediction, and 3D X-ray/panoramic tooth detection and classification. The table highlights their performance metrics, such as accuracy, precision, recall and F1-score, where applicable, along with the size and type of datasets used for training and evaluation. Notably, DenseNet121 achieved high accuracy (99%) and F1-score (99%) for oral lesion classification, while RPNN showed 99.08% accuracy in toothbrush posture detection. HRNet exhibited an 83.6% F1-score for smartphone-based oral cancer detection, and iCNN achieved 98.34% accuracy in dental caries prediction. For image-based detection, Faster R-CNN-GoogleNet and Faster R-CNN (Panoramic) showed strong performance in 3D X-ray and panoramic tooth detection, respectively.

3.1.2. Case Study: AI-Assisted Caries Detection in the HUNT4 Oral Health Study

A notable case study demonstrating the application of AI in dentistry is the HUNT4 Oral Health Study. In this study, researchers developed a Deep Learning (DL) model named AI-Dentify to detect proximal caries on bitewing radiographs. The model was trained on a dataset of 13,887 bitewing images annotated by six dental experts. Among various architectures tested, the YOLOv5 model achieved the highest performance, with a mAP of 0.647 and a mean F1-score of 0.548. Notably, the AI model outperformed human clinicians in terms of precision and reduced false negatives, highlighting its potential to assist in accurate and efficient caries detection [47]. This case study offers strong validation for the integration of DL in clinical dentistry, demonstrating that AI models can not only match but surpass expert-level diagnostic accuracy under controlled conditions.

3.1.3. Existing AI Products in Dentistry

The integration of AI into dental practice has led to the development of various products aimed at enhancing diagnostic accuracy, treatment planning, and patient management. Some prominent AI-driven dental products include:

• Pearl’s Second Opinion: An FDA-cleared AI platform that analyzes dental radio- graphs to detect multiple conditions such as caries, calculus, and periapical radiolucencies. It supports bitewing and periapical X-rays for patients aged 12 and above, aiding dentists in identifying pathologies that might be overlooked during manual examination [48].
• VideaAI: This AI assistant for dental service organizations (DSOs) and practices focuses on enhancing care quality and productivity. It assists in the early detection of dental issues, thereby improving diagnostic accuracy and patient outcomes [49].
• DentalXrai Pro: An AI-powered software that aids dental practitioners in analyzing radiographs more accurately and consistently. It serves as a decision support tool, helping clinicians identify problems and potential treatments with greater speed and precision [50].
• Overjet: Utilizes AI to analyze dental X-rays, providing insights into caries detection, bone loss measurement, and other conditions. It integrates with practice management systems to streamline workflows and improve patient communication [51].
• Diagnocat: Offers AI-driven analysis of dental images, including panoramic and CBCT scans, to assist in diagnosis and treatment planning across various dental specialties [52].

These AI tools exemplify the growing trend of incorporating ML algorithms into dental diagnostics and treatment planning, aiming to enhance clinical decision-making and patient care.

3.1.4. Security, Privacy, and Ethical Challenges for AI/ML in Dentistry

C.1

Ethical concerns in patient-facing AI simulations: AI-generated visual treatment outcomes (e.g., orthodontic simulations) could lead to persuasive bias, wherein patients might consent to procedures based on unrealistic or overly idealized results [16,24].

C.2

Risks in using patient data for AI model training in dental research: Studies like FL show promise, but improper data governance or aggregation without anonymization could result in re-identification risks, especially in image datasets [26].

To address the challenges outlined above, the following subsection presents targeted solutions aimed at improving trust, safety, and compliance in AI/ML applications within dentistry. It transitions from identifying risks to proposing practical strategies for ethical, secure, and transparent AI integration.

3.1.5. Future Directions and Solutions

Advancing AI in dentistry necessitates the development and rigorous validation of algorithms using diverse, demographically representative datasets to mitigate diagnostic bias. Concurrently, enhancing the explainability (XAI) of high-performing models such as YOLO and EfficientNet is essential to ensure transparency in clinical decision-making and foster trust among healthcare professionals and patients [53,54,55].

Securing biometric data from CBCT and intraoral imaging demands the adoption of advanced encryption standards alongside robust cloud access policies and secure data management frameworks. Drawing from broader healthcare data breaches, it is imperative to enforce regular security audits and ensure compliance with regulatory mandates such as the Health Insurance Portability and Accountability Act (HIPAA) and General Data Protection Regulation (GDPR) [56,57].

The deployment of dental chatbots and automated scheduling systems must integrate end-to-end encryption protocols to safeguard PHI. Furthermore, to counter automation bias and ensure patient safety, system designs should embed mechanisms for human oversight, particularly in scenarios involving critical care or diagnostic follow-ups [58,59].

AI-generated visual treatment outcomes, including orthodontic simulations, should undergo stringent validation to ensure both realism and interpretability. These visual- izations must transparently communicate inherent uncertainties and limitations, thereby supporting informed patient consent and minimizing persuasive bias [60].

As FL gains traction in dental AI applications, future research must emphasize the establishment of ethical and secure data governance protocols. This includes implementing advanced anonymization methods and secure aggregation techniques to mitigate re-identification risks while preserving patient privacy during distributed model training [61,62].

3.2. Large Language Models (LLMs) in Dentistry

The LLMs are revolutionizing dental informatics by enabling intelligent synthesis of clinical literature, conversational AI tools, and decision-support systems. These models are increasingly used for patient education, summarizing Electronic Health Record (EHR) records, answering dental queries, and generating treatment documentation. However, realizing their full potential requires a comprehensive development and monitoring ecosystem that aligns with clinical standards, data security policies, and regulatory frameworks.

Figure 5 presents a modular pipeline for LLM integration in dentistry, structured across four phases. Phase 1 involves data acquisition and preparation, including collection of dental records and text cleaning. Phase 2 covers LLM fine-tuning using dental domain-specific objectives and evaluation with standard NLP metrics. Phase 3 enables real-world deployment with prompt engineering, clinical oversight interfaces, and integration into existing dental workflows. Finally, Phase 4 ensures long-term quality through performance monitoring, feedback loops, retraining modules, and compliance audits, supporting continuous improvement and trustworthy AI deployment in dental practice.

3.2.1. Applications of LLMs in Dentistry

Large Language Models (LLMs) are emerging as powerful tools in dental informatics across multiple domains, such as patient monitoring, clinical decision support systems, and patient education.

In patient monitoring, LLMs such as ChatGPT and Google Bard are emerging as valuable adjuncts in dental patient monitoring, particularly for follow-ups, symptom tracking, and patient education in populations with limited mobility or access. Umer et al. [63] identified patient query handling––including post-operative care and behavioral assessments as among of the most common real-world applications of LLMs in dentistry, though most studies remain at deployment level 3, indicating partial clinical readiness. Lv et al. [64] conducted a blinded evaluation of ChatGPT-3.5, ChatGPT-4, and Google Bard across 40 oral health queries, where ChatGPT-4 achieved a mean appropriateness score of 9.34 (SD 0.47) and harmlessness score of 9.72, comparable to human dental experts. Notably, it also demonstrated the highest response stability (95% across repeated queries), highlighting its consistency under clinical simulation. Farhadi Nia et al. [65] emphasized LLMs’ potential in remote monitoring by integrating them with Electronic Dental Records (EDRs) to streamline documentation and enable real-time patient interactions, especially in oral surgery and endodontic workflows. Despite these advantages, Liu et al. [66] found that GPT-4’s accuracy in global dental licensing examinations averaged only 72%, with performance varying by language and geography, underscoring the need for domain-specific fine-tuning before clinical deployment. Additionally, Eggmann et al. [67] warned of risks including misinformation, automation bias, and a lack of contextual understanding, which limit LLMs’ standalone use for critical monitoring tasks. Nevertheless, when deployed with appropriate prompting, oversight, and evaluation protocols, LLMs can serve as scalable, multilingual, and patient-centric tools to enhance access, continuity, and personalization of oral healthcare [68].

In clinical decision support, LLMs enhance the retrieval of evidence-based practices by synthesizing large datasets and recommending treatment protocols tailored to individual cases. Applications also extend to prosthodontics, where LLMs assist in interpreting free-text clinical notes to suggest crown placements or denture designs [19]. Gaber et al. [69] benchmarked multiple LLMs, including Claude 3.5 Sonnet and a Retrieval-Augmented Generation (RAG) workflow, using 2000 real-world MIMIC-IV-ED cases. Claude 3.5 Sonnet achieved superior triage and diagnostic predictions, with 81.5% of outputs rated clinically correct by experts, although performance fluctuated in unstructured scenarios. Oniani et al. [70] enhanced four LLMs (GPT-4, GPT-3.5, LLaMA, PaLM 2) with clinical practice guidelines using methods like Binary Decision Trees and Chain-of-Thought prompting, finding that guideline integration significantly improved recommendation accuracy in synthetic COVID-19 cases. Özbay et al. [71] assessed ChatGPT-4, ChatGPT-3.5, and Bard on 40 endodontic questions; ChatGPT-4 scored highest with statistically significant accuracy (p = 0.004) and the lowest misinformation rate. Shankar and Mohan [72] explored ChatGPT-4’s feasibility for evidence-based decision-making (EBDM) among general dentists, reporting moderate CRSS (12/28) and C-GAM (46.4%) scores, suggesting time-saving benefits but limited transparency in sourcing. Similarly, Dermata et al. [73] compared six LLMs in pediatric dentistry, with ChatGPT-4 scoring highest (8.08/10) across diagnostic and educational tasks, though all models exhibited variability and required expert oversight. Collectively, while LLMs like ChatGPT-4 show strong potential for CDS in dentistry, their integration must be guided by structured prompts, clinical validation, and critical human evaluation to ensure reliability and ethical deployment.

For patient education, LLMs personalize care plans and explain complex procedures using simplified language and multilingual support. Moreover, advanced LLMs integrated with image interpretation capabilities act as virtual mentors, OPG-SHAP, analyzing radiographs or CBCT scans and identifying caries, impactions, or bone loss [55,74]. Islam et al. [75] conducted a comparative study assessing the effectiveness of AI-generated educational content, produced by ChatGPT, versus traditional instruction in enhancing patient understanding and reducing anxiety during endodontic and restorative procedures. Using structured knowledge assessments and Likert-scale surveys across 100 patients, AI-based materials outperformed traditional methods in clarity (4.42 vs. 3.25), usefulness (4.63 vs. 3.50), and anxiety reduction (2.63 vs. 3.38; p < 0.001). Expert review confirmed high content accuracy (k = 0.75), though the study’s cross-sectional design and lack of long-term behavioral outcomes limit its generalizability. In a broader review, Tassoker Murat et al. [76] analyzed ChatGPT’s role in augmenting diagnostic precision, communication, and clinical efficiency in dentistry. The review highlighted ChatGPT-4’s utility in oral surgery and endodontics, along with its application in text mining, treatment planning, and cross-modal diagnostics. However, limitations included variability in response accuracy, hallucination of references, high computational costs, and concerns around data privacy and reproducibility. The authors emphasized the need for human oversight, integration of neural-symbolic models, and inclusive trials to validate AI tools, especially in low- and middle-income regions. Collectively, both studies underscore LLMs’ promise in dental education and diagnostics, while cautioning against overreliance without rigorous validation and critical evaluation.

Lastly, EHR integration allows LLMs to extract diagnoses, suggest billing codes, and facilitate real-time patient interaction within digital record systems [77]. Mustafa et al. [78] evaluated ChatGPT-3.5 and ChatGPT-4 for ICD-10 coding of complex clinical records and found that LLMs rivaled or exceeded human accuracy in difficult cases, reducing false negatives, a crucial advancement for documentation integrity. Arian et al. [79] introduced the “Dental Loop Chatbot,” leveraging QLORA and Retrieval-Augmented Generation (RAG) to deliver real-time, evidence-based clinical guidance; although the prototype lacks extensive validation, it lays a scalable foundation for resource-constrained settings. Kim et al. [80] developed a bilingual LLM (Y-KNOT-med-base) for seamless EHR integration in English and Korean. It achieved 75.2% on PubMedQA and 55.8% on KorMedMCQA, outperforming several baselines, indicating feasibility for multilingual on-premise applications. Chuang et al. [81] demonstrated how GPT-J prompt engineering can enhance RoBERTa-based Named Entity Recognition (NER) models for periodontal diagnosis extraction from EDRs, achieving an F1-score of 0.72. Despite its promise, the system needs refinement for broader diagnostic domains. Finally, Sousa et al. [82] bench- marked 10 LLMs for EHR de-identification in English and German, showing that in-context learning––especially one-shot prompting––effectively anonymizes clinical text, although challenges like over-redaction and missed identifiers remain.

Recent literature reflects a rapid surge in the application of Large Language Models (LLMs) such as ChatGPT, Gemini, Claude, and domain-specific chatbots, across diverse areas of dental informatics. A notable trend across multiple studies is the evaluation of these models in clinical decision-making, diagnostic reasoning, and patient education. Works by Şişman and Acar [83], Özcivelek and Özcan [84], and Gökcek Taraç and Nale [85] highlight the growing trust in AI chatbots for addressing patient concerns in oral surgery, prosthodontics, and pediatric trauma, respectively, showing that LLMs can generate informative and coherent responses. Similarly, studies in diagnostic contexts, such as those by Grinberg et al. [86] and Tomo et al. [87], explore AI-driven differential diagnosis for oral diseases, with promising yet varied results. Several investigations also extend this evaluation into educational settings, where tools like ChatGPT were tested for their ability to solve dental student assessments and explain orthodontic procedures [88,89].

Critically comparing these studies reveals both alignment and divergence in methodology and depth. While some, like Şişman and Acar [83] and Grinberg et al. [86], leverage clinical scenarios for realistic validation, others adopt more descriptive or survey-based as- sessments [89,90], limiting generalizability. A key limitation across most studies is the lack of standardized evaluation frameworks; performance metrics, if reported, vary widely, making cross-study comparison challenging. Moreover, only a few, such as Buldur and Sezer [91], consider the regulatory and ethical implications of deploying AI-generated content in healthcare. Despite these inconsistencies, the collective findings suggest a clear shift toward the integration of generative AI in dentistry, driven by its versatility and patient-centric potential. However, robust clinical validation, standardized metrics, and attention to ethical deployment remain critical for these technologies to transition from experimental tools to trusted clinical assets [83,84,85,86,87,88,89,90,91,92].

Table 3. Comparison of LLM performance studies in dentistry.

Study (Year)	LLMs Evaluated	Dental Domain	Key Outcome
Giannakopoulos et al. (2023) [93]	ChatGPT-3.5, GPT-4, Bard, Bing Chat	Mixed dentistry (20 questions)	GPT-4 scored highest (7.2/10 avg), significantly outperforming others
Chatzopoulos et al. (2025) [94]	ChatGPT-4.0, Google Gemini, Gemini Advanced, Microsoft Copilot	Periodontology (furcation defect management, 10 Qs)	Gemini Advanced scored highest in clarity & comprehensiveness; Copilot lowest in relevance. No significant overall difference among models
Büker et al. (2025) [95]	ChatGPT-4.0, 3.5, Gemini, Copilot, Copilot Pro	Endodontics (76 Qs, 3 Levels)	GPT-4 highest overall accuracy (82.5%), followed by GPT-3.5 and Gemini. While, Copilot and Copilot Pro gave least accuracy
Makrygiannakis et al. (2024) [96]	ChatGPT-3.5, GPT-4, Bard, Bing Chat	Orthodontics (exam Qs)	Bing outranked GPT-3.5 & Bard; GPT-4 relevant differences

provides a comparative overview of recent studies evaluating the performance of various large language models (LLMs) in dental applications. Across different domains such as general dentistry, pediatric dentistry, endodontics, and orthodontics, GPT-4 consistently demonstrated superior accuracy and reliability compared to earlier models like GPT-3.5 and other generative models such as Bard, Bing Chat, Gemini, and Copilot. The results highlight not only the rapid advancements in LLM capabilities but also the variability in performance based on clinical context and question type.

3.2.2. Case Study: Evaluating LLMs in Dental Implant Decision Support

A recent study assessed the effectiveness of various general Large Language Models (LLMs) in answering dental implant-related questions. The study compared models like ChatGPT- 4.0, Gemini Pro 1.5, Claude 3 Opus, and Qwen 2.0 72B. The evaluation focused on the models’ ability to provide accurate and relevant information to dental professionals. Results indicated that while all models showed potential, there were variations in performance, highlighting the need for domain-specific fine-tuning to enhance clinical decision support in dentistry [97]. This case study illustrates the growing trend of leveraging general-purpose LLMs for domain-specific tasks in dentistry, reflecting a broader shift toward integrating generative AI in clinical environments despite current limitations in contextual accuracy.

3.2.3. Existing LLM Applications in Dentistry

The integration of LLMs into dental practice has led to the development of various applications aimed at improving patient care, education, and administrative efficiency. Notable LLM-driven dental products and platforms include:

• The AI chatbot initiative from the Harvard School of Dental Medicine, which focuses on leveraging generative AI, specifically LLMs, to revolutionize dental education [98]. The project aims to create advanced virtual patient systems by engineering and fine-tuning LLMs with de-identified patient data. These systems will generate synthetic patient cases, complete with dental records, clinical parameters, and imaging data, providing dental students with an unlimited supply of simulated scenarios. The primary goal is to enhance students’ diagnostic, treatment planning, and clinical decision-making skills through immersive, experiential learning.
• LLM on FHIR: An open-source mobile application that allows patients to interact with their health records using LLMs. Built on Stanford’s Spezi ecosystem and utilizing OpenAI’s GPT-4, the app translates complex medical data into patient-friendly language, enhancing health literacy and accessibility [99].
• Denota AI, an LLM-powered web application designed to significantly streamline the creation of patient and clinical notes for dental professionals [100]. It leverages specialized AI to generate detailed dental notes in under 15 s from minimal typed or dictated input. Key features include highly customizable templates, a voice-to-notes function for converting spoken descriptions into formatted notes, and tools for tracking earnings per note.
• The Dentistry Dashboard AI Chatbot [101], a customizable virtual assistant for UK dental practices. It integrates with websites to manage patient inquiries, provide real-time quotes, and act as a virtual treatment coordinator. Key features include custom branding, pre-built chat flows for common scenarios, and automatic lead capture with data extraction. It aims to reduce administrative workload, offer 24/7 patient engagement, improve lead conversion, and enhance overall practice efficiency.

These applications demonstrate the growing role of LLMs in enhancing various aspects of dental practice, from patient education and communication to clinical decision support and administrative tasks. It is also observed that the majority of the LLM-integrated dental solutions are currently available for virtual receptionist/assistants.

3.2.4. Security, Privacy, and Ethical Challenges for LLMs in Dentistry

C.1

Accountability and hallucination risk in LLM-based clinical decision support: LLMs can generate plausible but clinically incorrect outputs (hallucinations), leading to unsafe recommendations and undermining trust [102].

C.2

Misinformation and bias in LLM-based patient education: LLMs may present outdated, biased, or culturally insensitive health information, potentially misleading diverse patient populations [103].

C.3

Image data security in LLM-guided image interpretation: Cloud-based LLMs used for interpreting radiographs risk unauthorized access if secure protocols and anonymization are not enforced [104].

3.2.5. Future Directions and Solutions

To ensure responsible use of LLMs in dentistry, future efforts should focus on integrating real-time fact-checking layers to prevent hallucinations and deploying domain-specific fine-tuned models. Bias mitigation strategies, such as adversarial training and cross-cultural dataset expansion, are essential for equitable patient education.

Clear institutional and regulatory guidelines are needed for accountability in clinical use. Secure LLM–EHR bridges must employ end-to-end encryption, FL, and audit logs. Consent-based chatbot design and transparent user disclosures will also build trust and compliance in patient-facing systems.

3.3. Internet of Things (IoT)

The introduction of IoT devices into dental hygiene practices has given rise to smart toothbrush systems capable of monitoring, analyzing, and personalizing oral health behaviors. These devices interface with mobile applications and cloud infrastructures to provide users with real-time feedback, performance summaries, and longitudinal trend analysis. Central to this ecosystem is a multi-layered architecture that captures raw sensor data, processes it locally, and, when enabled, shares insights with secure backend systems for further analytics and data storage.

Figure 6 presents a modular architecture for a smart toothbrush data pipeline. The system begins with user interactions, wherein brushing behavior is captured via onboard sensors and sent to a connected mobile application. The data are first stored temporarily in a local cache and then cleaned and normalized by a processing module. A brushing analysis engine interprets behavior using embedded algorithms, sending personalized insights to an interactive dashboard. Optionally, data may be securely transmitted to a cloud backend via HTTPS for advanced trend analytics and long-term recordkeeping. This architecture supports real-time user engagement while maintaining data privacy and scalability.

3.3.1. Applications of IoT in Dentistry

The IoT has introduced significant advancements in dentistry through the integration of smart sensors and connected devices.

In oral hygiene monitoring, smart toothbrushes with embedded Bluetooth sensors provide real-time feedback on brushing duration and technique, improving preventive care [105,106]. Wearable biosensors can detect saliva composition, chewing patterns, and microbial activity, aiding in early disease detection [107].

In equipment management, IoT-enabled sterilizers and compressors monitor operational parameters like temperature and pressure, alerting clinicians to maintenance needs and preventing malfunctions [108].

Remote diagnostics are facilitated through intraoral cameras and connected diagnostic tools that transmit images and sensor data to cloud systems, allowing off-site clinicians to provide timely interventions [109].

Treatment enhancement includes devices like electronic apex locators and xerostomia-assist devices that rely on wireless feedback to guide procedures [110], as well as neuro-responsive silent communication systems using dental-mounted sensors [111].

3.3.2. Case Study: ToMoBrush—Acoustic Sensing for Dental Health Monitoring

A recent study introduced ToMoBrush, an innovative dental health sensing system that utilizes off-the-shelf sonic toothbrushes to detect dental conditions through acoustic analysis. By capturing tooth resonance signatures during brushing, ToMoBrush can identify common dental issues such as caries, calculus, and food impaction. In evaluations involving 19 participants, the system achieved high detection accuracy, with ROC–AUC scores of 0.90 for caries, 0.83 for calculus, and 0.88 for food impaction. This approach highlights the potential of integrating IoT technologies into everyday dental hygiene tools for real-time, at-home oral health monitoring [112]. This case exemplifies a broader trend toward the consumerization of IoT-driven diagnostic tools, where conventional personal care devices are being transformed into intelligent health monitoring systems.

Table 4. Comparative Analysis of IoT Applications in Dentistry.

Model/Platform	IoT Components	Performance Metrics	Dataset Size
Teeth Pitfall Measurement System [113]	Arduino + ESP32, Sensors (Force, SpO2, Temp), Cloud (ThingSpeak)	High accuracy (not numerically specified), Real-time visual monitoring	Two subjects demo
IoT-based Pediatric Dental Health Framework [114]	Smart toothbrushes, salivary sensors, cloud dashboards	Qualitative results, focus on usability and early diagnosis	Not specified
Smart Oral Health Monitor [115]	Sensors + ML model (Random Forest, SVM, CNN, XG-Boost)	CNN: Accuracy 94.9%, AUC 0.98	Not specified
Smart Dental Health-IoT Platform [116]	MASK R-CNN, Mobile terminal apps, AI back-end	Diagnosis time reduced by 37.5%, Patient throughput up 18.4%	Real-world deployment in 10 clinics
Deep Learning IoT Toothbrush [117]	Camera + MobileNet + WiFi sync	Accuracy 88.12%, Precision 0.83, Recall 0.92, AUC 0.93, F1 0.87	Private dataset

presents a comparative overview of IoT applications in dentistry. These systems integrate various sensors, machine learning models, and cloud platforms to enable real-time monitoring, early diagnosis, and improved treatment efficiency. Notably, the Smart Oral Health Monitor achieved a CNN accuracy of 94.9%, while the Smart Dental Health–IoT Platform demonstrated significant improvements in diagnosis time and patient throughput.

3.3.3. Existing IoT Products in Dentistry

The integration of IoT technologies into dental care has led to the development of various smart devices aimed at enhancing oral hygiene and patient engagement. Notable IoT-driven dental products include:

• Oral-B iO: An advanced electric toothbrush equipped with AI-driven pressure sensors, multiple brushing modes, and real-time feedback through a connected app. It utilizes a linear magnetic drive for smooth operation and provides users with detailed brushing analytics to improve oral hygiene practices [118,119].
• Playbrush: A gamified oral care device designed for children. Playbrush transforms toothbrushing into an interactive game. It connects to a smartphone app via Bluetooth, providing real-time feedback and encouraging proper brushing techniques through engaging gameplay [120].
• Kolibree Ara: Recognized as the first toothbrush with embedded AI, Kolibree Ara offers real-time feedback on brushing habits and integrates with mobile apps to track and improve oral care routines. Its motion sensors and AI algorithms help users maintain effective brushing techniques [121].
• Beam Brush: A smart toothbrush that tracks brushing frequency and duration, and transmits data to a mobile app. Beam Brush also integrates with dental insurance plans, offering incentives for consistent oral hygiene practices [122].
• Sunstar GUM Play: Developed to make brushing more engaging, GUM Play attaches to a manual toothbrush and connects to a smartphone app. It provides real-time feedback and visualizations of brushing habits, aiming to improve technique and duration [123].

These IoT-enabled devices exemplify the shift towards connected dental care solutions, offering users personalized feedback and fostering better oral health habits through technology-driven engagement.

3.3.4. Security, Privacy, and Ethical Challenges for IoT in Dentistry

C.1

Security vulnerabilities in smart toothbrushes and biosensors: Insecure firmware and lack of encrypted channels in consumer dental IoT devices can expose sensitive health metrics [124]. For example, a smart toothbrush with insecure firmware is exploited by an attacker to gain access to the user’s home network, or to collect and sell personal oral hygiene data on the dark web.

C.2

Cyber risks in IoT-enabled equipment (e.g., sterilizers, compressors): These systems, if exposed on open networks, can be hijacked to cause workflow interruptions or extort clinics [125]. For example, a dental clinic’s IoT-enabled sterilization unit may be targeted by ransomware, preventing its operation and forcing the clinic to halt procedures until the system is restored, and impacting patient safety and clinic revenue.

C.3

GDPR/HIPAA compliance in remote dental monitoring: Dental IoT systems must ensure encrypted transmission and proper access control to comply with privacy laws [126]. For example, a remote dental monitoring platform collects patient data via IoT sensors but fails to implement proper access controls or encrypted data transmission, leading to a HIPAA violation and a significant fine.

C.4

Wireless interference or jamming of smart treatment devices: Electromagnetic interference can affect the reliability of wireless dental tools, risking patient outcomes [127]. For example, during a complex dental procedure, a critical IoT-enabled device (e.g., a smart anesthetic delivery system) experiences wireless interference from another device in the clinic, leading to a malfunction and potential patient discomfort or safety risk.

C.5

Neuroprivacy and ethics in silent communication devices: As brain-computer interfaces evolve, devices that translate neural signals for dental commands may unintentionally record cognitive data [128]. For example, a future silent communication device used by a dentist, which interprets neural signals for hands-free operations, inadvertently collects and transmits sensitive cognitive data, raising profound questions about mental privacy and data ownership.

3.3.5. Future Directions and Solutions

Future work must address security from the hardware design phase, enforcing secure boot and update mechanisms in smart dental devices. Device authentication using biometric or hardware tokens can prevent unauthorized access.

Encryption protocols like TLS and AES must be standard for all data transmission. Establishing a regulatory framework for the Internet of Dental Things (IoDT), aligned with GDPR and HIPAA, is critical. Engineering research should also explore electromagnetic shielding and dynamic spectrum management to minimize wireless interference. Finally, ethical neuroprivacy guidelines are needed to govern the use of neural-interface dental tools.

3.4. Digital Twins in Dentistry

DT technology in dental education creates an immersive learning ecosystem where students engage with virtual replicas of clinical scenarios to develop procedural knowledge, psychomotor skills, and clinical reasoning. These platforms combine artificial intelligence, simulation engines, and feedback systems to replicate patient cases and monitor user progress. A robust software architecture is essential to ensure that these systems provide accurate simulations, timely feedback, and curriculum-aligned learning objectives, all while maintaining adaptability and performance.

Figure 7 presents a modular architecture for a DT learning platform tailored for dentistry. The architecture includes four core components: (1) Intelligence & Decision Logic for evaluating learning goals; (2) Curriculum Management for integrating with learning management systems; (3) DT Engine for scenario generation, assessment, simulation, and real-time feedback; and (4) User Interaction & Data Flow modules for handling user inputs and delivering personalized feedback. The feedback loop supports both corrective and reinforcement logic, ensuring a dynamic, learner-centered experience aligned with real-world competencies and educational standards.

3.4.1. Applications of Digital Twins in Dentistry

DTs in dentistry refer to high-fidelity, real-time virtual replicas of a patient’s oral anatomy created through the integration of imaging technologies (e.g., CBCT, intraoral scanners), sensor data, and computational models. These DTs are revolutionizing dental practice through enhanced personalization and precision.

In diagnostics, DTs provide detailed 3D visualizations of the oral cavity that assist in identifying caries, fractures, impacted teeth, and bone density issues [129]. For treatment planning, DTs enable simulation of surgical procedures, orthodontic movements, and implant placements, reducing the risk of intraoperative surprises [130,131].

DTs also enhance patient communication by providing immersive visual representations of treatment outcomes, improving understanding and engagement [132]. In dental research, DTs are being used to validate biomechanical models of jaw function, simulate long-term wear scenarios, and train ML models on synthetic yet realistic anatomical data [133].

Table 5. Performance Comparison of DT Platforms in Dental Education and Planning.

System/Platform	Metric Evaluated	Value/%	Application Area
Diagnocat with CBCT and 3D Printing [52]	Accuracy in Implant Placement	>95%	Diagnostic imaging and surgical planning
VR-based Framework (Adaptive Learning + Haptics) [134]	Performance Improvement in Tooth Drilling Skills	Statistically Significant	Skill-based training and error correction
Simodont Dental Trainer [135]	Novice vs. Expert Detection Sensitivity	High	Bridge and Crown Practice; Individual Feedback
DentalVerse Multiuser VR System [132]	Student Engagement and Collaboration Score	87% Positive Feedback	Multi-user psychomotor training in implant surgery

summarizes the performance of key Digital Twin (DT) platforms in dental education and planning. Diagnocat showed over 95% accuracy in implant placement, while VR-based systems like Simodont and DentalVerse enhanced skill training, engagement, and individualized feedback. Overall, these platforms demonstrate significant potential in improving clinical training and decision-making.

3.4.2. Case Study: AI-Driven DT for Orthodontic Tooth Movement Prediction

A collaborative study by the University of Copenhagen and 3Shape developed a DT system to simulate orthodontic tooth movements. Utilizing detailed CT scans, researchers created virtual patients to train AI models capable of predicting tooth movement trajectories under various orthodontic interventions. This approach aims to reduce the trial-and-error in brace adjustments, enhancing treatment efficiency and patient outcomes [136]. This case study serves as an illustration of how DT technologies can be operationalized in clinical orthodontics, bridging simulation with precision planning

3.4.3. Existing DT Products in Dentistry

The integration of DT technology in dentistry has led to the development of various products that enhance diagnostics, treatment planning, and patient communication:

• Planmeca Romexis: A comprehensive software platform offering 3D imaging and CAD/CAM integration, enabling the creation of detailed digital twins for precise treatment planning [137].
• Cybermed OnDemand3D: A dental imaging software providing advanced 3D visualization and analysis tools, facilitating the development of patient-specific digital twins for implant planning and orthodontic assessments [138].
• Shape Dental System: Offers a suite of tools for digital impression taking, design, and manufacturing, supporting the creation of accurate digital twins for restorative and orthodontic applications [139].
• EnvisionTEC 3D Printers: Provides high-precision 3D printing solutions that, when combined with DT models, allow for the fabrication of customized dental appliances and prosthetics [140].

3.4.4. Security, Privacy, and Ethical Challenges for Digital Twins in Dentistry

C.1

Ethical risks in over-reliance on simulated planning: Digital twins may foster automation bias, diminishing critical clinical judgment [141]. A dentist, overly reliant on a DT’s simulated treatment plan, overlooks a subtle anatomical variation that the twin model did not accurately capture, leading to complications during the actual procedure.

C.2

Patient coercion through immersive twin-based visuals: Visually persuasive simulations may unduly influence consent decisions [142]. A highly realistic DT simulation of a cosmetic dental procedure is so visually compelling that it subtly pressures a patient into accepting an expensive treatment they might not have otherwise chosen, raising questions about true informed consent.

3.4.5. Future Directions and Solutions

Future efforts must focus on secure data handling for DT systems by employing advanced storage platforms that support cryptographic access control. In particular, blockchain technologies and decentralized identity frameworks provide traceable and tamper-proof mechanisms for managing patient-specific data streams and simulation logs [143].

To reduce automation bias, hybrid decision-making frameworks should combine DT simulations with real-time clinician oversight. This preserves human judgment and con- textual reasoning while leveraging model predictions. Informed consent tools should incorporate simplified explanations of DT functionality and data reuse policies [144].

Moreover, real-time integrity of DT streaming can be protected using zero-trust architectures, which employ decentralized authentication and continuous verification instead of perimeter-based defenses [145]. Furthermore, adherence to rigorous Verification, Validation, and Uncertainty Quantification (VVUQ) practices, as recommended by regulatory bodies, will be essential to ensure the clinical safety of AI-enhanced DT deployments [146]. Finally, ethical governance guidelines must be co-developed with clinicians, ethicists, patients, and technologists to foster transparency, accountability, and trust in DT-driven dental care.

While the challenges discussed in the above sub-section highlight technology-specific risks, many of these concerns such as data privacy, bias, consent, and explainability cut across all digital dentistry paradigms. These cross-cutting challenges are synthesized and discussed in the following section.

3.5. Cross-Technology Ethical Challenges in Smart Dentistry

While each emerging technologies in smart dentistry, AI/ML, IoT, LLMs, and Digital Twins introduce unique capabilities and risks, several ethical and security concerns recur across domains. We consolidate these shared challenges here:

C.1

Bias, Representation, and Explainability: Algorithms trained on non-representative datasets, whether in AI/ML models, LLMs, or DT simulations, risk diagnostic bias and unequal treatment outcomes for underserved populations [13,147]. Lack of explainability (XAI) exacerbates this, as clinicians may find it difficult to interpret or trust black-box outputs [148]. Several studies emphasize that ensuring model interpretability is essential for responsible deployment in real-world settings, especially in high-stakes decision domains like dentistry [149].

C.2

Automation Bias and Over-Reliance: Whether in AI diagnostic tools, LLM-based triage, or DT-driven planning, there is a growing tendency to over-trust algorithmic outputs, even in the presence of contradictory clinical signs. This automation bias can diminish practitioner vigilance and decision quality [150,151]. Hybrid frameworks that combine algorithmic insights with clinician oversight are recommended to maintain decision accountability [152].

C.3

Privacy and Data Security Risks: Sensitive data such as 3D scans, PHI, conversational logs, or real-time sensor streams, are susceptible to interception, misuse, or reidentification without robust encryption and governance protocols. This is observed in AI cloud services, IoT communication, LLM interactions, and DT data storage [153,154,155,156].

C.4

Informed Consent and Ethical Reuse of Data: Across all domains, patients are often unaware of the downstream uses of their data, especially for research beyond original treatment purposes (e.g., secondary use of DTs, AI model training, LLM tuning) [157,158]. This lack of transparency challenges the ethical validity of consent, which should be informed, specific, and voluntary. Traditional consent mechanisms are typically designed for single-use or treatment-specific scenarios that are insufficient to accommodate the iterative, evolving nature of AI/LLM training pipelines or DT dataset reusability. Studies highlight that broad consent frameworks often fall short in communicating risks of data aggregation, potential re-identification, or commercial repurposing of de-identified health data [152].

4. Discussion

This structured narrative review reveals that the integration of advanced computational technologies in dentistry has rapidly expanded from concept to practice. AI and LLM are leading the transformation, particularly in diagnostic tasks, patient education, and decision support. Studies on ChatGPT, Gemini, and domain-specific dental LLMs demonstrate considerable potential in enhancing patient engagement and simplifying clinical explanations [83,84,85]. However, these benefits often come with caveats: while models are capable of producing coherent and medically relevant outputs, consistency, accuracy, and bias mitigation remain significant concerns. Clinical reliability varies with case complexity, and most LLMs lack interpretability, which limits their acceptability in high-risk settings such as oral cancer diagnosis or surgical planning.

In parallel, image-based AI applications dominate diagnostic automation. DL frame- works like CNNs and EfficientNet have been used for identifying caries, pulpitis, periodontal disease, and orthodontic misalignments. These studies generally report high accuracies, but few employ large-scale, multi-center datasets or conduct comparative benchmarking with human experts. Moreover, many studies do not mention training-validation-test splits or external validation, which raises reproducibility concerns. IoT-based dental systems, though promising, remain underrepresented. A few notable works discuss smart prosthetics and real-time oral monitoring systems, but broader clinical deployment and systematic evaluations are scarce. Similarly, Digital Twins (DTs) and virtual patient simulations are still at the conceptual or prototyping stage, with minimal real-world application documented in the literature.

A critical gap highlighted by this review is the lack of emphasis on ethics, security, and privacy in the implementation of these technologies. Although keywords like “security,” “bias,” and “transparency” occur frequently in abstracts and metadata, only a limited number of studies include substantial discussion on patient data protection, algorithmic fairness, or explainability frameworks. For instance, some works evaluated the credibility of chatbot responses against regulatory guidelines (e.g., FDA) [91], yet these remain exceptions. Without thorough consideration of compliance (e.g., HIPAA, GDPR), informed consent, and clinician oversight, the rapid adoption of these tools may outpace their trustworthiness.

In contrast, domains such as intensive care and cardiology have implemented XAI with audit trails and GDPR-compliant chatbots to promote safety and interpretability [148,159]. Smart hospitals are increasingly adopting FL and privacy-preserving IoT frameworks to ensure secure, distributed health data processing while meeting data protection standards [160,161].

The technological landscape also exhibits a lack of convergence. Most works explore individual modalities such AI, IoT, LLMs, or DTs in isolation. The potential of synergistic systems that combine real-time IoT monitoring, AI-based prediction, and DT-based visualization with conversational interfaces remains largely untapped. Drawing inspiration from cardiology, where multimodal digital twins integrate sensors, EHRs, and predictive modeling for precision care [162,163,164], future dental systems can be guided toward ethically grounded, interoperable solutions.

To advance this field responsibly, future research should prioritize cross-disciplinary collaborations and standardization. Methodologically, researchers must improve transparency in dataset design, report comprehensive performance metrics (e.g., sensitivity, specificity, AUC), and employ robust evaluation strategies. Ethically, frameworks for XAI, risk stratification, and human oversight must be embedded by design. With such measures, dental informatics can move closer to the vision of equitable, precise, and patient-centered intelligent care.

5. Conclusions

This structured narrative review comprehensively explores the convergence of advanced technologies, AI, IoT, Digital Twins, and LLMs, within the domain of dental informatics. By synthesizing 146 research articles published between 2020 and 2025, the study highlights the transformative potential of these technologies across diverse dental applications, ranging from diagnostics and treatment planning to patient education and remote monitoring. Our analysis reveals that AI and LLMs are increasingly being adopted for clinical decision support and patient communication, while IoT and DT technologies show significant promise for personalized, real-time dental healthcare delivery.

A notable finding is the growing attention to ethical, privacy, and security considerations, particularly in applications involving patient data and autonomous systems. However, the review also identifies gaps in the standardization of evaluation metrics, limited integration of explainable AI models, and underrepresentation of real-world validations, especially in IoT and DT deployments.

Future research must focus on harmonizing data privacy regulations with technological innovation, ensuring model transparency, and fostering interdisciplinary collaborations to build trustworthy, secure, and patient-centric dental AI ecosystems. This review serves as a reference point for researchers and practitioners aiming to responsibly integrate cutting-edge digital technologies into modern dental healthcare systems.