Artificial intelligence (AI) has been increasingly applied to periodontal diagnostics across periapical, bitewing, panoramic radiographs, cone-beam computed tomography (CBCT), and intraoral photographs. Recent multicenter, external validation, and explainability-focused studies have advanced the field, yet variability in datasets, anatomical sites, reference standards, model architectures, and reporting practices introduces significant heterogeneity. A structured synthesis of current evidence is therefore warranted.
Main text
This review synthesizes 35 studies published between 2019 and 2025, evaluating AI applications in four diagnostic domains: detection of periodontal bone loss, measurement of alveolar bone levels, identification of furcation involvement, and detection of periapical lesions. Convolutional neural network (CNN)-based models using periapical radiographs achieved moderate-to-high diagnostic accuracy (0.82–0.85) and AUCs above 0.88, comparable to clinician performance. Panoramic radiographs yielded lower sensitivity and specificity than CBCT, where deep learning systems reached higher accuracy (up to 0.91) and superior volumetric assessment. Intraoral photographic analyses showed variable performance (0.46–1.00), largely due to inconsistent imaging and reference standards. Emerging trends include hybrid segmentation–classification architectures, transformer-based networks, and clinician-in-the-loop approaches. Determinants of performance encompass reference standard quality, dataset diversity, anatomical complexity, and adherence to STARD-AI and TRIPOD-AI reporting frameworks.
Conclusions
AI demonstrates clinically relevant diagnostic accuracy in periodontal imaging, especially for measurement standardization and decision support. Although autonomous diagnosis remains premature, integrating explainable, externally validated AI systems within clinician-guided workflows supported by standardized reporting offers a practical route toward clinical translation.
artificial intelligencecone-beam computed tomographydiagnostic accuracyintraoral imagingmachine learningpanoramic radiographyperiodontal bone lossperiodonticsThe author(s) declared that financial support was not received for this work and/or its publication.section-at-acceptancePeriodonticsIntroduction
Periodontal diseases are among the most prevalent chronic inflammatory conditions globally, affecting billions of individuals and contributing substantially to the global burden of oral disease (1, 2). Accurate and early diagnosis is essential to prevent progression and tooth loss. Conventional diagnostic approaches rely primarily on clinical probing and two-dimensional (2D) radiographic interpretation, both of which have recognized limitations. Clinical measurements are prone to examiner variability and may underestimate localized defects, while 2D radiographs provide only planar projections that can distort anatomical relationships and obscure bone morphology (3–5).
Artificial intelligence (AI), particularly deep learning (DL) methods such as convolutional neural networks (CNNs), has emerged as a powerful adjunct to conventional diagnostic workflows in dentistry (6–9). Over the last decade, AI applications have rapidly expanded across dental disciplines, including periodontology, endodontics, caries detection, and implant planning. Various imaging modalities have been explored, including periapical and bitewing radiographs, panoramic radiographs, cone-beam computed tomography (CBCT), and intraoral photographs (8, 10–16). Across these modalities, AI has been applied to critical diagnostic tasks, such as detecting periodontal bone loss (PBL), measuring alveolar bone levels, identifying furcation involvement, and detecting periapical lesions.
On periapical radiographs, large-scale diagnostic studies have shown that CNN-based classifiers can achieve accuracies between 0.82 and 0.85, with AUC values above 0.88, often comparable to or exceeding average clinician performance (6, 12, 17). Systematic reviews and meta-analyses have affirmed the trend toward high diagnostic performance, and in individual AI imaging studies, specificities as high as 0.98 have been reported—underscoring the promise of AI-assisted periodontal diagnosis (15, 18, 19). In panoramic radiographs, CNN-based models have been reported to achieve diagnostic accuracies exceeding 90%, in some cases outperforming dentists whose accuracy ranges around 76%–78%, indicating balanced sensitivity–specificity profiles (20). However, despite these strong internal results, our DeepLabv3+ model demonstrated a notable decline in performance when evaluated on external data, reflecting a pattern widely observed in dental AI systems. For example, Chau et al. (21) reported high internal accuracy using a CNN-based photographic system for gingivitis detection, yet external validation of the same mHealth tool in Chau et al. (22) showed a substantial drop in specificity. Such discrepancies highlight the impact of dataset shift, arising from variations in imaging devices, acquisition protocols, and software versions; device heterogeneity across clinical sites; and differences in lighting conditions, angulation, and overall image quality, particularly in real-world dental imaging. Collectively, these factors contribute to external generalizability limitations, emphasizing the need for rigorous multi-center external validation before AI systems can be reliably deployed across diverse populations and imaging environments. Recent advancements have expanded to CBCT imaging, enabling volumetric assessments and detection of complex anatomical changes. Deep learning–based segmentation models have demonstrated promising performance, achieving accuracies around 0.91 and AUC values up to 0.96 for alveolar bone loss detection, with moderate performance in furcation defect identification (AUC ≈ 0.81). These findings highlight the potential of AI to enhance diagnostic precision in 3D periodontal imaging (13, 23). Intraoral photographic applications, increasingly relevant for screening and teledentistry, have shown variable diagnostic performance (with some studies reporting classification accuracies as low as ∼0.46, others approaching 1.00), though these extremes are not consistent across all works, owing to differences in protocol, device type, and reference standard (14, 24, 25).
Methodological developments have also matured. Frameworks such as STARD-AI and TRIPOD-AI provide standardized guidance for transparent reporting of AI-based diagnostic accuracy studies (25–28). Studies are increasingly incorporating external validation, multicenter datasets, and explainable AI techniques, aligning dental AI research with broader medical imaging standards (29). Recent dental CBCT/CT studies—for example in tooth segmentation meta-analyses, mandibular canal or root resorption quantification, and gender/age estimation using attention-based CBCT models—demonstrate this trend and show excellent performance in external/test datasets (30–34). Several recent overviews have synthesized AI performance across dental fields and highlighted the importance of structured reporting, external validation, and clinician engagement for successful clinical translation. Notably, Jundaeng et al. (35) review trends in AI for periodontitis diagnosis, and Tuygunov et al. (36) provide a comprehensive overview of challenges and enablers in integrating AI into dental practice (35, 36).
Given this rapidly expanding and heterogeneous evidence base, a thematic narrative review is warranted to synthesize current literature on AI diagnostic accuracy in periodontics. This review groups studies thematically by imaging modality, diagnostic task, and AI architecture, while examining determinants of performance and implications for clinical integration. This study was conducted as a thematic narrative review rather than a systematic review, and therefore did not follow a formal PRISMA-based screening or quantitative meta-analytic protocol.
Main text—thematic narrative reviewPeriapical and bitewing radiographs
Periapical and bitewing radiographs have been the most widely used modalities for AI applications in periodontal diagnostics, given their routine use in clinical practice and relatively high spatial resolution. CNN-based models have achieved diagnostic accuracies between 0.82 and 0.85 with AUC values above 0.88, comparable to average clinician performance (12, 15). An overview of the key characteristics and diagnostic performance of the included studies is summarized in Table 1. Similar performance trends have also been observed in panoramic radiographs (19). Subsequent studies have leveraged larger multicenter datasets, advanced segmentation–classification pipelines, and external validation strategies to enhance generalizability (37, 38). For example, Yadalam et al. (38) conducted a multicenter external validation of deep CNNs on periapical radiographs for alveolar bone loss detection, reporting sensitivities above 0.90 and consistent performance across centers (38).
Summary of the 35 synthesized studies on AI diagnostic accuracy in periodontics (2019–2025).
No.
Author(s), Year
Imaging modality
Diagnostic task/focus
AI model/approach
Key findings (accuracy/AUC)
Ref.
1
AlGhaihab et al. (2025)
Periapical/Bitewing
Alveolar bone loss detection
Deep CNN
Accuracy >0.85; strong correlation with manual measures
(10)
2
Çelik et al. (2023)
Panoramic
Periapical lesion detection
CNN
Reliable lesion detection vs. clinicians
(11)
3
Hoss et al. (2023)
Periapical
Periodontal bone loss detection
Multiple CNNs
AUC >0.88; comparable to dentists
(12)
4
Tariq et al. (2023)
Systematic Review
Radiographic PBL detection
Meta-analysis
Pooled specificity 0.91–0.98
(15)
5
Iacob et al. (2025)
Systematic Review
PBL in 2D radiographs
Meta-analysis
Consistent high accuracy across CNNs
(18)
6
Kabir et al. (2021)
Periapical
Periodontitis stage grading
HYNETS hybrid CNN
AUC ≈0.97
(39)
7
Chang et al. (2020)
Periapical
Bone loss and staging
Hybrid DL model
Accuracy 0.94; improved reproducibility
(62)
8
Muhammed Sunnetci et al. (2022)
Periapical
Bone loss classification
Hybrid DL + ML
Improved interpretability
(63)
9
Dujic et al. (2023)
Periapical
PBL detection
Vision Transformer
High accuracy, good generalization
(64)
10
Mei et al. (2025)
Periapical
Disease diagnosis
Clinical knowledge-guided hybrid
Enhanced accuracy vs. baseline CNN
(37)
11
Widyaningrum et al. (2025)
Panoramic
Periodontitis detection/staging
Two-stage CNN
Accuracy 0.91–0.94
(40)
12
Chen et al. (2024)
Periapical
Early bone loss diagnosis
CNN
Accuracy 0.90; supports early detection
(41)
13
Krois et al. (2021)
Cross-modal
Dental image analysis
CNN generalizability
Model transferability validated
(42)
14
Kurt-Bayrakdar et al. (2025)
CBCT
Bone loss pattern detection
Deep CNN
Accuracy 0.91; volumetric precision
(13)
15
Xue et al. (2024)
Panoramic
Bone loss & periodontitis stage
DL classifier
Accuracy >0.84; AUC 0.92
(16)
16
Chatzopoulos et al. (2025)
Panoramic
Furcation defect classification
Systematic review
Highlights AI potential in furcation analysis
(20)
17
Shetty et al. (2024)
CBCT
Furcation involvement
CNN
Accuracy 0.91; AUC 0.98
(45)
18
Palkovics et al. (2025)
CBCT
Periodontal bone segmentation
DL segmentation
Accurate 3D bone contour detection
(50)
19
Pan et al. (2025)
CBCT
Mandibular canal localization
CNN
Robust multicenter generalization
(51)
20
Rashid & Gaghor (2025)
CBCT
Bone quantity assessment
DL quantification
Accurate cross-sectional measurements
(57)
21
Widiasri et al. (2023)
CBCT
Bone & canal segmentation
U-Net
High Dice score; reliable segmentation
(58)
22
Naufal et al. (2024)
CBCT
3D reconstruction
YOLOv8 segmentation
Sub-mm accuracy vs. reference
(56)
23
Zhang et al. (2025)
Panoramic
Furcation classification
Vision Transformer
AUC >0.95
(59)
24
Zhou et al. (2025)
CBCT
Tooth instance segmentation
Open DL framework
Improved cross-task transferability
(47)
25
Chen et al. (2023)
CBCT
Dental segmentation
CTA-UNet (CNN–Transformer)
Outperformed U-Net baseline
(48)
26
Zhao et al. (2025)
CBCT
Implant classification/segmentation
Multi-task learning
AUC 0.94; effective joint learning
(49)
27
Liu et al. (2023)
CBCT + Intraoral
Multimodal 3D fusion
DL fusion
Superior structural reconstruction
(46)
28
Mao et al. (2025)
Intraoral Photo
Periodontitis detection
Systematic review
AUC 0.80–0.93; variable standards
(14)
29
Tao et al. (2025)
Intraoral Photo
Periodontitis screening
DL photo processing
AUC 0.93; high sensitivity
(53)
30
Wen et al. (2024)
Intraoral Photo
Gingival inflammation grading
CNN with removal strategy
Accuracy 0.84–0.88
(54)
31
Felsch et al. (2023)
Intraoral Photo
Caries/hypomineralization detection
Vision Transformer
AUROC 0.93
(52)
32
Yadalam et al. (2025)
Periapical
External validation of bone loss classifier
Dual-embedding few-shot CNN
Sensitivity >0.90
(38)
33
Kot et al. (2025)
CBCT/CT
Tooth segmentation meta-analysis
Systematic meta-analysis
Pooled Dice ≈0.92
(32)
34
Baena-de la Iglesia et al. (2025)
CBCT
External root resorption quantification
AI-aided 3D analysis
High volumetric agreement
(30)
35
da Andrade-Bortoletto et al. (2025)
CBCT
Mandibular canal segmentation
Comparative validation
Accuracy >0.90
(31)
Hybrid deep learning models that first segment alveolar bone or tooth structures and then classify bone loss severity or disease stage have increasingly outperformed classification-only approaches, especially in quantification tasks. For example, HYNETS (Kabir et al.) reports AUC ≈ 0.97 for stage grading from periapical radiographs, using entangled segmentation and classification modules (39). More recently, Mei et al. (37) introduced a clinical knowledge–guided hybrid classification network combining localization and classification branches to improve periodontal diagnosis (37). Another work by Widyaningrum et al. (40) demonstrated a two-stage pipeline (detect → classify) for periodontal bone loss detection and staging (40). Explainability tools such as Grad-CAM and clinician-in-the-loop interfaces have enhanced interpretability and acceptance among practitioners' trust in AI-assisted periodontal diagnostics (7, 8, 23, 36). A recent systematic review found that AI-based PBL detection on periapical radiographs consistently achieved high specificity (0.91–0.98) and improved measurement reproducibility compared to manual assessments (18, 41, 42).
Panoramic radiographs and CBCT
Panoramic radiographs offer wide coverage but have inherent distortions and overlapping structures, challenging AI detection accuracy. CNN models trained on panoramic images have achieved accuracies around 0.80–0.84, with balanced sensitivity and specificity (13, 16, 20). Large dataset studies (e.g., Kurt-Bayrakdar et al. (13) show that deep learning models can approach or match clinician diagnostic performance for bone loss classification using panoramic radiographs, although performance drops in certain defect types such as vertical bone loss, areas in posterior regions, or where anatomical overlap is substantial (13, 16, 43).
CBCT provides volumetric information that allows AI models to evaluate complex periodontal structures in three dimensions. Recent CBCT-based investigations, including high-resolution CBCT texture-analysis work (71) have demonstrated that deep CNNs and advanced segmentation architectures can achieve high diagnostic performance across selected periodontal tasks. For instance, a ResNet101V2 model achieved approximately 0.91 test accuracy and an AUC of 0.98 for detecting furcation involvement on axial CBCT slices, while segmentation models have reported AUC values approaching 0.96 for total alveolar bone loss detection. Although performance varies across tasks and datasets, these findings highlight the strong potential of AI for volumetric periodontal assessment and underscore the need for larger, multicenter validation studies (13, 44, 45). Recent work has explored multimodal fusion, combining CBCT with other imaging sources. For example, a deep-learning fusion model integrating CBCT and intraoral mesh scans achieved superior accuracy for structural reconstruction, and frameworks combining CBCT+ panoramic registration have shown improved detection of furcation and alveolar bone patterns compared with single modalities (46, 47).
Recent CNN–transformer hybrid and multi-task frameworks have improved CBCT performance by integrating segmentation, classification and measurement into unified pipelines; examples include CTA-UNet and Swin/UNetR-based systems that leverage global attention for improved tooth/ bone segmentation and downstream classification (47–49). External-validation studies on multicenter CBCT datasets have demonstrated robust generalizability for anatomical localization and segmentation tasks (e.g., mandibular canal localization, periodontal defect segmentation), supporting cross-site deployment after appropriate validation (45, 50, 51).
Intraoral photographs
Intraoral photographic applications are gaining traction, especially for screening and teledentistry settings where radiographic facilities may be limited. Reported diagnostic accuracies vary widely (≈0.46–1.00), primarily because of inconsistent imaging protocols, variable lighting conditions, and heterogeneous reference standards (14). Recent studies using standardized imaging workflows and modern CNN/transformer or hybrid CNN-transformer architectures have improved performance in controlled settings—for example, a ResNet50-based multi-instance model reached an AUROC of 0.93 on internal and external tests for stage II–IV periodontitis, and DenseNet/CNN models have reported AUROCs around 0.80–0.84 for gingival inflammation grading—indicating that protocol standardization plus modern architectures can substantially increase photographic diagnostic accuracy (52–54).
A systematic review by Mao et al. (14) highlighted the promise of AI for photographic detection of periodontal disease but emphasized the urgent need for standardized acquisition protocols and multicenter datasets (14). Clinician-in-the-loop systems are particularly relevant here, allowing AI to serve as an assistive tool for remote triaging rather than as an autonomous diagnostic system.
Themes by diagnostic task
Grouping studies by diagnostic task provides insight into how AI performance varies according to the clinical endpoint:
Periodontal Bone Loss Detection: Most studies focus here, particularly on periapical and panoramic radiographs. CNNs and hybrid segmentation-classification architectures achieve high specificity (>0.90) and good sensitivity, especially for moderate-to-severe defects (12, 19, 40, 55).
Alveolar Bone Measurement: Multi-task deep learning models using YOLOv8 and ResNet backbones have demonstrated sub-millimeter error in alveolar crest measurement compared to CBCT gold standards (37, 56–58).
Periapical Lesion Detection: Transformer-based networks validated on multinational datasets showed superior sensitivity and external generalizability compared to baseline CNNs (17, 60, 61).
This task-based perspective underscores that AI performance is highest for volumetric and measurement tasks using CBCT, while 2D modalities remain highly useful for screening and routine clinical support.
Themes by AI architecture
Early studies relied mainly on standard CNN architectures (e.g., U-Net, ResNet) for either segmentation or classification. Over time, three major trends have emerged:
Segmentation–Classification Hybrids: Hybrid models combining segmentation and classification have improved bone contour localization and bone loss quantification, particularly in periapical radiographs. Notable examples include Chang et al. (62), Sunnetci et al. (63), Dujic et al. (64), and Widyaningrum et al. (40), which applied CNN-based hybrids for periodontal bone loss assessment, and the U-Net architecture (29) as a foundational segmentation framework.
Transformer-Based Models: Enhanced contextual understanding, particularly in panoramic and CBCT tasks, improving AUC by 3%–5% over CNN baselines (65–67).
Clinician-in-the-Loop & Explainable AI: Integration of Grad-CAM, attention maps, and uncertainty quantification to increase interpretability and user trust (68–70).
Recent methodological frameworks such as TRIPOD-AI and STARD-AI have influenced model development and reporting, encouraging external validation, clear reference standards, and transparent evaluation metrics (28, 71).
Discussion
Determinants of Diagnostic Performance:
Imaging Modality and Anatomical Complexity: CBCT provides volumetric information that enables superior AI performance in furcation detection and alveolar bone measurements, with accuracies up to 0.91 and AUC values above 0.95 (37). U-Net architectures (29) provide the foundational segmentation framework, while reporting standards such as TRIPOD-AI and STARD-AI (25, 26, 72) ensure transparent evaluation and external validation. Panoramic and intraoral photographs, by contrast, are more prone to distortions and variability, resulting in wider performance ranges (14, 36).
Dataset Diversity and External Validation: Earlier single-center studies often reported high internal accuracies that did not generalize when evaluated on external datasets. Recent validation work has reinforced this limitation. For example, Hadzic et al. (60) demonstrated that a periapical lesion detection CNN trained on one dataset showed reduced performance when tested on an independent, clinically representative CBCT dataset, underscoring the impact of real-world variability (60). More recent multicenter and cross-architecture investigations have emphasized similar issues. The comparative study by Schneider et al. (73), evaluating CNNs, transformers, and hybrid models across multiple institutions, showed that diagnostic accuracy and model calibration varied substantially when models were applied across heterogeneous datasets (73). Likewise, Yadalam et al. (38) highlighted the sensitivity of AI systems to dataset diversity, noting that their few-shot learning framework for bone-loss classification on OPG data remained highly dependent on training-domain characteristics and would require rigorous external validation for broader deployment (38). Collectively, these findings demonstrate that population differences, device heterogeneity, and broader domain-shift effects can significantly degrade performance, reinforcing the need for robust, multi-institutional external validation to ensure real-world applicability.
Reference Standards: Studies using CBCT or clinically validated measurements as reference standards consistently outperform those relying on single-expert or 2D radiographic assessments. The choice of gold standard directly impacts reported sensitivity and specificity, especially for alveolar bone measurements and furcation classification (18, 37, 41, 42).
Methodological Rigor and Reporting Quality: Adherence to frameworks such as STARD-AI and TRIPOD-AI has increased over the past two years, leading to more transparent study designs and reproducible results (26, 72). Explainability features and uncertainty quantification also play a role in clinician acceptance (23, 36).
Clinical implications
AI has significant potential to enhance periodontal diagnosis through standardized measurements, workflow efficiency, and decision support. High specificity in periapical radiograph applications may reduce inter-examiner variability and improve longitudinal monitoring in both specialist and general practice settings (12, 18, 19, 38, 41, 42). CBCT-based AI systems, with their volumetric capabilities, are especially suited for complex diagnostic tasks such as furcation involvement and surgical planning, providing quantitative metrics that complement clinician judgment (37). Foundational segmentation frameworks such as U-Net (29) support model development, while reporting standards like TRIPOD-AI and STARD-AI (25, 26, 72) ensure transparency and reproducibility.
For panoramic and intraoral photographic modalities, AI can serve as a screening or triaging tool, particularly in primary care and teledentistry contexts where access to CBCT is limited (14, 33). Clinician-in-the-loop models further ensure that AI outputs are interpreted within a professional diagnostic framework, reducing the risk of over-reliance on automated systems.
Policy and integration considerations
For AI to transition from research to clinical use, several policy and integration factors must be addressed:
Regulatory Alignment: AI diagnostic systems should align with regulatory frameworks (e.g., FDA, EMA, local dental boards) that emphasize transparency, explainability, and post-market surveillance.
Infrastructure and Data Governance: Successful clinical integration requires secure data pipelines, federated learning frameworks, and standardized imaging protocols to enable robust external validation.
Education and Workforce Integration: Dental education must include AI literacy to ensure clinicians can critically interpret AI outputs and integrate them into diagnostic reasoning.
These considerations are consistent with trends in medical imaging AI, where regulatory and infrastructural readiness significantly influences clinical adoption (26, 28, 68, 72).
Limitations
Despite rapid advancements, the evidence base remains limited by several factors:
Heterogeneity of Study Design: Differences in dataset size, anatomical sites, labeling strategies, and evaluation metrics complicate direct comparisons across studies.
Limited Prospective Trials: Most included studies are retrospective; few prospective or real-time clinical validation studies exist.
Variable Reporting Standards: Although STARD-AI and TRIPOD-AI adoption is improving, many studies still lack clear descriptions of preprocessing, validation, or error analysis.
Underrepresentation of Certain Populations: Multicenter datasets remain geographically concentrated, raising questions about generalizability to diverse patient groups.
Addressing these limitations will require coordinated methodological efforts, standardization, and regulatory oversight.
Beyond methodological constraints, several additional sources of bias must be considered in AI-assisted periodontal diagnosis. Dataset bias—including imbalances in disease prevalence, demographic distribution, and imaging conditions—can lead models to learn spurious associations rather than true pathological features. Reference-standard variability, particularly when ground-truth labels are derived from heterogeneous clinical judgments or non-standardized criteria, further undermines model reliability. Even clinician-provided annotations may lack consistency without calibration exercises, resulting in systematic differences in “ground-truth” labels that propagate into model errors. In conjunction with these factors, the risk of automation bias remains a major concern: clinicians may over-rely on algorithmic outputs, accept incorrect AI predictions, or overlook contradictory clinical evidence. This phenomenon has been documented in medical AI settings, where over-trust in automated systems can result in diagnostic complacency, reduced critical appraisal, and delayed error detection—especially when models encounter out-of-distribution cases (69–71). Implementing clinician-in-the-loop workflows, standardized labeling protocols, calibration sessions for annotators, and explainable AI interfaces is therefore essential to mitigate these risks and preserve clinical accountability in AI-supported periodontal decision-making.
Future directions
Future research should prioritize:
Multicenter Prospective Clinical Trials using standardized protocols and external validation to evaluate AI under real-world conditions.
Task-Specific Architectures, including transformer-based multi-task networks, to improve performance on complex volumetric tasks.
Explainable and Uncertainty-Aware AI, fostering clinician trust and regulatory compliance.
Integration of AI into Longitudinal Care, enabling dynamic tracking of periodontal changes over time and personalized treatment planning.
These directions align with broader trends in AI for medical imaging and offer a roadmap for meaningful clinical integration in periodontics.
Conclusions
Artificial intelligence has progressed rapidly in periodontal diagnostics over the past five years, moving from proof-of-concept studies to externally validated, multicenter applications across multiple imaging modalities. CNN-based systems for periapical radiographs consistently achieve high specificity and accuracy, supporting their use for standardized measurements and clinical decision support. CBCT-based models offer the highest diagnostic performance for volumetric tasks, including alveolar bone measurement and furcation involvement detection, while panoramic and intraoral photographic applications provide valuable tools for screening and triage, particularly in primary care and teledentistry settings.
Despite these advances, methodological heterogeneity, limited prospective evidence, and variable reporting standards remain barriers to widespread clinical adoption. The integration of reporting frameworks such as STARD-AI and TRIPOD-AI, combined with explainable AI techniques and clinician-in-the-loop workflows, offers a realistic pathway toward safe and generalizable deployment.
Future efforts should focus on multicenter prospective trials, regulatory alignment, and standardized imaging protocols to ensure robust, equitable, and clinically meaningful integration of AI systems into periodontal practice. AI should be viewed not as a replacement for clinician expertise but as a complementary tool that enhances diagnostic accuracy, consistency, and patient care.
The author(s) declared that this work was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
Generative AI statement
The author(s) declared that generative AI was not used in the creation of this manuscript.
Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.
Publisher's note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
Supplementary material
The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fdmed.2026.1729825/full#supplementary-material
ReferencesKassebaumNJBernabéEDahiyaMBhandariBMurrayCJMarcenesW. Global burden of severe periodontitis in 1990–2010: a systematic review and meta-regression. J Dent Res. (2014) 93(11):1045–53. 10.1177/002203451455249125261053PetersenPEOgawaH. The global burden of periodontal disease: towards integration with chronic disease prevention and control. Periodontol 2000. (2012) 60(1):15–39. 10.1111/j.1600-0757.2011.00425.x22909104ChakrapaniSSirishaKSrilalithaASrinivasM. Choice of diagnostic and therapeutic imaging in periodontics and implantology. J Indian Soc Periodontol. (2013) 17(6):711–8. 10.4103/0972-124x.12447424554878MolA. Imaging methods in periodontology. Periodontol 2000. (2004) 34:34–48. 10.1046/j.0906-6713.2003.003423.x14717854PapapanouPNSanzMBuduneliNDietrichTFeresMFineDHPeriodontitis: consensus report of workgroup 2 of the 2017 world workshop on the classification of periodontal and peri-implant diseases and conditions. J Clin Periodontol. (2018) 45(Suppl 20):S162–s70. 10.1111/jcpe.1294629926490LeeJHKimDHJeongSNChoiSH. Detection and diagnosis of dental caries using a deep learning-based convolutional neural network algorithm. J Dent. (2018) 77:106–11. 10.1016/j.jdent.2018.07.01530056118LundervoldASLundervoldA. An overview of deep learning in medical imaging focusing on MRI. Z Med Phys. (2019) 29(2):102–27. 10.1016/j.zemedi.2018.11.00230553609SchwendickeFSamekWKroisJ. Artificial intelligence in dentistry: chances and challenges. J Dent Res. (2020) 99(7):769–74. 10.1177/002203452091571432315260TopolEJ. High-performance medicine: the convergence of human and artificial intelligence. Nat Med. (2019) 25(1):44–56. 10.1038/s41591-018-0300-730617339AlGhaihabAMorettiAJResideJTuzovaLHuangYSTyndallDA. Automatic detection of radiographic alveolar bone loss in bitewing and periapical intraoral radiographs using deep learning technology: a preliminary evaluation. Diagnostics (Basel). (2025) 15(5):576. 10.3390/diagnostics1505057640075823ÇelikBSavaştaerEFKayaHIÇelikME. The role of deep learning for periapical lesion detection on panoramic radiographs. Dentomaxillofac Radiol. (2023) 52(8):20230118. 10.1259/dmfr.20230118HossPMeyerOWölfleUCWülkAMeusburgerTMeierLDetection of periodontal bone loss on periapical radiographs-A diagnostic study using different convolutional neural networks. J Clin Med. (2023) 12(22):7189. 10.3390/jcm1222718938002799Kurt-BayrakdarSBayrakdarİKuranAÇelikÖOrhanKJagtapR. Advancing periodontal diagnosis: harnessing advanced artificial intelligence for patterns of periodontal bone loss in cone-beam computed tomography. Dentomaxillofac Radiol. (2025) 54(4):268–78. 10.1093/dmfr/twaf01139908459MaoKThuKMHungKFYuOYHsungRTLamWY. Artificial intelligence in detecting periodontal disease from intraoral photographs: a systematic review. Int Dent J. (2025) 75(5):100883. 10.1016/j.identj.2025.10088340639137TariqANakhiFBSalahFEltayebGAbdullaGJNajimNEfficiency and accuracy of artificial intelligence in the radiographic detection of periodontal bone loss: a systematic review. Imaging Sci Dent. (2023) 53(3):193–98. 10.5624/isd.2023009237799746XueTChenLSunQ. Deep learning method to automatically diagnose periodontal bone loss and periodontitis stage in dental panoramic radiograph. J Dent. (2024) 150:105373. 10.1016/j.jdent.2024.10537339332519LiuJLiuXShaoYGaoYPanKJinCPeriapical lesion detection in periapical radiographs using the latest convolutional neural network ConvNeXt and its integrated models. Sci Rep. (2024) 14(1):25429. 10.1038/s41598-024-75748-939455655IacobAMCastrillón FernándezMFernández RobledoLBarbeito CastroEEscobedo MartínezMF. Automated detection of periodontal bone loss in two-dimensional (2D) radiographs using artificial intelligence: a systematic review. Dent J (Basel). (2025) 13(9):413. 10.3390/dj1309041341002686JundaengJChamchongRNithikathkulC. Periodontitis diagnosis: a review of current and future trends in artificial intelligence. Technol Health Care. (2025) 33(1):473–84. 10.3233/thc-24116939302402ChatzopoulosGSKoidouVPTsalikisLKaklamanosEG. Artificial intelligence for detection and classification of furcation defects using radiographic imaging: a systematic review. Imaging Sci Dent. (2025) 55:322–44. 10.5624/isd.2025010141536891ChauRCWLiGHTewIMThuKMMcGrathCLoWLAccuracy of artificial intelligence-based photographic detection of gingivitis. Int Dent J. (2023) 73(5):724–30. 10.1016/j.identj.2023.03.00737117096ChauRCWChengACCMaoKThuKMLingZTewIMExternal validation of an AI mHealth tool for gingivitis detection among older adults at daycare centers: a pilot study. Int Dent J. (2025) 75(3):1970–78. 10.1016/j.identj.2025.01.00839864975SarakbiRMVarmaSRMuthiah AnnammaLSivaswamyV. Implications of artificial intelligence in periodontal treatment maintenance: a scoping review. Front Oral Health. (2025) 6:1561128. 10.3389/froh.2025.156112840438083CollinsGSReitsmaJBAltmanDGMoonsKG. Transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD): the TRIPOD statement. Br Med J. (2015) 350:g7594. 10.1136/bmj.g7594SounderajahVAshrafianHAggarwalRDe FauwJDennistonAKGreavesFDeveloping specific reporting guidelines for diagnostic accuracy studies assessing AI interventions: the STARD-AI steering group. Nat Med. (2020) 26(6):807–08. 10.1038/s41591-020-0941-132514173KellyCJKarthikesalingamASuleymanMCorradoGKingD. Key challenges for delivering clinical impact with artificial intelligence. BMC Med. (2019) 17(1):195. 10.1186/s12916-019-1426-231665002SounderajahVAshrafianHGolubRMShettySDe FauwJHooftLDeveloping a reporting guideline for artificial intelligence-centred diagnostic test accuracy studies: the STARD-AI protocol. BMJ Open. (2021) 11(6):e047709. 10.1136/bmjopen-2020-04770934183345SounderajahVGuniALiuXCollinsGSKarthikesalingamAMarkarSRThe STARD-AI reporting guideline for diagnostic accuracy studies using artificial intelligence. Nat Med. (2025) 31:3283–9. 10.1038/s41591-025-03953-840954311RonnebergerOFischerPBroxT. U-Net: convolutional networks for biomedical image segmentation (2015) 234–41 p.Baena-de la IglesiaTNavarro-FraileEIglesias-LinaresA. Validation of an AI-aided 3D method for enhanced volumetric quantification of external root resorption in orthodontics. Angle Orthod. (2025) 95(5):474–82. 10.2319/092324-781.140936626da Andrade-BortolettoMFSJindanilTFonteneleRCJacobsRFreitasDQ. Comparison of AI-powered tools for CBCT-based mandibular incisive canal segmentation: a validation study. Clin Oral Implants Res. (2025) 36(9):1086–94. 10.1111/clr.1445540481776KotWYAu YeungSYLeungYYLeungPHYangWF. Evolution of deep learning tooth segmentation from CT/CBCT images: a systematic review and meta-analysis. BMC Oral Health. (2025) 25(1):800. 10.1186/s12903-025-05984-640420051MureșanuSAlmășanOHedeșiuMDioșanLDinuCJacobsR. Artificial intelligence models for clinical usage in dentistry with a focus on dentomaxillofacial CBCT: a systematic review. Oral Radiol. (2023) 39(1):18–40. 10.1007/s11282-022-00660-9PishghadamNEsmaeilyfardRPaknahadM. Explainable deep learning for age and gender estimation in dental CBCT scans using attention mechanisms and multi task learning. Sci Rep. (2025) 15(1):18070. 10.1038/s41598-025-03305-z40413203JundaengJChamchongRNithikathkulC. Artificial intelligence-powered innovations in periodontal diagnosis: a new era in dental healthcare. Front Med Technol. (2024) 6:1469852. 10.3389/fmedt.2024.146985239866670TuygunovNSamaranayakeLKhurshidZRewthamrongsrisPSchwendickeFOsathanonTThe transformative role of artificial intelligence in dentistry: a comprehensive overview part 2: the promise and perils, and the international dental federation communique. Int Dent J. (2025) 75(2):397–404. 10.1016/j.identj.2025.02.00640011130MeiLDengKCuiZFangYLiYLaiHClinical knowledge-guided hybrid classification network for automatic periodontal disease diagnosis in x-ray image. Med Image Anal. (2025) 99:103376. 10.1016/j.media.2024.10337639536402YadalamPKPawarARNatarajanPMArdilaCM. A novel dual embedding few-shot learning approach for classifying bone loss using orthopantomogram radiographic notes. Head Face Med. (2025) 21(1):49. 10.1186/s13005-025-00528-340640850KabirTLeeC-TNelsonJShengSMengH-WChenLAn End-to-end entangled segmentation and classification convolutional neural network for periodontitis stage grading from periapical radiographic images. arXiv. (2021). 10.48550/arXiv.2109.13120WidyaningrumRAstutiERSoetojoAFaadiyaANNurrachmanASKinanggitNDHybrid two-stage CNN for detection and staging of periodontitis on panoramic radiographs. J Oral Biol Craniofac Res. (2025) 15(6):1392–99. 10.1016/j.jobcr.2025.08.01940927498ChenIHLinCHLeeMKChenTELanTHChangCMConvolutional-neural-network-based radiographs evaluation assisting in early diagnosis of the periodontal bone loss via periapical radiograph. J Dent Sci. (2024) 19(1):550–59. 10.1016/j.jds.2023.09.03238303886KroisJGarcia CantuAChaurasiaAPatilRChaudhariPKGaudinRGeneralizability of deep learning models for dental image analysis. Sci Rep. (2021) 11(1):6102. 10.1038/s41598-021-85454-533731732LeeSWHuzKGorelickKLiJBinaTMatsumuraSEvaluation by dental professionals of an artificial intelligence-based application to measure alveolar bone loss. BMC Oral Health. (2025) 25(1):329. 10.1186/s12903-025-05677-040025477HaylazEGumussoyIDumanSBKalabalikFErenMCDemirsoyMSAutomatic segmentation of the nasolacrimal canal: application of the nnU-Net v2 model in CBCT imaging. J Clin Med. (2025) 14(3):778. 10.3390/jcm1403077839941449ShettySTalaatWAlKawasSAl-RawiNReddySHamdoonZApplication of artificial intelligence-based detection of furcation involvement in mandibular first molar using cone beam tomography images- a preliminary study. BMC Oral Health. (2024) 24(1):1476. 10.1186/s12903-024-05268-539633335LiuJHaoJLinHPanWYangJFengYDeep learning-enabled 3D multimodal fusion of cone-beam CT and intraoral mesh scans for clinically applicable tooth-bone reconstruction. Patterns (N Y). (2023) 4(9):100825. 10.1016/j.patter.2023.10082537720330ZhouYXuYKhalilBNalleyATarceM. An open deep learning-based framework and model for tooth instance segmentation in dental CBCT. Clin Oral Investig. (2025) 29(10):473. 10.1007/s00784-025-06578-w40996470ChenZChenSHuF. CTA-UNet: CNN-transformer architecture UNet for dental CBCT images segmentation. Phys Med Biol. (2023) 68(17):175042. 10.1088/1361-6560/acf026ZhaoYZhuLWangWLvLLiQLiuYProgressive multi-task learning for fine-grained dental implant classification and segmentation in CBCT image. Comput Biol Med. (2025) 189:109896. 10.1016/j.compbiomed.2025.10989640073494PalkovicsDMolnarBPinterCGarcía-MatoDDiaz-PintoAWindischPAutomatic deep learning segmentation of mandibular periodontal bone topography on cone-beam computed tomography images. J Dent. (2025) 159:105813. 10.1016/j.jdent.2025.10581340373868PanXWangCLuoXDongQSunHZhangWDevelopment and verification of a convolutional neural network-based model for automatic mandibular canal localization on multicenter CBCT images. BMC Oral Health. (2025) 25(1):1352. 10.1186/s12903-025-06724-640841637FelschMMeyerOSchlickenriederAEngelsPSchönewolfJZöllnerFDetection and localization of caries and hypomineralization on dental photographs with a vision transformer model. NPJ Digit Med. (2023) 6(1):198. 10.1038/s41746-023-00944-237880375TaoLRLiYWuXYGuYXieYYuXYDeep learning photo processing for periodontitis screening. J Dent Res. (2025):220345251347508. 10.1177/0022034525134750840650464WenCBaiXYangJLiSWangXYangD. Deep learning based approach: automated gingival inflammation grading model using gingival removal strategy. Sci Rep. (2024) 14(1):19780. 10.1038/s41598-024-70311-y39187553RezallahNNSherifGAbdelkarimAZAfifiS. Enhancing periodontal bone loss diagnosis through advanced AI techniques. Appl Sci. (2025) 15(12):6832. 10.3390/app15126832NaufalMFatichahCAstutiEPutraR. YOLOv8-Based segmentation and 3D reconstruction of alveolar bone and mandibular canal in CBCT images. (2024) 425–30 p.RashidMOGaghorS. Clinical application of a deep learning system for automatic mandibular alveolar bone quantity assessment and suggested treatment options using CBCT cross-sections. Medicine (Baltimore). (2025) 104(30):e43257. 10.1097/md.000000000004325740725950WidiasriMSuciatiNFatichahCAstutiEPutraR. Alveolar bone and mandibular canal segmentation on cone beam computed tomography images using U-net. (2023) 36–41 p.ZhangXGuoELiuXZhaoHYangJLiWEnhancing furcation involvement classification on panoramic radiographs with vision transformers. BMC Oral Health. (2025) 25(1):153. 10.1186/s12903-025-05431-639881302HadzicAUrschlerMPressJARiedlRRuganiPŠternDEvaluating a periapical lesion detection CNN on a clinically representative CBCT dataset-A validation study. J Clin Med. (2023) 13(1):197. 10.3390/jcm1301019738202204LiuJJinCWangXPanKLiZYiXA comparative analysis of deep learning models for assisting in the diagnosis of periapical lesions in periapical radiographs. BMC Oral Health. (2025) 25(1):801. 10.1186/s12903-025-06104-040420083ChangHJLeeSJYongTHShinNYJangBGKimJEDeep learning hybrid method to automatically diagnose periodontal bone loss and stage periodontitis. Sci Rep. (2020) 10(1):7531. 10.1038/s41598-020-64509-z32372049Muhammed SunnetciKUlukayaSAlkanA. Periodontal bone loss detection based on hybrid deep learning and machine learning models with a user-friendly application. Biomed Signal Process Control. (2022) 77:103844. 10.1016/j.bspc.2022.103844DujicHMeyerOHossPWölfleUCWülkAMeusburgerTAutomatized detection of periodontal bone loss on periapical radiographs by vision transformer networks. Diagnostics (Basel). (2023) 13(23):3562. 10.3390/diagnostics1323356238066803Coşgun BaybarsSDaldalMParlak BaydoğanMArslan TuncerS. Evaluation of apical closure in panoramic radiographs using vision transformer architectures ViT-based apical closure classification. Diagnostics (Basel). (2025) 15(18):2350. 10.3390/diagnostics15182350GaoSLiXLiXLiZDengY. Transformer based tooth classification from cone-beam computed tomography for dental charting. Comput Biol Med. (2022) 148:105880. 10.1016/j.compbiomed.2022.10588035914362KüçükDBImakAÖzçelikSTAÇelebiATürkoğluMSengurAHybrid CNN-transformer model for accurate impacted tooth detection in panoramic radiographs. Diagnostics (Basel). (2025) 15(3):244. 10.3390/diagnostics15030244Di MartinoFDelmastroF. Explainable AI for clinical and remote health applications: a survey on tabular and time series data. Artif Intell Rev. (2023) 56(6):5261–315. 10.1007/s10462-022-10304-336320613SalviMSeoniSCampagnerAGertychAAcharyaURMolinariFExplainability and uncertainty: two sides of the same coin for enhancing the interpretability of deep learning models in healthcare. Int J Med Inform. (2025) 197:105846. 10.1016/j.ijmedinf.2025.10584639993336SinghYHathawayQAKeishingVSalehiSWeiYHorvatNBeyond post hoc explanations: a comprehensive framework for accountable AI in medical imaging through transparency, interpretability, and explainability. Bioengineering (Basel). (2025) 12(8):879. 10.3390/bioengineering1208087940868392UribeSEHamdanMHValenteNAYamaguchiSUmerFTichyAEvaluating dental AI research papers: key considerations for editors and reviewers. J Dent. (2025) 160:105867. 10.1016/j.jdent.2025.10586740451605SitarasSTsolakisIAGelsiniMTsolakisAISchwendickeFWolfTGApplications of artificial intelligence in dental medicine: a critical review. Int Dent J. (2025) 75(2):474–86. 10.1016/j.identj.2024.11.00939843259SchneiderLKrasowskiAPitchikaVBombeckLSchwendickeFBüttnerM. Assessment of CNNs, transformers, and hybrid architectures in dental image segmentation. J Dent. (2025) 156:105668. 10.1016/j.jdent.2025.10566840064460
Edited by: Camila Tirapelli, University of São Paulo, Ribeirão Preto, Brazil
Reviewed by: Walter Y. H. Lam, The University of Hong Kong, Hong Kong SAR, China
Alessio Rosa, University of Rome Tor Vergata, Italy