A Clinically Interpretable Machine Learning Model for Staging and Grading of Periodontitis: Development and Temporal External Validation as a Decision Support Tool

A Clinically Interpretable Machine Learning Model for Staging and Grading of Periodontitis: Development and Temporal External Validation as a Decision Support Tool | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article A Clinically Interpretable Machine Learning Model for Staging and Grading of Periodontitis: Development and Temporal External Validation as a Decision Support Tool Zuke Ya, Yaling Li, Feifei Mo, Zuwei Nong, Sheng Ni, Zuqiang Meng, and 1 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8504183/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 11 You are reading this latest preprint version Abstract Background The 2018 international classification for periodontitis enables individualized patient management through simultaneous staging (disease severity and complexity) and grading (progression rate). However, its clinical adoption is hindered by diagnostic inconsistency—particularly in grading and differentiation of advanced stages—due to the complexity of age-adjusted metrics such as radiographic bone loss–to–age ratio (RBL/age). While artificial intelligence (AI) shows promise in dental diagnostics, existing tools lack robust integration of multimodal clinical data and validation under real-world conditions. Methods We retrospectively collected data from 692 patients diagnosed with periodontitis at Hospital of Stomatology, Guangxi Medical University between June 2022 and June 2024. After data cleaning and feature selection, cases were labeled according to the 2018 international classification criteria. Machine learning models—including k-nearest neighbors (KNN), Random Forest (RF), and Decision Tree (DT)—were trained and optimized via hyperparameter tuning. Model performance was evaluated on internal test sets and further validated on a temporally held-out external cohort (n = 208). Results For staging, KNN, RF, and DT achieved 96.99% accuracy; DT showed the highest recall (94.74%) and F1-score (0.9231) for Stage II, while RF excelled in Stage III (recall and F1-score: 0.9747). For grading, KNN and RF reached 96.40% accuracy, with F1-scores > 0.97 for Grade C. Feature importance analysis identified clinical attachment loss (CAL) and probing depth (PD) as top predictors for staging, and RBL/age as the dominant feature for grading. In temporal external validation, the model achieved 93.75% accuracy for full diagnosis (extent + stage + grade). Conclusions The proposed model demonstrates high accuracy, interpretability, and generalizability, offering a promising decision-support tool for standardized periodontitis classification. Periodontitis Machine learning Staging Grading 2018 international classification Clinical decision support Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Background Periodontitis is a chronic inflammatory disease initiated by dental plaque biofilm, leading to host-mediated destruction of periodontal supporting tissues—including gingiva, periodontal ligament, alveolar bone, and cementum( 1 – 5 ). As one of the most prevalent chronic conditions globally, it progresses insidiously, often resulting in irreversible tooth loss, impaired masticatory function, reduced oral health–related quality of life, and substantial socioeconomic burdens ( 6 – 10 ). The Global Burden of Disease Study 2019 estimated that severe periodontitis affects over 1 billion people worldwide (age-standardized prevalence: 12.50%), with projections exceeding 1.5 billion by 2050—particularly in South Asia ( 11 ). In 2019 alone, periodontitis-related direct healthcare costs and indirect productivity losses totaled USD 186 billion and USD 142 billion, respectively ( 10 ), underscoring its public health urgency. To enable precision management, the 2018 international classification introduced a staging and grading system that replaced the outdated chronic/aggressive dichotomy( 12 ). Staging reflects current disease severity based on clinical attachment loss (CAL), radiographic bone loss (RBL), and tooth loss due to periodontitis (TLP), while grading estimates progression rate—primarily via the radiographic bone loss–to–age ratio (RBL/age)—and integrates risk modifiers such as smoking and systemic conditions( 12 , 13 ). Despite its clinical utility, real-world implementation remains challenging: diagnostic consistency is suboptimal, especially among non-specialists, with poor inter-rater agreement in grading (κ < 0.4) and frequent misclassification between Stage III and IV due to the complexity of age-adjusted metrics ( 14 – 16 ). Although decision-support tools—including rule-based algorithms ( 17 ) and dedicated software applications( 18 )—have improved diagnostic concordance, they often lack automation, multimodal data integration, and robust generalizability. Meanwhile, artificial intelligence (AI), particularly machine learning (ML), has shown promise in dentistry for tasks such as oral cancer detection ( 19 , 20 ), caries prognosis ( 21 ), and periapical diagnosis ( 22 ). Preliminary studies in periodontics suggest AI can enhance diagnostic consistency; however, large language models like ChatGPT exhibit only moderate staging agreement (κ = 0.447) and poor grading performance (κ = 0.284) ( 23 ), revealing limitations in capturing the nuanced logic of the 2018 international classification framework. Thus, there remains an unmet need for an intelligent, data-driven model that automatically integrates heterogeneous clinical and radiographic features, learns complex diagnostic rules, and demonstrates high generalizability. This study aims to develop and validate a ML system for joint staging and grading of periodontitis using real-world data. By evaluating multiple algorithms—including Random Forest (RF) and Support Vector Machines (SVM)—we seek not only to improve diagnostic accuracy but also to establish an interpretable, deployable framework to advance AI-enabled precision periodontics and support clinical education. Materials and methods This study adheres to the Transparent Reporting of a multivariable prediction model for Individual Prognosis Or Diagnosis (TRIPOD) guidelines and TRIPOD + AI statement( 24 – 26 ). Study Design and Data Source We conducted a single-center, retrospective observational cohort study to develop and validate machine learning models for automated staging and grading of periodontitis. Data were retrospectively collected from adult patients (aged 18–80 years) diagnosed with periodontitis at the Department of Periodontic and Oral Medicine, Hospital of Stomatology, Guangxi Medical University, between June 2022 and June 2024. Data were extracted from multiple institutional systems: Electronic Medical Records (EMR): demographic information, medical history, smoking status, dental examination findings. Picture Archiving and Communication System (PACS): full-mouth periapical radiographs, panoramic radiographs, and cone-beam computed tomography (CBCT) images. Laboratory Information System (LIS): biochemical markers including glycated hemoglobin (HbA1c) and high-sensitivity C-reactive protein (hsCRP). A total of 716 patients were initially identified. Exclusion criteria included: ( 1 ) incomplete clinical or imaging records; ( 2 ) non-periodontal causes of clinical attachment loss (e.g., traumatic recession, root fractures); ( 3 ) systemic conditions causing widespread periodontal destruction (e.g., uncontrolled leukemia); and ( 4 ) prior full-mouth periodontal surgery or orthodontic treatment affecting diagnostic assessment. After exclusions, 692 patients were included in the final analysis. Sample size estimation was based on the events per variable (EPV) criterion, with a minimum EPV ≥ 10( 27 ). Given an estimated prevalence of Stage III/IV periodontitis of 37.4% ( 28 )and approximately 15 candidate predictors, a minimum sample size of 395 was required. The final sample size (n = 692) satisfies this criterion and allows robust model development and validation. Data Extraction and Variable Definition Guided by the 2018 international classification of periodontal and peri-implant diseases and conditions, along with supporting evidence from the literature and considerations of clinical feasibility, we pre-specified a set of predictor variables that could be reliably extracted from structured fields in EMRs. The final feature set included: Disease extent: percentage of teeth affected; Periodontal parameters: PD and CAL; Radiographic progression metric: RBL/age; Structural and functional indicators: furcation involvement (FI), number of TLP, number of remaining functional tooth pairs, occlusal discrepancies (including bite collapse, tooth drifting, flaring, and masticatory dysfunction), and severe alveolar bone loss; Systemic modifier: smoking status, and HbA1c. All variables were extracted from standardized clinical examination notes and radiographic reports using institution-defined operational criteria. Missing or ambiguous entries were adjudicated by two calibrated periodontists, with discrepancies resolved through consensus. This approach ensured consistent, reproducible, and clinically meaningful data representation suitable for ML modeling. Diagnostic Criteria for Periodontitis Staging and Grading Diagnoses were assigned according to the 2018 international Classification of Periodontal and Peri-Implant Diseases and Conditions( 12 ). The detailed staging and grading diagnostic criteria are provided in Tables 1 and 2 . Staging (I–IV): Based on severity (CAL, RBL, TLP) and complexity (PD, FI, occlusal dysfunction). Grading (A–C): Assessed disease progression rate using: Direct evidence: Longitudinal changes in CAL or bone loss over 5 years. Indirect evidence: RBL/age. Phenotypic mismatch: Disproportionate tissue destruction relative to plaque levels. Modifying factors: Smoking (cigarettes/day), HbA1c, and hsCRP. Staging and grading were treated as separate multi-class classification outcomes, and two independent models were developed. Table 1 Staging Criteria for Periodontitis According to the 2018 International Classification of Periodontal and Peri-implant Diseases and Conditions Periodontitis stage Stage Ⅰ Stage Ⅱ Stage Ⅲ Stage Ⅳ Severity Interdental CAL at site of greatest loss 1–2 mm 3–4 mm ≥ 5 mm ≥ 5 mm Radiographic bone loss Coronal third 1/3(<15%) Coronal third 1/3(15%-33%) Extending to mid-third of root and beyond Extending to mid-third of root and beyond Tooth loss due to periodontitis No No ≤ 4 teeth ≥ 5 teeth Complexity Local Maximum probing depth ≤ 4mm Mostly horizontal bone loss Maximum probing depth ≤ 5mm Mostly horizontal bone loss In addition to stage II complexity: Probing depth ≥ 6mm Vertical bone loss ≥ 3mm Furcation involvement Class II or III Moderate ridge defect In addition to stage III complexity: Need for complex rehabilitation due to: Masticatory dysfunction Secondary occlusal trauma (tooth mobility degree ≥ 2) Severe ridge defect Bite collapse, drifting, flaring Less than20 remaining teeth (10 opposing pairs) Extent and distribution For each stage, describe extent as localized (< 30% of teeth involved), generalized (≥ 30% of teeth involved), or molar/incisor pattern CAL = clinical attachment loss; RBL = radiographic bone loss. Table 2 Grading Criteria for Periodontitis According to the 2018 International Classification of Periodontal and Peri-implant Diseases and Conditions Periodontitis grade A: Slow rate of progression B: Moderate rate of progression C: Rapid rate of progression Primary criteria Direct evidence of progression Longitudinal data (radiographic bone loss or CAL) Evidence of no loss over 5 years < 2mm over 5 years ≥ 2mm over 5 years Indirect evidence of progression % bone loss/age (RBL/age) 1.0 Case phenotype Heavy biofilm deposits with low levels of destruction Destruction commensurate with biofilm deposits Destruction exceeds expectation given biofilm deposits; specific clinical patterns suggestive of periods of rapid progression and/or early onset disease (e.g., molar/incisor pattern; lack of expected response to standard bacterial control therapies) Grade modifiers Risk factors Smoking Non-smoker Smoker <10 cigarettes/day Smoker ≥ 10 cigarettes/day Diabetes Normoglycemic/ no diagnosis of diabetes HbA1c<7.0% in patients with diabetes HbA1c ≥ 7.0% in patients with diabetes Risk of systemic impact of periodontitis Inflammatory burden High sensitivity CRP (hsCRP) 3 mg /L Biomarkers Indicators of CAL/bone loss Saliva, gingival crevicular fluid, serum ? ? ? HbA1c, glycated hemoglobin; hsCRP, high sensitivity C-reactive protein; PA, periapical; CAL, clinical attachment loss Diagnostic Label Reliability Assessment To ensure diagnostic accuracy and consistency, both inter-examiner and intra-examiner calibration were performed prior to data labeling. For inter-examiner reliability, 30 randomly selected cases were independently diagnosed by two experienced periodontists according to the 2018 international classification criteria. Inter-rater agreement was assessed using Cohen’s kappa coefficient, yielding substantial agreement (κ = 0.78). For intra-examiner reliability, an additional set of 30 cases was re-evaluated by one of the examiners after a 1-week interval, blinded to the initial assessments. The intra-rater agreement was also substantial (κ = 0.81). Based on these results, this calibrated examiner subsequently performed the definitive diagnosis for all study cases to ensure diagnostic consistency and data reliability. Data Preprocessing and statistical analyses Data preprocessing was performed using Python 3.10. The following libraries were used: pandas (2.0.3), NumPy (1.24.3), and scikit-learn (1.3.0). Data Preprocessing We applied a standardized preprocessing pipeline: ( 1 ) categorical variables were one-hot encoded; ( 2 ) missing values in inferable fields (e.g., smoking status) were resolved by consensus between two calibrated clinicians, while records with irrecoverable missingness in core variables were excluded; ( 3 ) data types were harmonized, and descriptive terms (e.g., “severe”) were mapped to predefined numeric codes; ( 4 ) key quantitative features (e.g., radiographic bone loss) were extracted from free-text clinical notes using rule-based parsing; ( 5 ) implausible outliers (e.g., PD > 15 mm) and invalid records were removed; and ( 6 ) predictors were pre-specified based on the 2018 international classification and supporting literature—ensuring clinical relevance and interpretability without reliance on data-driven feature selection. Feature Selection Strategy Rather than employing data-driven feature selection techniques (e.g., LASSO regression or recursive feature elimination), we adopted a clinically informed, hypothesis-driven approach to predictor selection. All candidate variables were pre-specified based on the 2018 international classification for periodontal and peri-implant diseases and conditions ( 3 , 12 )and supported by existing peer-reviewed literature on their association with disease severity, complexity, and progression rate. The final predictor set includes: extent and distribution of periodontitis (percentage of teeth affected), PD, CAL, RBL/age, FI, TLP, number of remaining functional tooth pairs, occlusal discrepancies (including bite collapse, tooth drifting, flaring, and masticatory dysfunction), presence of severe alveolar bone loss, smoking status, and HbA1c. These variables are routinely documented in standard periodontal examinations and can be reliably extracted from structured fields in EMRs. This a priori selection strategy ensures that the model is grounded in current diagnostic standards, enhances clinical interpretability, and avoids overfitting that may arise from exploratory feature screening in moderate-sized datasets. Dataset Splitting The cohort was partitioned into training (70%), validation (20%), and test (10%) sets using stratified random sampling to preserve the distribution of periodontitis stages across all subsets. Model Development and Validation Strategy Model Selection and Training Given the categorical nature of periodontitis staging and grading, we evaluated multiple established machine learning algorithms for classification, including Logistic Regression, k-nearest neighbors (KNN), Decision Tree (DT), Random Forest (RF), and Support Vector Machine (SVM). Each model was trained on the training set to learn the relationship between clinical predictors and diagnostic outcomes. Hyperparameters were optimized using stratified k-fold cross-validation (k = 5) on the training data, with performance assessed by area under the receiver operating characteristic curve (AUC) and balanced accuracy. This approach mitigated overfitting and enhanced generalizability by systematically identifying the configuration that maximized predictive performance while maintaining model stability. Model Evaluation and Validation Model performance was assessed on both the validation and held-out test sets using a comprehensive set of classification metrics, including accuracy, precision, recall (sensitivity), F1-score, and AUC. A normalized confusion matrix was generated to visualize class-wise prediction performance and identify potential misclassification patterns or systematic biases across periodontitis stages. Additionally, we conducted feature importance analysis—using mean decrease in impurity (for tree-based models) or permutation importance—to quantify the contribution of each predictor to model decisions, thereby enhancing interpretability and informing future clinical refinement. To mitigate the risk of overoptimistic performance estimates associated with random data splitting, we adopted a temporal validation approach: the model was trained on patients diagnosed between 2022 and 2024, and validated on a temporally distinct cohort diagnosed in 2025 from the same institution. This design simulates real-world deployment scenarios in which a model developed on historical data is applied to future patients, thereby providing a more realistic assessment of generalizability than cross-sectional random splits. All process was shown in Flowchart1. Although this strategy does not constitute multi-center external validation (which would assess transportability across different care settings), it represents a rigorous form of internal-external validation that accounts for temporal drift in clinical practice, documentation patterns, and patient demographics—key sources of model degradation in dynamic healthcare environments. Flowchart 1 Flowchart of the entire process Results Characteristics This study included EMRs from 692 patients. For each case, we collected demographic and clinical data—including sex, age, and smoking status—as well as a full set of periodontal clinical parameters: percentage of teeth affected, PD, CAL, RBL, presence and severity of FI, TLP, number of remaining functional tooth pairs, occlusal discrepancies (including bite collapse, tooth drifting, flaring, and masticatory dysfunction), and evidence of severe alveolar bone loss. HbA1c levels were also recorded. All cases were systematically staged and graded in accordance with the 2018 international classification of periodontal and peri-Implant diseases and conditions ( 12 ). A summary of the cohort characteristics is presented in Table 3 . Table 3 Clinical Characteristics and Staging/Grading Diagnoses of the 692 Included Case Records Clinical Characteristics n % M ± SD Sex Male 365 52.75 - Female 327 47.25 - Age - - - 47.62 ± 13.22 Maximum probing depth - - - 7.35 ± 2.14 Interdental CAL at site of greatest loss - - - 7.04 ± 3.35 % bone loss/age (RBL/age) 1.0 483 69.80 - Extent and distribution ≥ 30% 464 67.05 - <30% 228 32.95 - Furcation involvement(FI) 0 224 32.37 - 1 13 1.88 - 2 141 20.38 - 3 314 45.37 - Bite collapse No 653 94.36 - Yes 39 5.64 - Drifting, flaring No 578 83.53 - Yes 114 16.47 - Masticatory dysfunction No 549 79.34 - Yes 143 20.66 - Severe ridge defect No 541 78.18 - Yes 151 21.82 - Cigarettes per day Non-smoker 567 81.94 - Smoker <10 cigarettes/day 45 6.50 - Smoker ≥ 10 cigarettes/day 80 11.56 - HbA1c ≥ 7% 25 3.61 - <5.7%(normal) 627 90.61 - 5.7%-7% 40 5.78 - Table 3 (continue) Clinical Characteristics n % M ± SD Diagnosis Extent and distribution localized 228 32.95 - generalized 464 67.05 - Staging Stage Ⅰ 59 8.53 - Stage Ⅱ 93 13.44 - Stage Ⅲ 382 55.20 - Stage Ⅳ 158 22.93 - Grading Grade A 38 5.49 - Grade B 171 24.71 - Grade C 483 69.80 - Total 692 100 - Model comparison In the periodontitis staging task, KNN, RF, and DT achieved comparable top performance, each attaining an overall accuracy of 96.99%, with F1-scores ≥ 0.89 across all stages. Notably, the DT demonstrated superior sensitivity for Stage II, achieving the highest recall (94.74%) and F1-score (0.9231), suggesting enhanced capability in identifying this less prevalent class. In contrast, RF exhibited the greatest stability for Stage III—the most frequent stage—with both precision and recall reaching 0.9747. Logistic Regression performed least favorably (accuracy: 93.23%), particularly underperforming on Stage II (precision: 0.7391), indicating limited discriminative power for intermediate disease severity (Table 4 ). For the periodontitis grading task, KNN and RF again emerged as the best-performing models, yielding a joint highest accuracy of 96.40% and consistently high precision, recall, and F1-scores across all grades (A–C). Conversely, Logistic Regression and SVM showed reduced performance (accuracy: 92.81% for both), with notably lower F1-scores for Grade B and Grade C, reflecting diminished ability to distinguish moderate-to-rapid progression phenotypes (Table 5 ). Collectively, ensemble and instance-based methods (RF, KNN) demonstrated robust and balanced performance across both diagnostic dimensions, while linear models (Logistic Regression) and SVM exhibited limitations in handling class heterogeneity and nuanced clinical distinctions. Table 4 Comparison of Models for Periodontitis Staging Task (Stage I-IV) Model Class Precision Recall F1-Score Support Logistic Regression Stage I 1.00 1.00 1.00 10 Stage II 0.7391 0.8947 0.8095 19 Stage III 0.9605 0.9241 0.9419 79 Stage IV 1.00 0.96 0.9796 25 Accuracy 93.23% - - 133 k-nearest neighbors Stage I 1.00 1.00 1.00 10 Stage II 0.8571 0.9474 0.90 19 Stage III 0.9870 0.9620 0.9744 79 Stage IV 1.00 1.00 1.00 25 Accuracy 96.99% - - 133 Random Forest Stage I 1.00 1.00 1.00 10 Stage II 0.8947 0.8947 0.8947 19 Stage III 0.9747 0.9747 0.9747 79 Stage IV 1.00 1.00 1.00 25 Accuracy 96.99% - - 133 Support Vector Machine Stage I 1.00 1.00 1.00 10 Stage II 0.8095 0.8947 0.85 19 Stage III 0.9740 0.9494 0.9615 79 Stage IV 1.00 1.00 1.00 25 Accuracy 95.49% - - 133 Decision Tree Stage I 1.00 1.00 1.00 10 Stage II 0.90 0.9474 0.9231 19 Stage III 0.9747 0.9747 0.9747 79 Stage IV 1.00 0.96 0.9795 25 Accuracy 96.99% - - 133 Table 5 Comparison of Models for Periodontitis Grading Task (Grade A-C) Model Class Precision Recall F1-Score Support Logistic Regression Grade A 0.8333 0.7143 0.7692 7 Grade B 0.9250 0.8605 0.8916 43 Grade C 0.9355 0.9775 0.9560 89 Accuracy 92.81% - - 139 k-nearest neighbors Grade A 1.00 1.00 1.00 7 Grade B 0.9750 0.9070 0.9400 43 Grade C 0.9565 0.9888 0.9724 89 Accuracy 96.40% - - 139 Random Forest Grade A 1.00 1.00 1.00 7 Grade B 0.9750 0.9070 0.9400 43 Grade C 0.9565 0.9888 0.9724 89 Accuracy 96.40% - - 139 Support Vector Machine Grade A 0.6364 1.00 0.7778 7 Grade B 0.9714 0.7907 0.8718 43 Grade C 0.9462 0.9888 0.9670 89 Accuracy 92.81% - - 139 Decision Tree Grade A 0.778 1.00 0.875 7 Grade B 0.9744 0.8837 0.9268 43 Grade C 0.956 0.9775 0.9667 89 Accuracy 94.96% - - 139 The DT model for staging and the RF model for grading both exhibited high predictive performance, as evidenced by stable learning curves, clinically interpretable feature importance, and excellent classification accuracy (Figs. 1 and 2 ). These results indicate that the models are accurate, well-calibrated, and suitable for integration into clinical decision support systems for periodontitis diagnosis. External validation using a temporal split To assess generalizability under real-world conditions, we evaluated the model on an external test set comprising 208 periodontitis cases collected from the same institution during a subsequent time period (January 2025–June 2025), which was not involved in model development or hyperparameter tuning. On this temporally held-out cohort, the model achieved an accuracy of 93.75% for the complete diagnosis—defined as the simultaneous correct prediction of extent, stage, and grade according to the 2018 international classification. Model application This system is an intelligent decision-support tool designed for staging and grading the diagnosis of periodontitis. The front end utilizes the jQuery library from BootCDN to provide a user-friendly interface for interaction, while the back end is built on the Python Flask framework to handle business logic, data storage, and model computation. The front end and back end communicate via HTTP requests, implementing a decoupled architecture that enhances development efficiency and system maintainability. The business workflow is illustrated in Flowchart 2: Flowchart 2 Flowchart of the development of User interface A user-friendly web-based interface was developed to integrate the staging and grading models for clinical use (Fig. 3 ). The interface allows clinicians to input key periodontal parameters (e.g., PD, CAL, TLP, FI, RBL/age) and select from multiple ML algorithms (including RF, DT, KNN, and SVM). Upon submission, the system automatically generates a diagnosis prediction, including the stage and grade of periodontitis, along with model performance metrics (accuracy for staging and grading across different algorithms). This tool facilitates real-time decision support in clinical practice. Discussion Our study demonstrates that machine learning models—specifically KNN, RF, and DT—can achieve high diagnostic accuracy in both staging (96.99%) and grading (96.40%) of periodontitis according to the 2018 international classification. Crucially, this performance generalizes beyond internal validation: when evaluated on an external test set comprising cases from the same institution but collected during a subsequent time period (temporal split), the model maintained a high overall accuracy of 93.75% for full diagnosis (extent + stage + grade). This temporal external validation provides strong evidence that the model is not merely overfitting to cohort-specific patterns but captures stable, clinically relevant disease signatures—enhancing its potential for real-world deployment. The 2018 international classification, while conceptually rigorous, poses practical challenges in routine care, particularly for non-specialists. Prior studies report suboptimal inter-rater agreement, with general dentists showing reduced accuracy—especially in distinguishing Stage III from IV—and greater difficulty diagnosing mild disease ( 14 ). In this context, our results represent a significant advance: the models not only replicate expert-level judgment but do so consistently across all disease stages, as evidenced by high per-class F1-scores (≥ 0.89) and minimal misclassification in normalized confusion matrices. Notably, our approach substantially outperforms both human-assisted and generic AI–based strategies. Marini et al.( 18 )found that even with dedicated software support, general dentists achieved only 53.6% concordance in full diagnosis. More recently, Tastan et al. ( 23 ) reported that ChatGPT—a widely accessible large language model—achieved merely 59.5% accuracy in staging and 50.5% in grading, despite explicit prompting with the 2018 international classification criteria. These findings collectively suggest that domain-specific training on structured clinical data is essential for reliable periodontal classification, and that off-the-shelf AI tools lack the necessary granularity for this task. A key strength of our models lies in their clinical interpretability. Feature importance analysis identified CALand PD as the top predictors for staging—directly aligning with the 2018 international classification framework’s designation of CAL as the primary indicator of disease severity ( 12 ). This biologically grounded feature selection enhances model credibility and facilitates clinician trust. Secondary contributors—including severe alveolar bone loss, tooth displacement, and TLP—reflect established markers of advanced disease ( 29 ), while variables unrelated to core diagnostic criteria (e.g., remaining functional tooth pairs) showed negligible influence, indicating that the models focus on clinically meaningful signals rather than confounding factors. For grading, where longitudinal data on disease progression are rarely available in routine care, our models relied predominantly on the RBL/age—validating its utility as a surrogate for progression rate under real-world constraints. Smoking also emerged as a significant modifier, consistent with its well-documented role in impairing host response. Intriguingly, HbA1c exhibited limited predictive value, possibly due to sample size or overshadowing by stronger predictors. We observed a marked underrepresentation of Grade A cases (5.49%), likely attributable to the stringent RBL/age < 0.25 threshold. Although our model achieved 100% recall for this class internally, the extreme class imbalance (only 7 Grade A samples in the validation set) may limit generalizability to populations with milder disease. This finding resonates with growing concerns that the current grading thresholds may not reflect clinical reality ( 18 , 30 ). While we adhered strictly to the official criteria, our data provide empirical support for proposals to revise the cutoffs (e.g., to < 0.5 for Grade A), which could improve both clinical feasibility and model performance across the full spectrum of disease. Limitations include the single-center design and modest representation of early-stage disease. Nevertheless, the combination of high internal performance, robust temporal external validation (93.75% full-diagnosis accuracy), and interpretable feature reliance positions our models as a clinically actionable tool. Future prospective, multi-center studies should validate these findings in diverse populations and explore adaptive thresholds to further enhance equity and utility in routine periodontal care. A key methodological strength of our study is the theory-guided, guideline-concordant feature selection process. Instead of exhaustively evaluating hundreds of potential predictors—a common practice in machine learning that risks yielding statistically significant but clinically irrelevant features—we deliberately restricted our input variables to those explicitly recommended or implied by the 2018 international classification for periodontitis. This approach aligns our model with established clinical reasoning pathways and ensures that every predictor has a clear pathophysiological or diagnostic rationale. By prioritizing clinical validity over algorithmic novelty, we enhance the transparency, trustworthiness, and real-world applicability of our staging and grading system. Practitioners are more likely to adopt decision-support tools that reflect familiar diagnostic criteria rather than opaque “black-box” outputs derived from data-mining alone. While future studies may explore hybrid approaches combining expert knowledge with data-driven refinement, our current design reflects the pragmatic reality of clinical documentation and supports seamless integration into routine dental workflows—particularly in settings lacking advanced computational or laboratory infrastructure. Several features recommended in the 2018 international classification were not included in our models due to practical and methodological considerations. First, the phenotypic grading criterion based on the disproportionality between plaque accumulation and periodontal destruction—where excessive tissue loss relative to plaque burden suggests Grade C—was excluded due to its reliance on subjective clinical judgment and the absence of standardized, quantifiable thresholds in routine documentation. In machine learning applications that prioritize reproducibility and cross-institutional generalizability, such semi-qualitative judgments risk introducing measurement bias and undermining model robustness. Second, systemic inflammatory biomarkers—such as hsCRP—and host-derived inflammatory mediators, although biologically relevant to periodontitis progression, were not routinely measured in standard clinical practice during the study period. Their low prevalence and high rate of missingness in EMRs would have required either extensive imputation or the exclusion of a large number of cases, thereby reducing the generalizability of the model and introducing potential selection bias. Accordingly, our model was built exclusively on predictors that are routinely documented, objectively measurable, and consistently available in standard periodontal care—namely, PD, CAL, RBL/age, extent and distribution of periodontitis (percentage of teeth affected), FI, number of remaining functional tooth pairs, occlusal discrepancies (including bite collapse, tooth drifting, flaring, and masticatory dysfunction), severe alveolar bone loss, smoking status, and HbA1c. These variables collectively reflect the key domains of the 2018 international classification—disease severity, complexity, and estimated progression rate—and can be reliably extracted from electronic health records without reliance on specialized laboratory assays or subjective phenotypic assessments. This pragmatic feature selection strategy enhances model transparency, reproducibility, and clinical deployability, particularly in real-world dental settings where access to advanced diagnostics is limited. Although this approach may not fully capture host-mediated inflammatory dynamics or rare phenotypic subtypes, it aligns with current clinical workflows and prioritizes operational feasibility over theoretical comprehensiveness. Future prospective studies integrating standardized systemic biomarker panels and natural language processing of unstructured clinical notes could facilitate more nuanced, biologically informed staging and grading in automated decision-support systems. Conclusions In conclusion, the machine learning model developed in this study demonstrates high diagnostic accuracy and stability in the simultaneous staging and grading of periodontitis, achieving 93.75% full-diagnosis accuracy on a temporally held-out external cohort—a finding that underscores its robust generalizability under real-world conditions. Critically, the model’s decision logic aligns with biologically grounded clinical criteria: key predictors such as CAL for staging and RBL/age for grading directly reflect the core principles of the 2018 international classification. This alignment not only mirrors expert clinical reasoning but also enhances diagnostic consistency, particularly in complex cases where inter-rater variability is high. Nevertheless, due to limited Grade A cases and the single-center data source, the model should serve as a decision-support tool rather than a replacement for clinical judgment; comprehensive patient assessment remains essential. Future work should focus on multicenter data collection to enrich underrepresented categories (particularly Grade A), mitigate class imbalance, and validate generalizability across diverse populations—key steps toward realizing AI-enabled precision periodontics in real-world practice. Declarations Ethics approval and consent to participate The study protocol was approved by the Ethics Committee of School & Hospital of Stomatology, Guangxi Medical University (Approval No.: 2024057). All patient data were anonymized prior to analysis, and the requirement for informed consent was waived due to the retrospective nature of the study. Consent for publication Not appicable. Declaration of generative AI and AI-assisted technologies in the writing process Competing interests The authors declare no competing interests. Author details 1 Department of Periodontics and Oral Medicine, College of Stomatology, Hospital of Stomatology, Guangxi Medical University, Nanning, Guangxi 530021, P. R. China 2 Guangxi Health Commission Key Laboratory of Prevention and Treatment for Oral Infectious Diseases, Nanning, Guangxi 530021, P. R. China 3 Guangxi Key Laboratory of Oral and Maxillofacial Rehabilitation and Reconstruction, Guangxi Universities and Colleges Key Laboratory of Oral and Maxillofacial Surgery Disease Treatment, Guangxi Clinical Research Center for Craniofacial Deformity, Nanning, Guangxi 530021, P. R. China 4 Department of Information Technology, College of Stomatology, Hospital of Stomatology, Guangxi Medical University, Nanning, Guangxi 530021, P. R. China 5 College of Computer, Electronics and Information, Guangxi University, Nanning, Guangxi 530004, P.R. China 6 Department of Conservative Dentistry and Endodontics, College of Stomatology, Hospital of Stomatology, Guangxi Medical University, Nanning, Guangxi 530021, P. R. China Funding This work was conducted during the doctoral training of the first author at Guangxi Medical University. No external or internal grant funding was specifically allocated to this project. The research benefited from the clinical data infrastructure and academic supervision provided by Hospital of Stomatology, Guangxi Medical University. Author Contribution Z.Y. designed the study and drafted the manuscript; Y.L. and F.M. extracted and organized the clinical data; Z.N. and S.N. was responsible for data quality control, maintenance of the clinical information system, and statistical analyses; Z.M. provided critical guidance on the design of the machine learning model and the development of the user interface; R.T. supervised the research project and critically revised the manuscript. All authors contributed to the article and approved the submitted version. All authors take responsibility for all aspects of the reliability and freedom from bias of the data presented and their discussed interpretation. Acknowledgement We thank all participants at Hospital of Stomatology, Guangxi Medical University. Data Availability Data sharing is not applicable to this article as no datasets were generated or analyzed during the current study. References Chapple ILC, Mealey BL, Van Dyke TE, Bartold PM, Dommisch H, Eickholz P, et al. Periodontal health and gingival diseases and conditions on an intact and a reduced periodontium: Consensus report of workgroup 1 of the 2017 World Workshop on the Classification of Periodontal and Peri-Implant Diseases and Conditions. J Clin Periodontol. 2018;45(Suppl 20):S68–77. Murakami S, Mealey BL, Mariotti A, Chapple ILC. Dental plaque-induced gingival conditions. J Periodontol. 2018;89(Suppl 1):S17–27. Papapanou PN, Sanz M, Buduneli N, Dietrich T, Feres M, Fine DH, et al. Periodontitis: Consensus report of workgroup 2 of the 2017 World Workshop on the Classification of Periodontal and Peri-Implant Diseases and Conditions. J Periodontol. 2018;89(Suppl 1):S173–82. Kinane DF, Lappin DF, Culshaw S. The role of acquired host immunity in periodontal diseases. Periodontol 2000. 2024;00:1–15. Kinane DF, Bornstein MM. Introduction to the Diagnostics in Periodontology and Implant Dentistry issue. Periodontol 2000. 2024;95(1):7–9. Eke PI, Dye BA, Wei L, Slade GD, Thornton-Evans GO, Borgnakke WS, et al. Update on Prevalence of Periodontitis in Adults in the United States: NHANES 2009 to 2012. J Periodontol. 2015;86(5):611–22. Chen MX, Zhong YJ, Dong QQ, Wong HM, Wen YF. Global, regional, and national burden of severe periodontitis, 1990–2019: An analysis of the Global Burden of Disease Study 2019. J Clin Periodontol. 2021;48(9):1165–88. Jiao J, Jing W, Si Y, Feng X, Tai B, Hu D, et al. The prevalence and severity of periodontal disease in Mainland China: Data from the Fourth National Oral Health Survey (2015–2016). J Clin Periodontol. 2020;48(2):168–79. Tonetti MS, Jepsen S, Jin L, Otomo-Corgel J. Impact of the global burden of periodontal diseases on health, nutrition and wellbeing of mankind: A call for global action. J Clin Periodontol. 2017;44(5):456–62. Pattamatta M, Chapple I, Listl S. The value-for money of preventing and managing periodontitis: Opportunities and challenges. Periodontol 2000. 2024;00:1–9. Nascimento GG, Alves-Costa S, Romandini M. Burden of severe periodontitis and edentulism in 2021, with projections up to 2050: The Global Burden of Disease 2021 study. J Periodontal Res. 2024;59(5):823–67. Tonetti MS, Greenwell H, Kornman KS. Staging and grading of periodontitis: Framework and proposal of a new classification and case definition. J Clin Periodontol. 2018;45(Suppl 20):S149–61. Papapanou PN, Sanz M, Buduneli N, Dietrich T, Feres M, Fine DH, et al. Periodontitis: 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):S162–70. Marini L, Tonetti MS, Nibali L, Rojas MA, Aimetti M, Cairo F, et al. The staging and grading system in defining periodontitis cases: consistency and accuracy amongst periodontal experts, general dentists and undergraduate students. J Clin Periodontol. 2021;48(2):205–15. Abrahamian L, Pascual-LaRocca A, Barallat L, Valles C, Herrera D, Sanz M, et al. Intra- and inter-examiner reliability in classifying periodontitis according to the 2018 classification of periodontal diseases. J Clin Periodontol. 2022;49(8):732–9. Gandhi KK, Katwal D, Chang J, Blanchard S, Shin D, Maupome G, et al. Diagnosis and treatment planning using the 2017 classification of periodontal diseases among three dental schools. J Dent Educ. 2022;86(11):1521–8. Tonetti MS, Sanz M. Implementation of the new classification of periodontal diseases: Decision-making algorithms for clinical practice and education. J Clin Periodontol. 2019;46(4):398–405. Marini L, Tonetti MS, Nibali L, Sforza NM, Landi L, Cavalcanti R, et al. Implementation of a software application in staging and grading of periodontitis cases. Oral Dis. 2024;30(2):719–28. Chen W, Lin G, Chen Y, Cheng F, Li X, Ding J, et al. Prediction of the Ki-67 expression level in head and neck squamous cell carcinoma with machine learning-based multiparametric MRI radiomics: a multicenter study. BMC Cancer. 2024;24(1):418. Shan J, Jiang R, Chen X, Zhong Y, Zhang W, Xie L, et al. Machine Learning Predicts Lymph Node Metastasis in Early-Stage Oral Tongue Squamous Cell Carcinoma. J Oral Maxillofac Surg. 2020;78(12):2208–18. Reyes LT, Knorst JK, Ortiz FR, Ardenghi TM. Machine Learning in the Diagnosis and Prognostic Prediction of Dental Caries: A Systematic Review. Caries Res. 2022;56(3):161–70. Chau KK, Zhu M, AlHadidi A, Wang C, Hung K, Wohlgemuth P, et al. A novel AI model for detecting periapical lesion on CBCT: CBCT-SAM. J Dent. 2025;153:105526. Tastan Eroglu Z, Babayigit O, Ozkan Sen D, Ucan Yarkac F. Performance of ChatGPT in classifying periodontitis according to the 2018 classification of periodontal diseases. Clin Oral Investig. 2024;28(7):407. Collins GS, Dhiman P, Andaur Navarro CL, Ma J, Hooft L, Reitsma JB, et al. Protocol for development of a reporting guideline (TRIPOD-AI) and risk of bias tool (PROBAST-AI) for diagnostic and prognostic prediction model studies based on artificial intelligence. BMJ Open. 2021;11(7):e048008. Yan MH, Zhao YY, Liu X, Li W, Wang Y. [Interpretation of the transparent reporting of a multivariable prediction model for individual prognosis or diagnosis based on regression or machine learning methods (TRIPOD + AI)]. Zhonghua nei ke za zhi. 2025;64(1):4–10. Collins GS, Moons KGM, Dhiman P, Riley RD, Beam AL, Van Calster B, et al. TRIPOD + AI statement: updated guidance for reporting clinical prediction models that use regression or machine learning methods. BMJ. 2024;385:e078378. Peduzzi P, Concato J, Kemper E, Holford TR, Feinstein AR. A simulation study of the number of events per variable in logistic regression analysis. J Clin Epidemiol. 1996;49(12):1373–9. Chen Y, Lu P, Lin C, Li S, Zhu Y, Tan J, et al. Hyperuricemia and elevated uric acid/creatinine ratio are associated with stages III/IV periodontitis: a population-based cross-sectional study (NHANES 2009–2014). BMC Oral Health. 2024;24(1):1389. Zhao L, Xu T, Hu W, Chung KH. Preservation and augmentation of molar extraction sites affected by severe bone defect due to advanced periodontitis: A prospective clinical trial. Clin Implant Dent Relat Res. 2018;20(3):333–44. Dietrich T, Ower P, Tank M, West NX, Walter C, Needleman I, et al. Periodontal diagnosis in the context of the 2017 classification system of periodontal diseases and conditions - implementation in clinical practice. Br Dent J. 2019;226(1):16–22. Additional Declarations No competing interests reported. Cite Share Download PDF Status: Under Review Version 1 posted Reviews received at journal 26 Feb, 2026 Reviews received at journal 19 Feb, 2026 Reviewers agreed at journal 14 Feb, 2026 Reviewers agreed at journal 13 Feb, 2026 Reviews received at journal 07 Feb, 2026 Reviewers agreed at journal 07 Feb, 2026 Reviewers invited by journal 05 Feb, 2026 Editor invited by journal 12 Jan, 2026 Editor assigned by journal 10 Jan, 2026 Submission checks completed at journal 10 Jan, 2026 First submitted to journal 02 Jan, 2026 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-8504183","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":587492011,"identity":"31e5f5fa-a083-4b2f-9c95-a9222320785e","order_by":0,"name":"Zuke Ya","email":"","orcid":"","institution":"Guangxi Medical University","correspondingAuthor":false,"prefix":"","firstName":"Zuke","middleName":"","lastName":"Ya","suffix":""},{"id":587492013,"identity":"91fe549c-4460-4aeb-b72e-a3b3cb201090","order_by":1,"name":"Yaling Li","email":"","orcid":"","institution":"Guangxi Medical University","correspondingAuthor":false,"prefix":"","firstName":"Yaling","middleName":"","lastName":"Li","suffix":""},{"id":587492020,"identity":"bb1a77a0-1529-41f7-86b6-57565e0b1c53","order_by":2,"name":"Feifei Mo","email":"","orcid":"","institution":"Guangxi Medical University","correspondingAuthor":false,"prefix":"","firstName":"Feifei","middleName":"","lastName":"Mo","suffix":""},{"id":587492021,"identity":"ef6f6f48-76d9-47af-b949-f57d5059ac73","order_by":3,"name":"Zuwei Nong","email":"","orcid":"","institution":"Guangxi Medical University","correspondingAuthor":false,"prefix":"","firstName":"Zuwei","middleName":"","lastName":"Nong","suffix":""},{"id":587492022,"identity":"472e60e3-4c85-40b7-a705-ba0897ec0597","order_by":4,"name":"Sheng Ni","email":"","orcid":"","institution":"Guangxi Medical University","correspondingAuthor":false,"prefix":"","firstName":"Sheng","middleName":"","lastName":"Ni","suffix":""},{"id":587492023,"identity":"9a869e79-3589-46a4-a634-e7fdd2eacfec","order_by":5,"name":"Zuqiang Meng","email":"","orcid":"","institution":"Guangxi University","correspondingAuthor":false,"prefix":"","firstName":"Zuqiang","middleName":"","lastName":"Meng","suffix":""},{"id":587492024,"identity":"a451f29d-af78-4378-be2c-6099bd3aaf11","order_by":6,"name":"Renchuan Tao","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA8ElEQVRIiWNgGAWjYBCDBCBmfADlGOBVyoOkhRmmlHgtbBJEabFnP3zsw8cddnkGx88eq/zati2xgb15mwRDzR3ctvCkJc+ceSa52OBMXtptmTO3Ext4jpVJMBx7hsdhOcbMvG0HEjccyDG7LVEB1CKRYybB2HAYtxb+N8bMf0Fazr8xK5YwAGqRf0NAiwTQFkaQlhs5ZowfwLbwENBy41kyY29bcuLMG2+MpRnO3DZu40krtkg4hlsLe3/yYYafbXaJfedzDD/+bLst289+eOONDzW4taAAZlAssYFYCcRpAKaYH8SqHAWjYBSMghEFAI0BV28NEAD2AAAAAElFTkSuQmCC","orcid":"","institution":"Guangxi Medical University","correspondingAuthor":true,"prefix":"","firstName":"Renchuan","middleName":"","lastName":"Tao","suffix":""}],"badges":[],"createdAt":"2026-01-03 03:53:13","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-8504183/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8504183/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":102261049,"identity":"8b591017-f3a4-4baf-8cf5-d3722017e112","added_by":"auto","created_at":"2026-02-10 00:38:05","extension":"jpeg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":316216,"visible":true,"origin":"","legend":"\u003cp\u003ePerformance Analysis of Decision Tree Models for Periodontitis Staging Tasks. \u003cstrong\u003e\u0026nbsp;A:\u003c/strong\u003e Learning Curve. \u003cstrong\u003eB: \u003c/strong\u003eFeature Importance. \u003cstrong\u003eC: \u003c/strong\u003eConfusion Matrix. \u003cstrong\u003eD:\u003c/strong\u003e Precision-Recall (PF) Curve.\u003c/p\u003e","description":"","filename":"floatimage2.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8504183/v1/69326ab0187802e363e18583.jpeg"},{"id":102261046,"identity":"63af8e58-9d77-4a83-afc5-7d78a4da3b91","added_by":"auto","created_at":"2026-02-10 00:38:04","extension":"jpeg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":275517,"visible":true,"origin":"","legend":"\u003cp\u003ePerformance Analysis of Random Forest Models for Periodontitis Grading Tasks. \u003cstrong\u003e\u0026nbsp;A:\u003c/strong\u003e Learning Curve. \u003cstrong\u003eB: \u003c/strong\u003eFeature Importance. \u003cstrong\u003eC: \u003c/strong\u003eConfusion Matrix. \u003cstrong\u003eD:\u003c/strong\u003e Precision-Recall (PF) Curve.\u003c/p\u003e","description":"","filename":"floatimage3.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8504183/v1/8a266a40f02dfefbf9c5c070.jpeg"},{"id":102261047,"identity":"825d6a6e-8ffc-4a94-95f2-a43fd534b7f4","added_by":"auto","created_at":"2026-02-10 00:38:05","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":118139,"visible":true,"origin":"","legend":"\u003cp\u003eUser Interface of the Periodontitis Diagnostic System\u003c/p\u003e","description":"","filename":"floatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-8504183/v1/cd3ad717ff74af2894476a80.png"},{"id":102297999,"identity":"2ccd7ab8-3cb7-49a4-88bd-ee72d88899ef","added_by":"auto","created_at":"2026-02-10 10:30:03","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":45630,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFlowchart 1 \u003c/strong\u003eFlowchart of the entire process\u003c/p\u003e","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-8504183/v1/8d1922c2869cbd4e60447c9e.png"},{"id":102261050,"identity":"806c4b19-fb5a-4519-9f74-7a0e5c3c4e87","added_by":"auto","created_at":"2026-02-10 00:38:05","extension":"jpeg","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":30492,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFlowchart 2 \u003c/strong\u003eFlowchart of the development of User interface\u003c/p\u003e","description":"","filename":"floatimage4.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8504183/v1/575b456ad95796c734e0e4f7.jpeg"},{"id":102299653,"identity":"c66c3a6e-f3c1-44d6-9d87-7a24844b6075","added_by":"auto","created_at":"2026-02-10 11:07:32","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2072795,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8504183/v1/2a84d264-b6bf-48f6-a0e3-8a13a972bb28.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"A Clinically Interpretable Machine Learning Model for Staging and Grading of Periodontitis: Development and Temporal External Validation as a Decision Support Tool","fulltext":[{"header":"Background","content":"\u003cp\u003ePeriodontitis is a chronic inflammatory disease initiated by dental plaque biofilm, leading to host-mediated destruction of periodontal supporting tissues\u0026mdash;including gingiva, periodontal ligament, alveolar bone, and cementum(\u003cspan additionalcitationids=\"CR2 CR3 CR4\" citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e). As one of the most prevalent chronic conditions globally, it progresses insidiously, often resulting in irreversible tooth loss, impaired masticatory function, reduced oral health\u0026ndash;related quality of life, and substantial socioeconomic burdens (\u003cspan additionalcitationids=\"CR7 CR8 CR9\" citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e). The Global Burden of Disease Study 2019 estimated that severe periodontitis affects over 1\u0026nbsp;billion people worldwide (age-standardized prevalence: 12.50%), with projections exceeding 1.5\u0026nbsp;billion by 2050\u0026mdash;particularly in South Asia (\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e). In 2019 alone, periodontitis-related direct healthcare costs and indirect productivity losses totaled USD 186\u0026nbsp;billion and USD 142\u0026nbsp;billion, respectively (\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e), underscoring its public health urgency.\u003c/p\u003e \u003cp\u003eTo enable precision management, the 2018 international classification introduced a staging and grading system that replaced the outdated chronic/aggressive dichotomy(\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e). Staging reflects current disease severity based on clinical attachment loss (CAL), radiographic bone loss (RBL), and tooth loss due to periodontitis (TLP), while grading estimates progression rate\u0026mdash;primarily via the radiographic bone loss\u0026ndash;to\u0026ndash;age ratio (RBL/age)\u0026mdash;and integrates risk modifiers such as smoking and systemic conditions(\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e, \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e). Despite its clinical utility, real-world implementation remains challenging: diagnostic consistency is suboptimal, especially among non-specialists, with poor inter-rater agreement in grading (κ\u0026thinsp;\u0026lt;\u0026thinsp;0.4) and frequent misclassification between Stage III and IV due to the complexity of age-adjusted metrics (\u003cspan additionalcitationids=\"CR15\" citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eAlthough decision-support tools\u0026mdash;including rule-based algorithms (\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e) and dedicated software applications(\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e)\u0026mdash;have improved diagnostic concordance, they often lack automation, multimodal data integration, and robust generalizability. Meanwhile, artificial intelligence (AI), particularly machine learning (ML), has shown promise in dentistry for tasks such as oral cancer detection (\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e, \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e), caries prognosis (\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e), and periapical diagnosis (\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e). Preliminary studies in periodontics suggest AI can enhance diagnostic consistency; however, large language models like ChatGPT exhibit only moderate staging agreement (κ\u0026thinsp;=\u0026thinsp;0.447) and poor grading performance (κ\u0026thinsp;=\u0026thinsp;0.284) (\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e), revealing limitations in capturing the nuanced logic of the 2018 international classification framework.\u003c/p\u003e \u003cp\u003eThus, there remains an unmet need for an intelligent, data-driven model that automatically integrates heterogeneous clinical and radiographic features, learns complex diagnostic rules, and demonstrates high generalizability. This study aims to develop and validate a ML system for joint staging and grading of periodontitis using real-world data. By evaluating multiple algorithms\u0026mdash;including Random Forest (RF) and Support Vector Machines (SVM)\u0026mdash;we seek not only to improve diagnostic accuracy but also to establish an interpretable, deployable framework to advance AI-enabled precision periodontics and support clinical education.\u003c/p\u003e"},{"header":"Materials and methods","content":"\u003cp\u003eThis study adheres to the Transparent Reporting of a multivariable prediction model for Individual Prognosis Or Diagnosis (TRIPOD) guidelines and TRIPOD\u0026thinsp;+\u0026thinsp;AI statement(\u003cspan additionalcitationids=\"CR25\" citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e).\u003c/p\u003e \u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eStudy Design and Data Source\u003c/h2\u003e \u003cp\u003eWe conducted a single-center, retrospective observational cohort study to develop and validate machine learning models for automated staging and grading of periodontitis.\u003c/p\u003e \u003cp\u003eData were retrospectively collected from adult patients (aged 18\u0026ndash;80 years) diagnosed with periodontitis at the Department of Periodontic and Oral Medicine, Hospital of Stomatology, Guangxi Medical University, between June 2022 and June 2024. Data were extracted from multiple institutional systems:\u003c/p\u003e \u003cp\u003e \u003cul\u003e \u003cli\u003e \u003cp\u003eElectronic Medical Records (EMR): demographic information, medical history, smoking status, dental examination findings.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003ePicture Archiving and Communication System (PACS): full-mouth periapical radiographs, panoramic radiographs, and cone-beam computed tomography (CBCT) images.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eLaboratory Information System (LIS): biochemical markers including glycated hemoglobin (HbA1c) and high-sensitivity C-reactive protein (hsCRP).\u003c/p\u003e \u003c/li\u003e \u003c/ul\u003e \u003c/p\u003e \u003cp\u003eA total of 716 patients were initially identified. Exclusion criteria included: (\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e) incomplete clinical or imaging records; (\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e) non-periodontal causes of clinical attachment loss (e.g., traumatic recession, root fractures); (\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e) systemic conditions causing widespread periodontal destruction (e.g., uncontrolled leukemia); and (\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e) prior full-mouth periodontal surgery or orthodontic treatment affecting diagnostic assessment. After exclusions, 692 patients were included in the final analysis.\u003c/p\u003e \u003cp\u003eSample size estimation was based on the events per variable (EPV) criterion, with a minimum EPV\u0026thinsp;\u0026ge;\u0026thinsp;10(\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e). Given an estimated prevalence of Stage III/IV periodontitis of 37.4% (\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e)and approximately 15 candidate predictors, a minimum sample size of 395 was required. The final sample size (n\u0026thinsp;=\u0026thinsp;692) satisfies this criterion and allows robust model development and validation.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eData Extraction and Variable Definition\u003c/h3\u003e\n\u003cp\u003eGuided by the 2018 international classification of periodontal and peri-implant diseases and conditions, along with supporting evidence from the literature and considerations of clinical feasibility, we pre-specified a set of predictor variables that could be reliably extracted from structured fields in EMRs. The final feature set included:\u003c/p\u003e \u003cp\u003e \u003cul\u003e \u003cli\u003e \u003cp\u003eDisease extent: percentage of teeth affected;\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003ePeriodontal parameters: PD and CAL;\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eRadiographic progression metric: RBL/age;\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eStructural and functional indicators: furcation involvement (FI), number of TLP, number of remaining functional tooth pairs, occlusal discrepancies (including bite collapse, tooth drifting, flaring, and masticatory dysfunction), and severe alveolar bone loss;\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eSystemic modifier: smoking status, and HbA1c.\u003c/p\u003e \u003c/li\u003e \u003c/ul\u003e \u003c/p\u003e \u003cp\u003eAll variables were extracted from standardized clinical examination notes and radiographic reports using institution-defined operational criteria. Missing or ambiguous entries were adjudicated by two calibrated periodontists, with discrepancies resolved through consensus. This approach ensured consistent, reproducible, and clinically meaningful data representation suitable for ML modeling.\u003c/p\u003e\n\u003ch3\u003eDiagnostic Criteria for Periodontitis Staging and Grading\u003c/h3\u003e\n\u003cp\u003eDiagnoses were assigned according to the 2018 international Classification of Periodontal and Peri-Implant Diseases and Conditions(\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e). The detailed staging and grading diagnostic criteria are provided in Tables\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e and \u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003cul\u003e \u003cli\u003e \u003cp\u003eStaging (I\u0026ndash;IV): Based on severity (CAL, RBL, TLP) and complexity (PD, FI, occlusal dysfunction).\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eGrading (A\u0026ndash;C): Assessed disease progression rate using:\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eDirect evidence: Longitudinal changes in CAL or bone loss over 5 years.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eIndirect evidence: RBL/age.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003ePhenotypic mismatch: Disproportionate tissue destruction relative to plaque levels.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eModifying factors: Smoking (cigarettes/day), HbA1c, and hsCRP.\u003c/p\u003e \u003c/li\u003e \u003c/ul\u003e \u003c/p\u003e \u003cp\u003eStaging and grading were treated as separate multi-class classification outcomes, and two independent models were developed.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eStaging Criteria for Periodontitis According to the 2018 International Classification of Periodontal and Peri-implant Diseases and Conditions\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"6\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePeriodontitis stage\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eStage Ⅰ\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eStage Ⅱ\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eStage Ⅲ\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eStage Ⅳ\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003eSeverity\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eInterdental CAL at site of greatest loss\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1\u0026ndash;2 mm\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e3\u0026ndash;4 mm\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u0026ge;\u0026thinsp;5 mm\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e\u0026ge;\u0026thinsp;5 mm\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eRadiographic bone loss\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eCoronal third 1/3(\u0026lt;15%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eCoronal third 1/3(15%-33%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eExtending to mid-third of root and beyond\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eExtending to mid-third of root and beyond\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTooth loss due to periodontitis\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eNo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u0026le;\u0026thinsp;4 teeth\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e\u0026ge;\u0026thinsp;5 teeth\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eComplexity\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eLocal\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMaximum probing depth\u0026thinsp;\u0026le;\u0026thinsp;4mm\u003c/p\u003e \u003cp\u003eMostly horizontal bone loss\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eMaximum probing depth\u0026thinsp;\u0026le;\u0026thinsp;5mm\u003c/p\u003e \u003cp\u003eMostly horizontal bone loss\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eIn addition to stage II complexity: Probing depth\u0026thinsp;\u0026ge;\u0026thinsp;6mm Vertical bone loss\u0026thinsp;\u0026ge;\u0026thinsp;3mm Furcation involvement Class II or III\u003c/p\u003e \u003cp\u003eModerate ridge defect\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eIn addition to stage III complexity: Need for complex rehabilitation due to: Masticatory dysfunction Secondary occlusal trauma (tooth mobility degree\u0026thinsp;\u0026ge;\u0026thinsp;2) Severe ridge defect Bite collapse, drifting, flaring\u003c/p\u003e \u003cp\u003eLess than20 remaining teeth (10 opposing pairs)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eExtent and distribution\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"5\" nameend=\"c6\" namest=\"c2\"\u003e \u003cp\u003eFor each stage, describe extent as localized (\u0026lt;\u0026thinsp;30% of teeth involved), generalized (\u0026ge;\u0026thinsp;30% of teeth involved), or molar/incisor pattern\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eCAL\u0026thinsp;=\u0026thinsp;clinical attachment loss; RBL\u0026thinsp;=\u0026thinsp;radiographic bone loss.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eGrading Criteria for Periodontitis According to the 2018 International Classification of Periodontal and Peri-implant Diseases and Conditions\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"6\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePeriodontitis grade\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eA: Slow rate of progression\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eB: Moderate rate of progression\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eC: Rapid rate of progression\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003ePrimary criteria\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eDirect evidence of progression\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eLongitudinal data (radiographic bone loss or CAL)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eEvidence of no loss over 5 years\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;2mm over 5 years\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e\u0026ge;\u0026thinsp;2mm over 5 years\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eIndirect evidence of progression\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e% bone loss/age (RBL/age)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u0026lt;0.25\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.25-1.0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e\u0026gt;1.0\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eCase phenotype\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eHeavy biofilm deposits with low levels of destruction\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eDestruction commensurate with biofilm deposits\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eDestruction exceeds expectation given biofilm deposits; specific clinical patterns suggestive of periods of rapid progression and/or early onset disease (e.g., molar/incisor pattern; lack of expected response to standard bacterial control therapies)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eGrade modifiers\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eRisk factors\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eSmoking\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNon-smoker\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eSmoker \u0026lt;10 cigarettes/day\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eSmoker\u0026thinsp;\u0026ge;\u0026thinsp;10 cigarettes/day\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eDiabetes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNormoglycemic/ no diagnosis of diabetes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eHbA1c\u0026lt;7.0% in patients with diabetes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eHbA1c\u0026thinsp;\u0026ge;\u0026thinsp;7.0% in patients with diabetes\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRisk of systemic impact of periodontitis\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eInflammatory burden\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eHigh sensitivity CRP (hsCRP)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u0026lt;1 mg /L\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e1 to 3 mg /L\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e\u0026gt;3 mg /L\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBiomarkers\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eIndicators of CAL/bone loss\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eSaliva, gingival crevicular fluid, serum\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e?\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e?\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e?\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eHbA1c, glycated hemoglobin; hsCRP, high sensitivity C-reactive protein; PA, periapical; CAL, clinical attachment loss\u003c/p\u003e\n\u003ch3\u003eDiagnostic Label Reliability Assessment\u003c/h3\u003e\n\u003cp\u003eTo ensure diagnostic accuracy and consistency, both inter-examiner and intra-examiner calibration were performed prior to data labeling.\u003c/p\u003e \u003cp\u003eFor inter-examiner reliability, 30 randomly selected cases were independently diagnosed by two experienced periodontists according to the 2018 international classification criteria. Inter-rater agreement was assessed using Cohen\u0026rsquo;s kappa coefficient, yielding substantial agreement (κ\u0026thinsp;=\u0026thinsp;0.78).\u003c/p\u003e \u003cp\u003eFor intra-examiner reliability, an additional set of 30 cases was re-evaluated by one of the examiners after a 1-week interval, blinded to the initial assessments. The intra-rater agreement was also substantial (κ\u0026thinsp;=\u0026thinsp;0.81).\u003c/p\u003e \u003cp\u003eBased on these results, this calibrated examiner subsequently performed the definitive diagnosis for all study cases to ensure diagnostic consistency and data reliability.\u003c/p\u003e\n\u003ch3\u003eData Preprocessing and statistical analyses\u003c/h3\u003e\n\u003cp\u003eData preprocessing was performed using Python 3.10. The following libraries were used: pandas (2.0.3), NumPy (1.24.3), and scikit-learn (1.3.0).\u003c/p\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003eData Preprocessing\u003c/h2\u003e \u003cp\u003eWe applied a standardized preprocessing pipeline: (\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e) categorical variables were one-hot encoded; (\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e) missing values in inferable fields (e.g., smoking status) were resolved by consensus between two calibrated clinicians, while records with irrecoverable missingness in core variables were excluded; (\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e) data types were harmonized, and descriptive terms (e.g., \u0026ldquo;severe\u0026rdquo;) were mapped to predefined numeric codes; (\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e) key quantitative features (e.g., radiographic bone loss) were extracted from free-text clinical notes using rule-based parsing; (\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e) implausible outliers (e.g., PD\u0026thinsp;\u0026gt;\u0026thinsp;15 mm) and invalid records were removed; and (\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e) predictors were pre-specified based on the 2018 international classification and supporting literature\u0026mdash;ensuring clinical relevance and interpretability without reliance on data-driven feature selection.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eFeature Selection Strategy\u003c/h3\u003e\n\u003cp\u003eRather than employing data-driven feature selection techniques (e.g., LASSO regression or recursive feature elimination), we adopted a clinically informed, hypothesis-driven approach to predictor selection. All candidate variables were pre-specified based on the 2018 international classification for periodontal and peri-implant diseases and conditions (\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e, \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e)and supported by existing peer-reviewed literature on their association with disease severity, complexity, and progression rate.\u003c/p\u003e \u003cp\u003eThe final predictor set includes: extent and distribution of periodontitis (percentage of teeth affected), PD, CAL, RBL/age, FI, TLP, number of remaining functional tooth pairs, occlusal discrepancies (including bite collapse, tooth drifting, flaring, and masticatory dysfunction), presence of severe alveolar bone loss, smoking status, and HbA1c.\u003c/p\u003e \u003cp\u003eThese variables are routinely documented in standard periodontal examinations and can be reliably extracted from structured fields in EMRs.\u003c/p\u003e \u003cp\u003eThis a priori selection strategy ensures that the model is grounded in current diagnostic standards, enhances clinical interpretability, and avoids overfitting that may arise from exploratory feature screening in moderate-sized datasets.\u003c/p\u003e\n\u003ch3\u003eDataset Splitting\u003c/h3\u003e\n\u003cp\u003eThe cohort was partitioned into training (70%), validation (20%), and test (10%) sets using stratified random sampling to preserve the distribution of periodontitis stages across all subsets.\u003c/p\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003eModel Development and Validation Strategy\u003c/h2\u003e \u003cdiv id=\"Sec12\" class=\"Section3\"\u003e \u003ch2\u003eModel Selection and Training\u003c/h2\u003e \u003cp\u003eGiven the categorical nature of periodontitis staging and grading, we evaluated multiple established machine learning algorithms for classification, including Logistic Regression, k-nearest neighbors (KNN), Decision Tree (DT), Random Forest (RF), and Support Vector Machine (SVM). Each model was trained on the training set to learn the relationship between clinical predictors and diagnostic outcomes. Hyperparameters were optimized using stratified k-fold cross-validation (k\u0026thinsp;=\u0026thinsp;5) on the training data, with performance assessed by area under the receiver operating characteristic curve (AUC) and balanced accuracy. This approach mitigated overfitting and enhanced generalizability by systematically identifying the configuration that maximized predictive performance while maintaining model stability.\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003eModel Evaluation and Validation\u003c/h2\u003e \u003cp\u003eModel performance was assessed on both the validation and held-out test sets using a comprehensive set of classification metrics, including accuracy, precision, recall (sensitivity), F1-score, and AUC. A normalized confusion matrix was generated to visualize class-wise prediction performance and identify potential misclassification patterns or systematic biases across periodontitis stages. Additionally, we conducted feature importance analysis\u0026mdash;using mean decrease in impurity (for tree-based models) or permutation importance\u0026mdash;to quantify the contribution of each predictor to model decisions, thereby enhancing interpretability and informing future clinical refinement.\u003c/p\u003e \u003cp\u003eTo mitigate the risk of overoptimistic performance estimates associated with random data splitting, we adopted a temporal validation approach: the model was trained on patients diagnosed between 2022 and 2024, and validated on a temporally distinct cohort diagnosed in 2025 from the same institution. This design simulates real-world deployment scenarios in which a model developed on historical data is applied to future patients, thereby providing a more realistic assessment of generalizability than cross-sectional random splits. All process was shown in Flowchart1.\u003c/p\u003e \u003cp\u003eAlthough this strategy does not constitute multi-center external validation (which would assess transportability across different care settings), it represents a rigorous form of internal-external validation that accounts for temporal drift in clinical practice, documentation patterns, and patient demographics\u0026mdash;key sources of model degradation in dynamic healthcare environments.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003cb\u003eFlowchart 1\u003c/b\u003e Flowchart of the entire process\u003c/p\u003e \u003c/div\u003e"},{"header":"Results","content":"\u003cdiv id=\"Sec15\" class=\"Section2\"\u003e \u003ch2\u003eCharacteristics\u003c/h2\u003e \u003cp\u003eThis study included EMRs from 692 patients. For each case, we collected demographic and clinical data\u0026mdash;including sex, age, and smoking status\u0026mdash;as well as a full set of periodontal clinical parameters: percentage of teeth affected, PD, CAL, RBL, presence and severity of FI, TLP, number of remaining functional tooth pairs, occlusal discrepancies (including bite collapse, tooth drifting, flaring, and masticatory dysfunction), and evidence of severe alveolar bone loss. HbA1c levels were also recorded. All cases were systematically staged and graded in accordance with the 2018 international classification of periodontal and peri-Implant diseases and conditions (\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e). A summary of the cohort characteristics is presented in Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e3\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eClinical Characteristics and Staging/Grading Diagnoses of the 692 Included Case Records\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"8\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c8\" colnum=\"8\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colspan=\"3\" nameend=\"c3\" namest=\"c1\"\u003e \u003cp\u003eClinical Characteristics\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003en\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"3\" nameend=\"c7\" namest=\"c5\"\u003e \u003cp\u003e%\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c8\"\u003e \u003cp\u003e\u003cem\u003eM\u0026thinsp;\u0026plusmn;\u0026thinsp;SD\u003c/em\u003e\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eSex\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003eMale\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e365\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"3\" nameend=\"c7\" namest=\"c5\"\u003e \u003cp\u003e52.75\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003eFemale\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e327\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"3\" nameend=\"c7\" namest=\"c5\"\u003e \u003cp\u003e47.25\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAge\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"3\" nameend=\"c7\" namest=\"c5\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e47.62\u0026thinsp;\u0026plusmn;\u0026thinsp;13.22\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMaximum probing\u003c/p\u003e \u003cp\u003edepth\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"3\" nameend=\"c7\" namest=\"c5\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e7.35\u0026thinsp;\u0026plusmn;\u0026thinsp;2.14\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eInterdental CAL at site of greatest loss\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"3\" nameend=\"c7\" namest=\"c5\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e7.04\u0026thinsp;\u0026plusmn;\u0026thinsp;3.35\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003e% bone loss/age (RBL/age)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003e\u0026lt;0.25\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e38\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"3\" nameend=\"c7\" namest=\"c5\"\u003e \u003cp\u003e5.49\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003e0.25-1.0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e171\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"3\" nameend=\"c7\" namest=\"c5\"\u003e \u003cp\u003e24.71\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003e\u0026gt;1.0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e483\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"3\" nameend=\"c7\" namest=\"c5\"\u003e \u003cp\u003e69.80\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eExtent and distribution\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003e\u0026ge;\u0026thinsp;30%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e464\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"3\" nameend=\"c7\" namest=\"c5\"\u003e \u003cp\u003e67.05\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003e\u0026lt;30%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e228\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"3\" nameend=\"c7\" namest=\"c5\"\u003e \u003cp\u003e32.95\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"3\" rowspan=\"4\"\u003e \u003cp\u003eFurcation involvement(FI)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e224\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"3\" nameend=\"c7\" namest=\"c5\"\u003e \u003cp\u003e32.37\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e13\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"3\" nameend=\"c7\" namest=\"c5\"\u003e \u003cp\u003e1.88\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e141\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"3\" nameend=\"c7\" namest=\"c5\"\u003e \u003cp\u003e20.38\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003e3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e314\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"3\" nameend=\"c7\" namest=\"c5\"\u003e \u003cp\u003e45.37\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eBite collapse\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003eNo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e653\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"3\" nameend=\"c7\" namest=\"c5\"\u003e \u003cp\u003e94.36\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003eYes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e39\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"3\" nameend=\"c7\" namest=\"c5\"\u003e \u003cp\u003e5.64\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eDrifting, flaring\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003eNo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e578\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"3\" nameend=\"c7\" namest=\"c5\"\u003e \u003cp\u003e83.53\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003eYes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e114\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"3\" nameend=\"c7\" namest=\"c5\"\u003e \u003cp\u003e16.47\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eMasticatory dysfunction\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003eNo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e549\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"3\" nameend=\"c7\" namest=\"c5\"\u003e \u003cp\u003e79.34\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003eYes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e143\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"3\" nameend=\"c7\" namest=\"c5\"\u003e \u003cp\u003e20.66\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eSevere ridge defect\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003eNo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e541\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"3\" nameend=\"c7\" namest=\"c5\"\u003e \u003cp\u003e78.18\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003eYes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e151\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"3\" nameend=\"c7\" namest=\"c5\"\u003e \u003cp\u003e21.82\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003eCigarettes per day\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003eNon-smoker\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e567\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"3\" nameend=\"c7\" namest=\"c5\"\u003e \u003cp\u003e81.94\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003eSmoker \u0026lt;10 cigarettes/day\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e45\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"3\" nameend=\"c7\" namest=\"c5\"\u003e \u003cp\u003e6.50\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003eSmoker\u0026thinsp;\u0026ge;\u0026thinsp;10 cigarettes/day\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e80\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"3\" nameend=\"c7\" namest=\"c5\"\u003e \u003cp\u003e11.56\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003eHbA1c\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003e\u0026ge;\u0026thinsp;7%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e25\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"3\" nameend=\"c7\" namest=\"c5\"\u003e \u003cp\u003e3.61\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003e\u0026lt;5.7%(normal)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e627\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"3\" nameend=\"c7\" namest=\"c5\"\u003e \u003cp\u003e90.61\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003e5.7%-7%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e40\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"3\" nameend=\"c7\" namest=\"c5\"\u003e \u003cp\u003e5.78\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab4\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003e(continue)\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"5\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e \u003cp\u003eClinical Characteristics\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003en\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003e%\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u003cem\u003eM\u0026thinsp;\u0026plusmn;\u0026thinsp;SD\u003c/em\u003e\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"11\" rowspan=\"12\"\u003e \u003cp\u003eDiagnosis\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eExtent and distribution\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003elocalized\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e228\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e32.95\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003egeneralized\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e464\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e67.05\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eStaging\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eStage Ⅰ\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e59\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e8.53\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eStage Ⅱ\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e93\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e13.44\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eStage Ⅲ\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e382\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e55.20\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eStage Ⅳ\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e158\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e22.93\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eGrading\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eGrade A\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e38\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e5.49\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eGrade B\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e171\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e24.71\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eGrade C\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e483\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e69.80\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTotal\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e692\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e100\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec16\" class=\"Section2\"\u003e \u003ch2\u003eModel comparison\u003c/h2\u003e \u003cp\u003eIn the periodontitis staging task, KNN, RF, and DT achieved comparable top performance, each attaining an overall accuracy of 96.99%, with F1-scores\u0026thinsp;\u0026ge;\u0026thinsp;0.89 across all stages. Notably, the DT demonstrated superior sensitivity for Stage II, achieving the highest recall (94.74%) and F1-score (0.9231), suggesting enhanced capability in identifying this less prevalent class. In contrast, RF exhibited the greatest stability for Stage III\u0026mdash;the most frequent stage\u0026mdash;with both precision and recall reaching 0.9747. Logistic Regression performed least favorably (accuracy: 93.23%), particularly underperforming on Stage II (precision: 0.7391), indicating limited discriminative power for intermediate disease severity (Table\u0026nbsp;\u003cspan refid=\"Tab5\" class=\"InternalRef\"\u003e4\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eFor the periodontitis grading task, KNN and RF again emerged as the best-performing models, yielding a joint highest accuracy of 96.40% and consistently high precision, recall, and F1-scores across all grades (A\u0026ndash;C). Conversely, Logistic Regression and SVM showed reduced performance (accuracy: 92.81% for both), with notably lower F1-scores for Grade B and Grade C, reflecting diminished ability to distinguish moderate-to-rapid progression phenotypes (Table\u0026nbsp;\u003cspan refid=\"Tab6\" class=\"InternalRef\"\u003e5\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eCollectively, ensemble and instance-based methods (RF, KNN) demonstrated robust and balanced performance across both diagnostic dimensions, while linear models (Logistic Regression) and SVM exhibited limitations in handling class heterogeneity and nuanced clinical distinctions.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab5\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 4\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eComparison of Models for Periodontitis Staging Task (Stage I-IV)\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"6\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eModel\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eClass\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003ePrecision\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eRecall\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eF1-Score\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eSupport\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLogistic Regression\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eStage I\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e1.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e10\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eStage II\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.7391\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.8947\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.8095\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e19\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eStage III\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.9605\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.9241\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.9419\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e79\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eStage IV\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.96\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.9796\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e25\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAccuracy\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e93.23%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e133\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ek-nearest neighbors\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eStage I\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e1.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e10\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eStage II\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.8571\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.9474\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.90\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e19\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eStage III\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.9870\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.9620\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.9744\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e79\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eStage IV\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e1.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e25\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAccuracy\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e96.99%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e133\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRandom Forest\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eStage I\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e1.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e10\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eStage II\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.8947\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.8947\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.8947\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e19\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eStage III\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.9747\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.9747\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.9747\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e79\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eStage IV\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e1.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e25\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAccuracy\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e96.99%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e133\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSupport Vector Machine\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eStage I\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e1.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e10\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eStage II\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.8095\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.8947\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.85\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e19\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eStage III\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.9740\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.9494\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.9615\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e79\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eStage IV\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e1.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e25\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAccuracy\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e95.49%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e133\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDecision Tree\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eStage I\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e1.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e10\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eStage II\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.90\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.9474\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.9231\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e19\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eStage III\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.9747\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.9747\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.9747\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e79\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eStage IV\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.96\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.9795\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e25\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAccuracy\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e96.99%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e133\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab6\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 5\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eComparison of Models for Periodontitis Grading Task (Grade A-C)\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"6\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eModel\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eClass\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003ePrecision\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eRecall\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eF1-Score\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eSupport\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLogistic Regression\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eGrade A\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.8333\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.7143\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.7692\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e7\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eGrade B\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.9250\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.8605\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.8916\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e43\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eGrade C\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.9355\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.9775\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.9560\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e89\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAccuracy\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e92.81%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e139\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ek-nearest neighbors\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eGrade A\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e1.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e7\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eGrade B\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.9750\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.9070\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.9400\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e43\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eGrade C\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.9565\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.9888\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.9724\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e89\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAccuracy\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e96.40%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e139\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRandom Forest\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eGrade A\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e1.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e7\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eGrade B\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.9750\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.9070\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.9400\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e43\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eGrade C\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.9565\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.9888\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.9724\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e89\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAccuracy\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e96.40%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e139\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSupport Vector Machine\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eGrade A\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.6364\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.7778\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e7\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eGrade B\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.9714\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.7907\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.8718\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e43\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eGrade C\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.9462\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.9888\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.9670\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e89\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAccuracy\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e92.81%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e139\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDecision Tree\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eGrade A\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.778\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.875\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e7\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eGrade B\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.9744\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.8837\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.9268\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e43\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eGrade C\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.956\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.9775\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.9667\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e89\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAccuracy\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e94.96%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e139\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eThe DT model for staging and the RF model for grading both exhibited high predictive performance, as evidenced by stable learning curves, clinically interpretable feature importance, and excellent classification accuracy (Figs.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e and \u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e). These results indicate that the models are accurate, well-calibrated, and suitable for integration into clinical decision support systems for periodontitis diagnosis.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec17\" class=\"Section2\"\u003e \u003ch2\u003eExternal validation using a temporal split\u003c/h2\u003e \u003cp\u003eTo assess generalizability under real-world conditions, we evaluated the model on an external test set comprising 208 periodontitis cases collected from the same institution during a subsequent time period (January 2025\u0026ndash;June 2025), which was not involved in model development or hyperparameter tuning. On this temporally held-out cohort, the model achieved an accuracy of 93.75% for the complete diagnosis\u0026mdash;defined as the simultaneous correct prediction of extent, stage, and grade according to the 2018 international classification.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec18\" class=\"Section2\"\u003e \u003ch2\u003eModel application\u003c/h2\u003e \u003cp\u003eThis system is an intelligent decision-support tool designed for staging and grading the diagnosis of periodontitis. The front end utilizes the jQuery library from BootCDN to provide a user-friendly interface for interaction, while the back end is built on the Python Flask framework to handle business logic, data storage, and model computation. The front end and back end communicate via HTTP requests, implementing a decoupled architecture that enhances development efficiency and system maintainability. The business workflow is illustrated in Flowchart 2:\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003cb\u003eFlowchart 2\u003c/b\u003e Flowchart of the development of User interface\u003c/p\u003e \u003cp\u003eA user-friendly web-based interface was developed to integrate the staging and grading models for clinical use (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e). The interface allows clinicians to input key periodontal parameters (e.g., PD, CAL, TLP, FI, RBL/age) and select from multiple ML algorithms (including RF, DT, KNN, and SVM). Upon submission, the system automatically generates a diagnosis prediction, including the stage and grade of periodontitis, along with model performance metrics (accuracy for staging and grading across different algorithms). This tool facilitates real-time decision support in clinical practice.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e"},{"header":"Discussion","content":"\u003cp\u003eOur study demonstrates that machine learning models\u0026mdash;specifically KNN, RF, and DT\u0026mdash;can achieve high diagnostic accuracy in both staging (96.99%) and grading (96.40%) of periodontitis according to the 2018 international classification. Crucially, this performance generalizes beyond internal validation: when evaluated on an external test set comprising cases from the same institution but collected during a subsequent time period (temporal split), the model maintained a high overall accuracy of 93.75% for full diagnosis (extent\u0026thinsp;+\u0026thinsp;stage\u0026thinsp;+\u0026thinsp;grade). This temporal external validation provides strong evidence that the model is not merely overfitting to cohort-specific patterns but captures stable, clinically relevant disease signatures\u0026mdash;enhancing its potential for real-world deployment.\u003c/p\u003e \u003cp\u003eThe 2018 international classification, while conceptually rigorous, poses practical challenges in routine care, particularly for non-specialists. Prior studies report suboptimal inter-rater agreement, with general dentists showing reduced accuracy\u0026mdash;especially in distinguishing Stage III from IV\u0026mdash;and greater difficulty diagnosing mild disease (\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e). In this context, our results represent a significant advance: the models not only replicate expert-level judgment but do so consistently across all disease stages, as evidenced by high per-class F1-scores (\u0026ge;\u0026thinsp;0.89) and minimal misclassification in normalized confusion matrices.\u003c/p\u003e \u003cp\u003eNotably, our approach substantially outperforms both human-assisted and generic AI\u0026ndash;based strategies. Marini et al.(\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e)found that even with dedicated software support, general dentists achieved only 53.6% concordance in full diagnosis. More recently, Tastan et al. (\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e) reported that ChatGPT\u0026mdash;a widely accessible large language model\u0026mdash;achieved merely 59.5% accuracy in staging and 50.5% in grading, despite explicit prompting with the 2018 international classification criteria. These findings collectively suggest that domain-specific training on structured clinical data is essential for reliable periodontal classification, and that off-the-shelf AI tools lack the necessary granularity for this task.\u003c/p\u003e \u003cp\u003eA key strength of our models lies in their clinical interpretability. Feature importance analysis identified CALand PD as the top predictors for staging\u0026mdash;directly aligning with the 2018 international classification framework\u0026rsquo;s designation of CAL as the primary indicator of disease severity (\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e). This biologically grounded feature selection enhances model credibility and facilitates clinician trust. Secondary contributors\u0026mdash;including severe alveolar bone loss, tooth displacement, and TLP\u0026mdash;reflect established markers of advanced disease (\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e), while variables unrelated to core diagnostic criteria (e.g., remaining functional tooth pairs) showed negligible influence, indicating that the models focus on clinically meaningful signals rather than confounding factors.\u003c/p\u003e \u003cp\u003eFor grading, where longitudinal data on disease progression are rarely available in routine care, our models relied predominantly on the RBL/age\u0026mdash;validating its utility as a surrogate for progression rate under real-world constraints. Smoking also emerged as a significant modifier, consistent with its well-documented role in impairing host response. Intriguingly, HbA1c exhibited limited predictive value, possibly due to sample size or overshadowing by stronger predictors.\u003c/p\u003e \u003cp\u003eWe observed a marked underrepresentation of Grade A cases (5.49%), likely attributable to the stringent RBL/age\u0026thinsp;\u0026lt;\u0026thinsp;0.25 threshold. Although our model achieved 100% recall for this class internally, the extreme class imbalance (only 7 Grade A samples in the validation set) may limit generalizability to populations with milder disease. This finding resonates with growing concerns that the current grading thresholds may not reflect clinical reality (\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e, \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e). While we adhered strictly to the official criteria, our data provide empirical support for proposals to revise the cutoffs (e.g., to \u0026lt;\u0026thinsp;0.5 for Grade A), which could improve both clinical feasibility and model performance across the full spectrum of disease.\u003c/p\u003e \u003cp\u003eLimitations include the single-center design and modest representation of early-stage disease. Nevertheless, the combination of high internal performance, robust temporal external validation (93.75% full-diagnosis accuracy), and interpretable feature reliance positions our models as a clinically actionable tool. Future prospective, multi-center studies should validate these findings in diverse populations and explore adaptive thresholds to further enhance equity and utility in routine periodontal care.\u003c/p\u003e \u003cp\u003e A key methodological strength of our study is the theory-guided, guideline-concordant feature selection process. Instead of exhaustively evaluating hundreds of potential predictors\u0026mdash;a common practice in machine learning that risks yielding statistically significant but clinically irrelevant features\u0026mdash;we deliberately restricted our input variables to those explicitly recommended or implied by the 2018 international classification for periodontitis. This approach aligns our model with established clinical reasoning pathways and ensures that every predictor has a clear pathophysiological or diagnostic rationale.\u003c/p\u003e \u003cp\u003eBy prioritizing clinical validity over algorithmic novelty, we enhance the transparency, trustworthiness, and real-world applicability of our staging and grading system. Practitioners are more likely to adopt decision-support tools that reflect familiar diagnostic criteria rather than opaque \u0026ldquo;black-box\u0026rdquo; outputs derived from data-mining alone. While future studies may explore hybrid approaches combining expert knowledge with data-driven refinement, our current design reflects the pragmatic reality of clinical documentation and supports seamless integration into routine dental workflows\u0026mdash;particularly in settings lacking advanced computational or laboratory infrastructure.\u003c/p\u003e \u003cp\u003eSeveral features recommended in the 2018 international classification were not included in our models due to practical and methodological considerations. First, the phenotypic grading criterion based on the disproportionality between plaque accumulation and periodontal destruction\u0026mdash;where excessive tissue loss relative to plaque burden suggests Grade C\u0026mdash;was excluded due to its reliance on subjective clinical judgment and the absence of standardized, quantifiable thresholds in routine documentation. In machine learning applications that prioritize reproducibility and cross-institutional generalizability, such semi-qualitative judgments risk introducing measurement bias and undermining model robustness. Second, systemic inflammatory biomarkers\u0026mdash;such as hsCRP\u0026mdash;and host-derived inflammatory mediators, although biologically relevant to periodontitis progression, were not routinely measured in standard clinical practice during the study period. Their low prevalence and high rate of missingness in EMRs would have required either extensive imputation or the exclusion of a large number of cases, thereby reducing the generalizability of the model and introducing potential selection bias.\u003c/p\u003e \u003cp\u003eAccordingly, our model was built exclusively on predictors that are routinely documented, objectively measurable, and consistently available in standard periodontal care\u0026mdash;namely, PD, CAL, RBL/age, extent and distribution of periodontitis (percentage of teeth affected), FI, number of remaining functional tooth pairs, occlusal discrepancies (including bite collapse, tooth drifting, flaring, and masticatory dysfunction), severe alveolar bone loss, smoking status, and HbA1c. These variables collectively reflect the key domains of the 2018 international classification\u0026mdash;disease severity, complexity, and estimated progression rate\u0026mdash;and can be reliably extracted from electronic health records without reliance on specialized laboratory assays or subjective phenotypic assessments.\u003c/p\u003e \u003cp\u003eThis pragmatic feature selection strategy enhances model transparency, reproducibility, and clinical deployability, particularly in real-world dental settings where access to advanced diagnostics is limited. Although this approach may not fully capture host-mediated inflammatory dynamics or rare phenotypic subtypes, it aligns with current clinical workflows and prioritizes operational feasibility over theoretical comprehensiveness. Future prospective studies integrating standardized systemic biomarker panels and natural language processing of unstructured clinical notes could facilitate more nuanced, biologically informed staging and grading in automated decision-support systems.\u003c/p\u003e"},{"header":"Conclusions","content":"\u003cp\u003eIn conclusion, the machine learning model developed in this study demonstrates high diagnostic accuracy and stability in the simultaneous staging and grading of periodontitis, achieving 93.75% full-diagnosis accuracy on a temporally held-out external cohort\u0026mdash;a finding that underscores its robust generalizability under real-world conditions. Critically, the model\u0026rsquo;s decision logic aligns with biologically grounded clinical criteria: key predictors such as CAL for staging and RBL/age for grading directly reflect the core principles of the 2018 international classification. This alignment not only mirrors expert clinical reasoning but also enhances diagnostic consistency, particularly in complex cases where inter-rater variability is high. Nevertheless, due to limited Grade A cases and the single-center data source, the model should serve as a decision-support tool rather than a replacement for clinical judgment; comprehensive patient assessment remains essential. Future work should focus on multicenter data collection to enrich underrepresented categories (particularly Grade A), mitigate class imbalance, and validate generalizability across diverse populations\u0026mdash;key steps toward realizing AI-enabled precision periodontics in real-world practice.\u003c/p\u003e"},{"header":"Declarations","content":" \u003cp\u003e \u003cstrong\u003eEthics approval and consent to participate\u003c/strong\u003e \u003cp\u003eThe study protocol was approved by the Ethics Committee of School \u0026amp; Hospital of Stomatology, Guangxi Medical University (Approval No.: 2024057). All patient data were anonymized prior to analysis, and the requirement for informed consent was waived due to the retrospective nature of the study.\u003c/p\u003e \u003c/p\u003e \u003cp\u003e \u003cstrong\u003eConsent for publication\u003c/strong\u003e \u003cp\u003eNot appicable.\u003c/p\u003e \u003c/p\u003e \u003cdiv class=\"Heading\"\u003e\u003c/div\u003e \u003cp\u003e \u003cb\u003eDeclaration of generative AI and AI-assisted technologies in the writing process\u003c/b\u003e \u003c/p\u003e\u003cp\u003e \u003ch2\u003eCompeting interests\u003c/h2\u003e \u003cp\u003eThe authors declare no competing interests.\u003c/p\u003e \u003ch2\u003eAuthor details\u003c/h2\u003e \u003cp\u003e\u003csup\u003e1\u003c/sup\u003eDepartment of Periodontics and Oral Medicine, College of Stomatology, Hospital of Stomatology, Guangxi Medical University, Nanning, Guangxi 530021, P. R. China\u003c/p\u003e \u003cp\u003e\u003csup\u003e2\u003c/sup\u003eGuangxi Health Commission Key Laboratory of Prevention and Treatment for Oral Infectious Diseases, Nanning, Guangxi 530021, P. R. China\u003c/p\u003e \u003cp\u003e\u003csup\u003e3\u003c/sup\u003eGuangxi Key Laboratory of Oral and Maxillofacial Rehabilitation and Reconstruction, Guangxi Universities and Colleges Key Laboratory of Oral and Maxillofacial Surgery Disease Treatment, Guangxi Clinical Research Center for Craniofacial Deformity, Nanning, Guangxi 530021, P. R. China\u003c/p\u003e \u003cp\u003e \u003csup\u003e4\u003c/sup\u003e Department of Information Technology, College of Stomatology, Hospital of Stomatology, Guangxi Medical University, Nanning, Guangxi 530021, P. R. China\u003c/p\u003e \u003cp\u003e \u003csup\u003e5\u003c/sup\u003eCollege of Computer, Electronics and Information, Guangxi University, Nanning, Guangxi 530004, P.R. China\u003c/p\u003e \u003cp\u003e \u003csup\u003e6\u003c/sup\u003e Department of Conservative Dentistry and Endodontics, College of Stomatology, Hospital of Stomatology, Guangxi Medical University, Nanning, Guangxi 530021, P. R. China\u003c/p\u003e \u003c/p\u003e\u003ch2\u003eFunding\u003c/h2\u003e \u003cp\u003eThis work was conducted during the doctoral training of the first author at Guangxi Medical University. No external or internal grant funding was specifically allocated to this project. The research benefited from the clinical data infrastructure and academic supervision provided by Hospital of Stomatology, Guangxi Medical University.\u003c/p\u003e\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eZ.Y. designed the study and drafted the manuscript; Y.L. and F.M. extracted and organized the clinical data; Z.N. and S.N. was responsible for data quality control, maintenance of the clinical information system, and statistical analyses; Z.M. provided critical guidance on the design of the machine learning model and the development of the user interface; R.T. supervised the research project and critically revised the manuscript. All authors contributed to the article and approved the submitted version. All authors take responsibility for all aspects of the reliability and freedom from bias of the data presented and their discussed interpretation.\u003c/p\u003e\u003ch2\u003eAcknowledgement\u003c/h2\u003e\u003cp\u003e We thank all participants at Hospital of Stomatology, Guangxi Medical University.\u003c/p\u003e\u003ch2\u003eData Availability\u003c/h2\u003e\u003cp\u003eData sharing is not applicable to this article as no datasets were generated or analyzed during the current study.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eChapple ILC, Mealey BL, Van Dyke TE, Bartold PM, Dommisch H, Eickholz P, et al. Periodontal health and gingival diseases and conditions on an intact and a reduced periodontium: Consensus report of workgroup 1 of the 2017 World Workshop on the Classification of Periodontal and Peri-Implant Diseases and Conditions. J Clin Periodontol. 2018;45(Suppl 20):S68\u0026ndash;77.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMurakami S, Mealey BL, Mariotti A, Chapple ILC. Dental plaque-induced gingival conditions. J Periodontol. 2018;89(Suppl 1):S17\u0026ndash;27.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePapapanou PN, Sanz M, Buduneli N, Dietrich T, Feres M, Fine DH, et al. Periodontitis: Consensus report of workgroup 2 of the 2017 World Workshop on the Classification of Periodontal and Peri-Implant Diseases and Conditions. J Periodontol. 2018;89(Suppl 1):S173\u0026ndash;82.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKinane DF, Lappin DF, Culshaw S. The role of acquired host immunity in periodontal diseases. Periodontol 2000. 2024;00:1\u0026ndash;15.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKinane DF, Bornstein MM. Introduction to the Diagnostics in Periodontology and Implant Dentistry issue. Periodontol 2000. 2024;95(1):7\u0026ndash;9.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eEke PI, Dye BA, Wei L, Slade GD, Thornton-Evans GO, Borgnakke WS, et al. Update on Prevalence of Periodontitis in Adults in the United States: NHANES 2009 to 2012. J Periodontol. 2015;86(5):611\u0026ndash;22.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen MX, Zhong YJ, Dong QQ, Wong HM, Wen YF. Global, regional, and national burden of severe periodontitis, 1990\u0026ndash;2019: An analysis of the Global Burden of Disease Study 2019. J Clin Periodontol. 2021;48(9):1165\u0026ndash;88.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJiao J, Jing W, Si Y, Feng X, Tai B, Hu D, et al. The prevalence and severity of periodontal disease in Mainland China: Data from the Fourth National Oral Health Survey (2015\u0026ndash;2016). J Clin Periodontol. 2020;48(2):168\u0026ndash;79.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTonetti MS, Jepsen S, Jin L, Otomo-Corgel J. Impact of the global burden of periodontal diseases on health, nutrition and wellbeing of mankind: A call for global action. J Clin Periodontol. 2017;44(5):456\u0026ndash;62.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePattamatta M, Chapple I, Listl S. The value-for money of preventing and managing periodontitis: Opportunities and challenges. Periodontol 2000. 2024;00:1\u0026ndash;9.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNascimento GG, Alves-Costa S, Romandini M. Burden of severe periodontitis and edentulism in 2021, with projections up to 2050: The Global Burden of Disease 2021 study. J Periodontal Res. 2024;59(5):823\u0026ndash;67.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTonetti MS, Greenwell H, Kornman KS. Staging and grading of periodontitis: Framework and proposal of a new classification and case definition. J Clin Periodontol. 2018;45(Suppl 20):S149\u0026ndash;61.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePapapanou PN, Sanz M, Buduneli N, Dietrich T, Feres M, Fine DH, et al. Periodontitis: 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\u0026ndash;70.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMarini L, Tonetti MS, Nibali L, Rojas MA, Aimetti M, Cairo F, et al. The staging and grading system in defining periodontitis cases: consistency and accuracy amongst periodontal experts, general dentists and undergraduate students. J Clin Periodontol. 2021;48(2):205\u0026ndash;15.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAbrahamian L, Pascual-LaRocca A, Barallat L, Valles C, Herrera D, Sanz M, et al. Intra- and inter-examiner reliability in classifying periodontitis according to the 2018 classification of periodontal diseases. J Clin Periodontol. 2022;49(8):732\u0026ndash;9.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGandhi KK, Katwal D, Chang J, Blanchard S, Shin D, Maupome G, et al. Diagnosis and treatment planning using the 2017 classification of periodontal diseases among three dental schools. J Dent Educ. 2022;86(11):1521\u0026ndash;8.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTonetti MS, Sanz M. Implementation of the new classification of periodontal diseases: Decision-making algorithms for clinical practice and education. J Clin Periodontol. 2019;46(4):398\u0026ndash;405.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMarini L, Tonetti MS, Nibali L, Sforza NM, Landi L, Cavalcanti R, et al. Implementation of a software application in staging and grading of periodontitis cases. Oral Dis. 2024;30(2):719\u0026ndash;28.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen W, Lin G, Chen Y, Cheng F, Li X, Ding J, et al. Prediction of the Ki-67 expression level in head and neck squamous cell carcinoma with machine learning-based multiparametric MRI radiomics: a multicenter study. BMC Cancer. 2024;24(1):418.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eShan J, Jiang R, Chen X, Zhong Y, Zhang W, Xie L, et al. Machine Learning Predicts Lymph Node Metastasis in Early-Stage Oral Tongue Squamous Cell Carcinoma. J Oral Maxillofac Surg. 2020;78(12):2208\u0026ndash;18.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eReyes LT, Knorst JK, Ortiz FR, Ardenghi TM. Machine Learning in the Diagnosis and Prognostic Prediction of Dental Caries: A Systematic Review. Caries Res. 2022;56(3):161\u0026ndash;70.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChau KK, Zhu M, AlHadidi A, Wang C, Hung K, Wohlgemuth P, et al. A novel AI model for detecting periapical lesion on CBCT: CBCT-SAM. J Dent. 2025;153:105526.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTastan Eroglu Z, Babayigit O, Ozkan Sen D, Ucan Yarkac F. Performance of ChatGPT in classifying periodontitis according to the 2018 classification of periodontal diseases. Clin Oral Investig. 2024;28(7):407.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCollins GS, Dhiman P, Andaur Navarro CL, Ma J, Hooft L, Reitsma JB, et al. Protocol for development of a reporting guideline (TRIPOD-AI) and risk of bias tool (PROBAST-AI) for diagnostic and prognostic prediction model studies based on artificial intelligence. BMJ Open. 2021;11(7):e048008.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYan MH, Zhao YY, Liu X, Li W, Wang Y. [Interpretation of the transparent reporting of a multivariable prediction model for individual prognosis or diagnosis based on regression or machine learning methods (TRIPOD\u0026thinsp;+\u0026thinsp;AI)]. Zhonghua nei ke za zhi. 2025;64(1):4\u0026ndash;10.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCollins GS, Moons KGM, Dhiman P, Riley RD, Beam AL, Van Calster B, et al. TRIPOD\u0026thinsp;+\u0026thinsp;AI statement: updated guidance for reporting clinical prediction models that use regression or machine learning methods. BMJ. 2024;385:e078378.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePeduzzi P, Concato J, Kemper E, Holford TR, Feinstein AR. A simulation study of the number of events per variable in logistic regression analysis. J Clin Epidemiol. 1996;49(12):1373\u0026ndash;9.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen Y, Lu P, Lin C, Li S, Zhu Y, Tan J, et al. Hyperuricemia and elevated uric acid/creatinine ratio are associated with stages III/IV periodontitis: a population-based cross-sectional study (NHANES 2009\u0026ndash;2014). BMC Oral Health. 2024;24(1):1389.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhao L, Xu T, Hu W, Chung KH. Preservation and augmentation of molar extraction sites affected by severe bone defect due to advanced periodontitis: A prospective clinical trial. Clin Implant Dent Relat Res. 2018;20(3):333\u0026ndash;44.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDietrich T, Ower P, Tank M, West NX, Walter C, Needleman I, et al. Periodontal diagnosis in the context of the 2017 classification system of periodontal diseases and conditions - implementation in clinical practice. Br Dent J. 2019;226(1):16\u0026ndash;22.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"bmc-medical-informatics-and-decision-making","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"midm","sideBox":"Learn more about [BMC Medical Informatics and Decision Making](http://bmcmedinformdecismak.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/midm/default.aspx","title":"BMC Medical Informatics and Decision Making","twitterHandle":"BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Periodontitis, Machine learning, Staging, Grading, 2018 international classification, Clinical decision support","lastPublishedDoi":"10.21203/rs.3.rs-8504183/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8504183/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eBackground\u003c/h2\u003e \u003cp\u003eThe 2018 international classification for periodontitis enables individualized patient management through simultaneous staging (disease severity and complexity) and grading (progression rate). However, its clinical adoption is hindered by diagnostic inconsistency\u0026mdash;particularly in grading and differentiation of advanced stages\u0026mdash;due to the complexity of age-adjusted metrics such as radiographic bone loss\u0026ndash;to\u0026ndash;age ratio (RBL/age). While artificial intelligence (AI) shows promise in dental diagnostics, existing tools lack robust integration of multimodal clinical data and validation under real-world conditions.\u003c/p\u003e\u003ch2\u003eMethods\u003c/h2\u003e \u003cp\u003eWe retrospectively collected data from 692 patients diagnosed with periodontitis at Hospital of Stomatology, Guangxi Medical University between June 2022 and June 2024. After data cleaning and feature selection, cases were labeled according to the 2018 international classification criteria. Machine learning models\u0026mdash;including k-nearest neighbors (KNN), Random Forest (RF), and Decision Tree (DT)\u0026mdash;were trained and optimized via hyperparameter tuning. Model performance was evaluated on internal test sets and further validated on a temporally held-out external cohort (n\u0026thinsp;=\u0026thinsp;208).\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e \u003cp\u003eFor staging, KNN, RF, and DT achieved 96.99% accuracy; DT showed the highest recall (94.74%) and F1-score (0.9231) for Stage II, while RF excelled in Stage III (recall and F1-score: 0.9747). For grading, KNN and RF reached 96.40% accuracy, with F1-scores\u0026thinsp;\u0026gt;\u0026thinsp;0.97 for Grade C. Feature importance analysis identified clinical attachment loss (CAL) and probing depth (PD) as top predictors for staging, and RBL/age as the dominant feature for grading. In temporal external validation, the model achieved 93.75% accuracy for full diagnosis (extent\u0026thinsp;+\u0026thinsp;stage\u0026thinsp;+\u0026thinsp;grade).\u003c/p\u003e\u003ch2\u003eConclusions\u003c/h2\u003e \u003cp\u003eThe proposed model demonstrates high accuracy, interpretability, and generalizability, offering a promising decision-support tool for standardized periodontitis classification.\u003c/p\u003e","manuscriptTitle":"A Clinically Interpretable Machine Learning Model for Staging and Grading of Periodontitis: Development and Temporal External Validation as a Decision Support Tool","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-02-10 00:38:00","doi":"10.21203/rs.3.rs-8504183/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"editorInvitedReview","content":"","date":"2026-02-26T19:10:36+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-02-19T12:48:46+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"19145634273609608340779924209882447206","date":"2026-02-14T07:19:49+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"54444291187683708687955994447383728559","date":"2026-02-13T11:53:33+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-02-07T15:50:08+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"66837341144450387482752875211850013692","date":"2026-02-07T08:06:04+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2026-02-05T07:20:30+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2026-01-12T13:27:18+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2026-01-10T13:39:48+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2026-01-10T13:39:47+00:00","index":"","fulltext":""},{"type":"submitted","content":"BMC Medical Informatics and Decision Making","date":"2026-01-03T03:36:10+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"bmc-medical-informatics-and-decision-making","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"midm","sideBox":"Learn more about [BMC Medical Informatics and Decision Making](http://bmcmedinformdecismak.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/midm/default.aspx","title":"BMC Medical Informatics and Decision Making","twitterHandle":"BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"d312489a-eb56-48b7-9ef3-bab779c6062e","owner":[],"postedDate":"February 10th, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2026-02-10T00:38:00+00:00","versionOfRecord":[],"versionCreatedAt":"2026-02-10 00:38:00","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8504183","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8504183","identity":"rs-8504183","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}