Periodontal Disease
Conditions
Brief summary
Background and Objective: Periodontitis and gingivitis are highly prevalent oral diseases that require accurate diagnostic classification and continuous gingival health monitoring. This study aims to develop, internally validate, and externally evaluate the diagnostic accuracy of artificial intelligence (AI) models for periodontitis staging and gingival inflammation assessment at both tooth and patient levels. Study Design: This is a multi-center observational study utilizing a large-scale primary clinical dataset for model development. To rigorously evaluate the generalizability of the trained AI models, two distinct pathways of independent external validation will be implemented across multiple clinical sites. Research Phases & Validation Architecture: Phase 1 (Periodontitis Diagnosis via Probing): Development of an AI model to diagnose periodontitis (binary classification: stage 0/I vs. stage II/III/IV) at both tooth and patient levels, using comprehensive clinical periodontal probing as the gold standard. External Validation I will be performed using an independent cohort from another campus of the primary hospital to test the model's diagnostic accuracy. Phase 2 (Periodontitis Diagnosis via Radiographs): Development of an AI model to diagnose periodontitis (binary classification: stage 0/I vs. stage II/III/IV) at both tooth and patient levels, using digital panoramic radiographs as the reference standard. External Validation II will be conducted using distinct, independent image datasets acquired from two separate regional hospitals to evaluate geographic generalizability. Phase 3 (Gingival Inflammation Monitoring): Development of an AI model to monitor and assess gingival inflammation at both tooth and patient levels, based on Probing Depth (PD) and Bleeding on Probing (BOP) as the gold standard. This model's performance will also be evaluated through External Validation I using the independent dataset from the primary hospital's alternative campus. Significance: By validating the AI models across varied institutional workflows and imaging systems, this study will provide high-level evidence on the clinical utility and robustness of AI-driven digital systems for automated periodontal screening and long-term health monitoring.
Interventions
DIAGNOSTIC_TESTAI-driven Periodontal Diagnostic and Monitoring Algorithms
The intervention evaluated in this observational study is the deployment of deep learning/artificial intelligence (AI) software models. The AI algorithms process two streams of standard clinical data to perform three automated diagnostic tasks without altering patient care: Automated classification of periodontitis stages (Stage 0/I vs. Stage II/III/IV) utilizing full-mouth clinical charting metrics. Automated classification of periodontitis stages (Stage 0/I vs. Stage II/III/IV) utilizing digital panoramic radiographs. Automated assessment and monitoring of gingival inflammation flags based on Probing Depth (PD) and Bleeding on Probing (BOP) patterns. The outputs of these AI models will be directly compared against clinical and radiographic gold standards to calculate diagnostic accuracy metrics.
Sponsors
Shanghai Ninth People's Hospital Affiliated to Shanghai Jiao Tong University
Study design
Observational model
COHORT
Time perspective
CROSS_SECTIONAL
Eligibility
Sex/Gender
ALL
Age
18 Years to No maximum
Healthy volunteers
Yes
Inclusion criteria
1. Patients aged \> 18 years at the time of their clinical periodontal examination. 2. Availability of complete full-mouth periodontal charting records, which must include Probing Depth (PD) and Bleeding on Probing (BOP) documented at 6 sites per tooth. 3. Availability of a digital panoramic radiograph of acceptable diagnostic quality, taken within one months of the clinical periodontal examination.
Exclusion criteria
1. Patients who are completely edentulous or those who have undergone full-arch dental implant rehabilitation (not applicable for natural teeth periodontitis staging). 2. Panoramic radiographs with severe image degradation, including major motion artifacts, severe positioning errors, or poor contrast/exposure that obscures the alveolar bone crest. 3. Presence of extensive metal artifacts or massive bilateral multiple fixed crowns/bridges that completely shadow the marginal bone level of interest. 4. Incomplete clinical electronic medical records or missing core diagnostic descriptors required to establish the clinical gold standard for periodontitis staging or gingival inflammation.
Design outcomes
Primary
| Measure | Time frame | Description |
|---|---|---|
| Diagnostic Accuracy of the AI Model for Probing-Based Periodontitis Staging | Baseline (At a single point in time for each participant (cross-sectional assessment)) | The diagnostic performance of the deep learning AI model in classifying periodontitis stages (binary classification: Stage 0/I vs. Stage II/III/IV) at both individual tooth and patient levels, using comprehensive clinical periodontal probing as the gold standard. Performance will be evaluated using the internal development dataset and verified using External Validation Dataset I (secondary campus data). Metrics will include Area Under the Receiver Operating Characteristic curve (AUC), Sensitivity, Specificity, and F1-score. |
| Diagnostic Accuracy of the AI Model for Radiograph-Based Periodontitis Staging | Baseline (At a single point in time for each participant (cross-sectional assessment)) | The diagnostic performance of the deep learning AI model in classifying periodontitis stages (binary classification: Stage 0/I vs. Stage II/III/IV) at both individual tooth and patient levels, using digital panoramic radiographs as the reference standard. Performance will be evaluated using the internal development dataset and verified using External Validation Dataset II (multi-center data from two separate regional hospitals). Metrics will include Area Under the Receiver Operating Characteristic curve (AUC), Sensitivity, Specificity, and F1-score. |
| Diagnostic Accuracy of the AI Model for Gingival Inflammation Monitoring | Baseline (At a single point in time for each participant (cross-sectional assessment)) | The performance of the deep learning AI model in detecting and monitoring gingival inflammation flags at both individual tooth and patient levels, using Probing Depth (PD) and Bleeding on Probing (BOP) metrics as the clinical gold standard. Performance will be evaluated using the internal development dataset and verified using External Validation Dataset I (secondary campus data). Metrics will include Sensitivity, Specificity, Positive Predictive Value (PPV), and Negative Predictive Value (NPV). |
Countries
China
Outcome results
None listed