Agent-based Multimodal Periodontal Disease Simulator: An In-silico Study of Bone Loss Dynamics
Year: 2026
Type: article
Abstract: Introduction: Periodontitis is a chronic inflammatory disease that leads to irreversible damage to the gums, periodontal ligament, and alveolar bone. Its progression is driven by a complex interaction of microbial imbalance, host immune responses, and patient-specific risk factors like smoking and diabetes. Digital Twin (DT) technology- virtual models that mimic anatomical and physiological behaviours- has shown promise in orthodontics and implantology, but its use in periodontal disease remains unexplored. Hereby, authors introduce an agent-based, multimodal periodontal disease simulator that generates synthetic patient groups, models inflammation and bone loss over time, and produces synthetic radiographic images. Aim: To develop and evaluate an agent-based, multimodal in-silico simulator to model periodontal disease progression by simulating inflammatory dynamics, alveolar bone loss, and synthetic radiographic features, and to assess the performance of machine learning models trained on the generated data. Materials and Methods: Three interacting agents were created: Data agent generated a cohort of 600 patients with demographic data, risk factors, and baseline imaging. The computational simulation used synthetic data and agentbased modeling without involving human participants, clinical facilities, or time-bound data collection, so Institutional location and study duration are not applicable. Inflammation agent simulated month-by-month inflammatory trajectories based on risk factors and graph-based neighbour interactions, and bone loss agent translated inflammation into bone loss scores and updated lesion masks. Results: Machine learning models trained on this synthetic data achieved high accuracy in classifying severity {accuracy ≈ 0.98, macro F1≈0.97, Receiver Operation Characteristic-Area Under Curve (ROC-AUC)≈0.997}, although continuous bone loss regression performed poorly (R²≈0.10). Simple adaptive thresholding for lesion segmentation yielded moderate mean Intersection over Union (IoU) (≈ approximately 0.67) and Dice (≈approximately 0.75) scores. An ablation study of the network indicated that even small interactions among neighbours increased average bone loss, suggesting population-level effects. Statistical evaluation employed a Random Forest classifier and regressor for severity classification and bone loss prediction, with segmentation assessed via IoU, dice, precision, and recall. Conclusion: Although the dataset is synthetic and not validated, this work demonstrates the potential of combining agent-based modeling, graph theory, synthetic imaging, and machine learning to explore the dynamics of periodontal disease.
Language: English
Subfield: Periodontics
Field: Dentistry
Domain: Health Sciences
Open Access status: gold
Estimated APC paid: $275
OpenAlex ID: W7155197752
FWCI:
Cites:
Cited by:
Related to: