PERIODONTAL-AI HARMONIZATION PROTOCOL (PAIHP): A MODULAR FRAMEWORK FOR STANDARDIZING AI PIPELINES IN PERIODONTAL CLINICAL

Background:AI is transforming periodontal research and clinical diagnostics, but lack of standardized methods for data handling, modeling, and evaluation limits reproducibility and regulatory approval. Reporting standards like CONSORT-AI and STARD-AI guide reporting but not technical implementation. This study introduces the Periodontal-AI Harmonization Protocol (PAIHP), a modular, explainable, and FAIR-compliant framework for reproducible AI in periodontal research.
Materials and Methods: PAIHP consists of five modules:

(1) Modular Data Schema Template (MDST), for HL7 FHIRcompliant dataset structures;

(2) Preprocessing and Harmonization Protocol (PHP), covering normalization, encoding, and imputation;

(3) Explainable AI Modeling Workflow (XAI-MW), containerized deep learning with explainability modules;

(4) Multi-Metric Evaluation Protocol (MMEP), defining clinical assessment metrics;

(5) Benchmarking and Registry System,
supporting version-controlled repositories. Validation used open-access periodontal imaging and clinical datasets, assessing accuracy, calibration, interpretability, and robustness.
Results: PAIHP improved reproducibility, transparency, and interoperability. Harmonized data schemas increased reusability by 37%, preprocessing reduced feature variance by 22%, and explainability modules boosted interpretability by 31%. Multimetric evaluation established clinical benchmarks for sensitivity and generalizability. PAIHP-compliant pipelines showed higher consistency and reproducibility than conventional workflows.
Conclusions: PAIHP offers a unified, modular, and FAIR-compliant framework for developing, evaluating, and
benchmarking AI in periodontal research. By extending beyond reporting standards into technical implementation, it
overcomes key barriers to reproducibility and regulatory acceptance. Future work will expand the protocol to broader dental AI applications and integrate real-world clinical validation