Advancing Personalized Periodontal Therapy Using AI and Multimodal Data Integration - PROJECT SUMMARY/ABSTRACT Periodontitis, a multifaceted chronic inflammatory disease affecting the specialized tissues supporting teeth, leads to irreversible bone and tooth loss and contributes to or exacerbates systemic inflammatory conditions. Its severe forms affect one-fifth of the population over the age of 15, making periodontitis a major health burden, particularly in an aging global population, and presenting significant challenges in both clinical and public health domains. Despite decades of research and the growing recognition of oral health's impact on systemic conditions, current treatment protocols still adhere to a “one-size-fits-all” approach—relying primarily on mechanical cleaning, surgery, and antibiotic regimens for severe cases. This treatment approach only accounts for a few specific risk factors but mostly overlooks individual patient phenotypes and profound inter-individual differences, leading to inconsistent outcomes. Recent advances in disease understanding and the emergence of artificial intelligence and machine learning provide an unprecedented opportunity to overcome these limitations and move periodontology toward precision medicine. We hypothesize that periodontal patients present with different, unique disease phenotypes characterized by distinct clinical, microbiological, and immunological signatures. Consequently, patients with different disease phenotypes respond differently to various periodontal treatment protocols. Based on this hypothesis, the overall goal of this project is the development of a personalized approach to periodontal diagnosis, risk assessment, and treatment planning. We will integrate a high-volume, granular dataset, sourced from the most robust raw randomized clinical trials spanning up to five years of various periodontal therapies, alongside thousands of new oral samples evaluated through next- generation sequencing, and hundreds of thousands of electronic dental records; we will apply advanced machine learning methods for analysis. First, we will use unsupervised clustering to stratify patients based on their disease phenotypes, which are more complex and personalized than traditional, human-defined categories. This stratification could explain the variable clinical manifestations of periodontitis and serve as a foundation for tailored treatments. Next, we will train a recurrent neural network on a longitudinal, time series dataset to predict individual responses to various periodontal treatments. Finally, we will integrate a large language model, fine- tuned on a large dataset of electronic dental records, with a language-vision model, trained using dental radiographs, to develop a generative artificial intelligence-based, provider-facing clinical decision-support system. To our knowledge, this is the first study to apply contemporary machine learning methods to high- resolution dental datasets to develop personalized periodontal care approaches. The findings will provide direct guidance for implementing precision treatments in clinical practice, potentially improving the oral and systemic health of millions of patients affected by periodontitis worldwide.
