Oral Microbe Community and Pyramid Scene Parsing Network-based Periodontitis Risk Prediction

Periodontitis (PD) is a common chronic inflammatory disease affecting the gums and supporting tooth structures. It is often diagnosed only after significant irreversible tissue damage - such as gum recession and bone loss - has occurred, leading to tooth loss and systemic complications. Early detection of PD risk is therefore critical. This study integrates the Pyramid Scene Parsing Network (PSPNet), a deep learning model, with dental plaque microbial profiling data to generate a Periodontitis Risk Score (PRS) for identifying individuals at high risk of developing PD.

Microbial profiling data from dental plaque samples of 90 healthy controls (CON) and 514 PD patients were obtained from the Gene Expression Omnibus database (GSE32159). A preprocessing algorithm identified predictive indicators for PD and calculated actual PRS values (PRS_Actual) for both groups. The maximum theoretical PRS was set to '1' for clinically diagnosed PD patients and '0' for CON. The differential algorithm was embedded into PSPNet, which was trained using the generated dataset. The model's predictive ability was evaluated by comparing PSPnet-based PRS (PRS_PSPnet) with PRS_Actual.

After preprocessing, 27 indicators were identified for PD risk prediction. The PRS_Actual range ranged from 0.011 to 0.524 (mean 0.485) for CON and from 0.589 to 0.700 (mean 0.682) for PD patients, successfully distinguishing between the groups. The mean absolute error between PRS_PSPnet and PRS_Actual was 0.027, with an average computation time per sample of 10^-5 seconds, demonstrating both accuracy and efficiency.

By combining microbial profiling with PSPNet, this study offers a reliable, efficient, and noninvasive method for early screening of individuals at high risk of PD. This approach can help prevent irreversible periodontal damage, improve oral health, and reduce the associated health and economic burdens.