Construction Of Individual Tooth Prognostic Assessment Models For Periodontitis Patients

To construct precise individual tooth prognostic assessment models using machine learning algorithms for reliable clinical treatment decisions. Analyzed data from 1,340 teeth (590 non-molars, 750 molars) of 50 randomly selected patients out of 641 periodontitis cases. Six indicators, including probing depth (PD) and radiographic bone loss, were assessed. Risk factors were scored according to standards, and prognostic grades assigned. Machine learning algorithms like Gradient Boosting and Linear Regression were used to build specific models for non-molars and molars, optimized through cross-validation. Performance was evaluated using R² and RMSE. Gradient Boosting performed best in both models. The non-molar model achieved an R² of 0.997, and the molar model reached 0.999, with both having an RMSE of 0.15. “Tooth mobility” was identified as the most critical factor, while “probing depth” and “crown-root ratio (CRR)” also significantly impacted outcomes. Learning curves showed good generalization ability, with residuals concentrated around zero, indicating high fit between actual and predicted values. This study successfully constructed highly accurate and generalizable individual tooth prognostic models, supporting clinicians in treatment planning. Limitations include insufficient sample representativeness and lack of dynamic monitoring. Future work should expand sample size, incorporate more influencing factors, and optimize model algorithms to enhance clinical practice in dentistry.