Predicting Tooth Mobility and Implant Stability using Periapical Radiographic Features and Implant Stability Test Data

Abstract

This study proposes a machine learning framework for predicting tooth mobility and implant stability by integrating anatomical features extracted from periapical radiographs with biomechanical measurements (IST values). A total of 407 annotated radiographs were expanded to 2,038 via geometric augmentation. Structural indices—such as head-to-root area ratios, periodontal ligament visibility, and root morphology—were engineered into composite features. A stacked ensemble model, incorporating LightGBM, XGBoost, and Random Forest with a Ridge Regression meta-learner, was trained on these features. The best-performing model achieved an R² of 0.6840, MAE of 4.0132, and MSE of 46.6392, demonstrating robust alignment between predicted and actual IST values. SHAP analysis revealed that root type and crown-root ratios were the most influential predictors. Although ligament annotations were sparse, their inclusion improved model accuracy in well-annotated cases. These findings highlight the potential of anatomy-aware, image-based regression models to non-invasively assess periodontal support and implant stability. The proposed framework bridges radiographic morphology and objective biomechanics, offering a reproducible, data-driven approach for clinical decision support in dentistry.