IADR Abstract Archives
Identify Factors Associated With Zirconia Failures Using Machine Learning
Objectives: Zirconia, a type of ceramic material, is widely used in dentistry for prostheses like crowns, bridges, and implants. Studies show zirconia crowns have high survival rates, typically 90% to 95% over 5-10 years, making them reliable for restorations. Failures are generally associated with specific clinical circumstances rather than material flaws. Little evidence exists to determine the factors responsible for zirconia failure. This study aimed to utilize Artificial Intelligence (AI) methods, including machine learning (ML), to identify factors contributing to zirconia crown failure.
Methods: Electronic dental records from patients treated at Temple University Kornberg School of Dentistry between 2021-2024 were analyzed, and two cohorts were created (patients with failure versus success) for ML analysis. Several ML models, such as Support Vector Machine, Naïve Bayes, Random Forest, XGBoost, and Light Gradient Boosting Machine (LGBM) were applied using a 80% training and 20% testing dataset split. Due to an imbalance in the dataset, over-sampling and under-sampling methods were employed to adjust the data. The ML performance was assessed through metrics such as accuracy, F1-score, precision, recall, sensitivity, specificity, and Brier score. SHapley Additive exPlanations (SHAP) values were used for interpretable ML to explain the impact of each feature on the model’s predictions.
Results: The cohort consisted of 3,146 zirconia crowns with no failure and 84 crowns with failure. The ML model achieved moderate to high accuracy of 81% F-1 score, 95% precision, and 71% recall. Factors contributing to zirconia crown failures included Decayed, Missed, and Filled Surfaces (DMFS), poor dental prognosis , younger age, teeth clenching/grinding, infrequent brushing, lack of insurance, more periodontal treatments, high ASA, Cardiology and Rheumatology diseases, and high caries risk.
Conclusions: The AI-driven model produced promising results in identifying factors contributing to zirconia crown failure. This information can be used chairside for patient education and for clinical decision-making once the study findings are validated with larger datasets. Although ML applications in prosthodontics are limited, further studies are needed to explore long-term clinical outcomes.
Methods: Electronic dental records from patients treated at Temple University Kornberg School of Dentistry between 2021-2024 were analyzed, and two cohorts were created (patients with failure versus success) for ML analysis. Several ML models, such as Support Vector Machine, Naïve Bayes, Random Forest, XGBoost, and Light Gradient Boosting Machine (LGBM) were applied using a 80% training and 20% testing dataset split. Due to an imbalance in the dataset, over-sampling and under-sampling methods were employed to adjust the data. The ML performance was assessed through metrics such as accuracy, F1-score, precision, recall, sensitivity, specificity, and Brier score. SHapley Additive exPlanations (SHAP) values were used for interpretable ML to explain the impact of each feature on the model’s predictions.
Results: The cohort consisted of 3,146 zirconia crowns with no failure and 84 crowns with failure. The ML model achieved moderate to high accuracy of 81% F-1 score, 95% precision, and 71% recall. Factors contributing to zirconia crown failures included Decayed, Missed, and Filled Surfaces (DMFS), poor dental prognosis , younger age, teeth clenching/grinding, infrequent brushing, lack of insurance, more periodontal treatments, high ASA, Cardiology and Rheumatology diseases, and high caries risk.
Conclusions: The AI-driven model produced promising results in identifying factors contributing to zirconia crown failure. This information can be used chairside for patient education and for clinical decision-making once the study findings are validated with larger datasets. Although ML applications in prosthodontics are limited, further studies are needed to explore long-term clinical outcomes.