Abstract
Objective:
This longitudinal study aimed to evaluate how artificial intelligence (AI) models could predict the long‑term effects of teeth whitening products on enamel health among patients treated at tertiary dental hospitals in Lahore and Islamabad, Pakistan.
Methods:
We collected detailed clinical and demographic data from 500 participants over 24 months who underwent professional or at‑home teeth whitening between January 2023 and December 2024. Enamel health was assessed at baseline, 6, 12, 18, and 24 months via indices including enamel surface roughness, mineral density (measured with quantitative light‑induced fluorescence), and standardized visual scoring. AI predictive models — including random forests, support vector machines (SVM), and deep neural networks — were trained on both clinical and lifestyle predictors to estimate enamel degradation risk scores over long‑term follow‑up. Model performance was evaluated using cross‑validation accuracy, F1 score, and receiver operating characteristic (ROC) curves.
Results:
Predictive models exhibited high performance in estimating enamel changes over time, with the best‑performing model achieving an ROC area under the curve (AUC) of 0.88 (95% CI: 0.85–0.91), precision of 82%, and recall of 79%. AI predictions displayed statistically significant correlations with observed clinical enamel parameters (p < 0.01), and identified key risk features such as bleaching agent concentration, application frequency, baseline enamel microhardness, and dietary acid exposure.
Conclusion:
AI‑based modeling accurately predicted long‑term enamel changes following whitening treatments and identified modifiable clinical and lifestyle risk factors. Implementation of such predictive systems can aid clinicians in personalizing whitening protocols and mitigating adverse enamel outcomes. Future research should broaden data sources and validate predictive tools in diverse populations.