Automated Dental Age Estimation from Panoramic Radiographs using an Interpretable Deep Learning Model.

Dental age estimation serves as a crucial foundation for forensic identification and pediatric dental assessment. Yet traditional methods, including the Demirjian and Willems standards, depend on subjective staging of tooth mineralization, resulting in observer variability and reduced applicability in routine practice. We developed an interpretable deep learning model trained on 7,207 panoramic radiographs from individuals aged 3.00–14.99 years. In five-fold cross-validation, the model achieved a mean absolute error (MAE) of 0.67 years, a root mean squared error (RMSE) of 0.87 years, and an R² of 0.92, outperforming benchmark deep architectures. Compared with expert assessments, the model yielded lower MAEs (0.78 vs. 0.92 years) and markedly shortened processing time. To enhance clinical reliability, the model self-assesses its confidence for each estimation, providing a predictive range. The 95% prediction intervals with a mean width of 2.76 years and 94.32% coverage. Gradient-weighted Class Activation Mapping confirmed biologically plausible attention to dental regions such as molars and root apices. The proposed model enables accurate, efficient, and interpretable dental age estimation with quantified uncertainty, supporting its potential application in forensic and clinical settings.