Towards clinically interpretable and externally validated models for chronological age estimation from panoramic radiographs

Radiographic dental age estimation is widely used in clinical, legal and forensic contexts, particularly when reliable documentary evidence is unavailable. However, existing methods remain highly heterogeneous, differing in their target age ranges, radiographic requirements, population specificity, observer dependence and reported accuracy. Although traditional manual approaches have been extensively evaluated, automated methods based on machine learning and deep learning still require stronger external validation, greater interpretability and more consistent assessment across populations and imaging systems. This doctoral thesis addresses these limitations through a combined research approach comprising a systematic review of radiographic dental age-estimation methods and the development and external validation of interpretable predictive models for individuals aged 6 to under 16 years. Using automated tooth-specific measurements from panoramic radiographs, the study identifies mandibular second molar height as the most informative predictor, achieving external validation errors of approximately one year. The results support the use of simple, robust and biologically grounded models, while highlighting the need for further validation in geographically and clinically diverse populations.

Palabras clave: Dental age estimation, chronological age estimation, Explainable Machine Learning,