A hierarchical age-aware framework for dental age estimation

Abstract

Dental Age Estimation (DAE) is a critical component of forensic odontology, yet many existing models struggle with performance stability across varying data scales. This study proposes a novel DAE framework that synergistically combines a Hierarchical Age-Aware Machine Learning Pipeline with a Hybrid Feature Approach. The methodology integrates deep convolutional features from EfficientNetV2-S with handcrafted texture descriptors including HOG, LBP, and GLCM. Evaluated on the extensive IPCTI dataset, the model achieved a high test accuracy of 0.9206. However, comparative experiments on the smaller OPG-800 dataset revealed a significant performance drop to an accuracy of 0.6206. These results confirm that while the hybrid hierarchical approach offers superior performance for large-scale dental repositories, the efficacy of hierarchical classification boundaries is deeply dependent on data volume.