Accurate Dental Age Estimation (DAE) is fundamental for forensic identification, particularly regarding legal age thresholds. This study is intended for forensic decision support using CBCT scans that are already available or acquired for justified clinical or medico-legal purposes, rather than advocating CBCT acquisition solely for age estimation. While Cone Beam Computed Tomography (CBCT) provides high-resolution volumetric data, current deep learning methods often rely on single-planar analysis, neglecting the anisotropic changes in the dental pulp cavity. We propose a multimodal deep fusion approach that integrates orthogonal cone-beam computed tomography (CBCT) views (coronal and sagittal) with clinical metadata to support forensic age assessment. Our dual-stream architecture, based on EfficientNet-V2, extracts visual features from maxillary central incisors and fuses them with biological sex and an automated Pulp-to-Tooth Ratio (PTR) index. Evaluated on the IPCTI dataset, the framework demonstrated a specialized performance profile: while the single-view model achieved superior global stability (MAE 5.49 years), the proposed multi-view fusion established a new state-of-the-art for the young adult demographic (18–32 years), reducing the MAE by up to 31.3% in the youngest cohort. Grad-CAM interpretability confirmed that the network targets biologically relevant markers, specifically the pulp chamber and cervical root canal. This approach advances automated DAE by providing a reproducible research prototype and benchmark evidence for CBCT-based DAE in forensic medicine settings.
Moreira et al. (Tue,) studied this question.