Multimodal Swin Transformer With Attention-Based Fusion of Panoramic and Alveolar Bone Loss Features for Age Estimation
IEEE Access · 2026 · 14 · 68650-68661
DOI: 10.1109/access.2026.3685318 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Access, 2026. Verfasst von Shaohua Tang und 6 weiteren Autor:innen. Die Arbeit nennt 26 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: ‘‘Regional Innovation System and Education (RISE)’’ through Gwangju RISE Center, Ministry of Education (MOE) and Gwangju Metropolitan Government, Republic of Korea.
Abstract (Originalsprache)
Dental age estimation is vital in forensic and clinical dentistry, particularly when valid dental records are unavailable. This study proposes a dual-channel Swin Transformer that combines global contextual features from panoramic dental radiographs (PDRs) with localized structural information from automatically extracted alveolar bone loss (ABL) maps through an attention-based fusion mechanism. The network integrates gated Feature-wise Linear Modulation (FiLM) and Convolutional Block Attention Module (CBAM) units to adaptively balance and refine cross-modal representations. Trained on 23,475 radiographs from individuals aged 10–99 years, the proposed framework achieved a mean absolute error (MAE) of 3.41 years and an R2 of 0.916, outperforming single-modality models (PDR-only 3.63, ABL-only 6.30) and representative baselines such as ResNet-50 and Swin-T. These results indicate that PDR and ABL provide complementary age-related information and that the proposed multimodal framework yields a consistent, albeit modest, improvement over single-modality models and simpler fusion baselines for large-scale panoramic radiograph-based age estimation.
Förderung
- ‘‘Regional Innovation System and Education (RISE)’’ through Gwangju RISE Center
- Ministry of Education (MOE) and Gwangju Metropolitan Government, Republic of Korea
Zitieren
APA
Tang, S., Kim, S., Ko, J., Kim, H., Lee, K., Han, D., & Jung, Y. (2026). Multimodal Swin Transformer With Attention-Based Fusion of Panoramic and Alveolar Bone Loss Features for Age Estimation. IEEE Access, 14, 68650-68661. https://doi.org/10.1109/access.2026.3685318
Vancouver
Tang S, Kim S, Ko J, Kim H, Lee K, Han D, et al. Multimodal Swin Transformer With Attention-Based Fusion of Panoramic and Alveolar Bone Loss Features for Age Estimation. IEEE Access. 2026;14:68650-68661. doi:10.1109/access.2026.3685318
BibTeX
@article{Tang2026,
author = {Tang, Shaohua and Kim, Seung-Min and Ko, Ji-Hyeong and Kim, Hyeong-Jae and Lee, Kyu-Hwan and Han, Dong-Hun and Jung, Young-Jin},
title = {Multimodal Swin Transformer With Attention-Based Fusion of Panoramic and Alveolar Bone Loss Features for Age Estimation},
journal = {IEEE Access},
year = {2026},
volume = {14},
pages = {68650--68661},
doi = {10.1109/access.2026.3685318}
}
EUVIMED