Automated Multi-Task Deep Learning Framework for Simultaneous Identification of Dental Pathologies and Arch Localization in Orthopantomograms (OPG)

Interpreting orthopantomograms (OPGs) accurately and efficiently remains challenging due to overlapping anatomical structures and the subtle presentation of many dental pathologies. While deep learning (DL) has shown considerable potential in dental imaging, most existing models focus on single-task detection and do not address the need for simultaneous analysis of both structural and pathological features within a unified system. To fill this gap, an automated multi-task deep learning framework is proposed that concurrently identifies ten dental pathology classes such as fractured roots and periodontally compromised teeth alongside arch-region classification of edentulous spans in OPGs. This method makes use of the YOLOv11 object detection architecture, which is powered by the C2PSA attention module and sophisticated post-processing algorithms that were trained on a larger collection of ten classes from the Dental OPG Kennedy dataset. With a mean Average Precision (mAP) 0.854 and an F1-score of 0.961 for every class assessed, the model demonstrated strong overall performance. Even in more complicated situations like root fractures (F1 = 0.928), the model was able to recognize well-defined edentulous areas with almost perfect sensitivity (Recall = 0.997) and good mAP. By integrating diagnostic analysis and anatomical techniques in a single optimized inference process, the YOLOv11 system offers interpretable and scalable solutions for comprehensive screening, thereby representing an advance for Computer-Assisted Dentistry Diagnostics.

@article{Elhag2026,
  title     = {Automated Multi-Task Deep Learning Framework for Simultaneous Identification of Dental Pathologies and Arch Localization in Orthopantomograms (OPG)},
  journal   = {International Journal of Advanced Computer Science and Applications},
  volume    = {17},
  number    = {7},
  year      = {2026},
  publisher = {The Science and Information Organization},
  author    = {Hatim Elhag and Norehan Binti Mokhtar and Siti Noor Fazliah Binti Mohd Noor and Gururajaprasad Kaggal Lakshmana Rao and Nordin Zakaria and Fakhitah Ridzuan and Mohammed Bin Salah},
  doi       = {10.14569/IJACSA.2026.0170714},
  url       = {https://doi.org/10.14569/IJACSA.2026.0170714}
}