نشریه علمی پژوهشی طب انتظامی Journal of Police Medicine
Volume 14, Issue 1 (2025)
J Police Med 2025, 14(1) |
Back to browse issues page
Download citation:
BibTeX | RIS | EndNote | Medlars | ProCite | Reference Manager | RefWorks
Send citation to:
BibTeX | RIS | EndNote | Medlars | ProCite | Reference Manager | RefWorks
Send citation to:
Baghery S, Hosseini V. Design of an Artificial Intelligence System for Assessing the Spatial Relationship Between the Mandibular Third Molar and the Inferior Alveolar Nerve on Panoramic Radiographs.. , J Police Med. 2025; 14 (1) : e15
URL: http://jpmed.ir/article-1-1368-en.html
URL: http://jpmed.ir/article-1-1368-en.html
1- , bagheriss961@mums.ac.ir
English Extended Abstract: (31 Views)
Aims: The mandibular third molar, is the most frequently impacted tooth in the human dentition. Given the high prevalence of mandibular third molar impaction and the frequent need for surgical extraction, injury to the inferior alveolar nerve (IAN) is considered one of the most significant and serious postoperative complications, potentially leading to persistent sensory disturbances in the lower lip and chin. In this study, an artificial intelligence–based system was designed, implemented, and evaluated for the automatic assessment of the spatial relationship between the mandibular third molar (MTM) and the inferior alveolar nerve (IAN) on dental panoramic radiographs.
Materials and methods: The designed system consists of three main components: (1) semantic segmentation for anatomical structure localization, (2) binary classification to determine the presence or absence of MTM–IAN contact, and (3) an integrated inference pipeline that sequentially combines both models. In the segmentation stage, a U-Net architecture with a ResNet-34 encoder was employed, while the classification stage was based on an EfficientNet-B0 model adapted to accept a four-channel input (RGB image combined with the segmentation mask).
Results: The results demonstrated that the proposed integrated system achieved an overall accuracy of 92.57%, sensitivity of 88.31%, and specificity of 97.18%, indicating the high effectiveness of the segmentation-informed classification approach.
Conclusions: Overall, the findings suggest that anatomically informed deep learning models—particularly those integrating segmentation and classification—can serve as valuable decision-support tools in clinical dentistry.
Materials and methods: The designed system consists of three main components: (1) semantic segmentation for anatomical structure localization, (2) binary classification to determine the presence or absence of MTM–IAN contact, and (3) an integrated inference pipeline that sequentially combines both models. In the segmentation stage, a U-Net architecture with a ResNet-34 encoder was employed, while the classification stage was based on an EfficientNet-B0 model adapted to accept a four-channel input (RGB image combined with the segmentation mask).
Results: The results demonstrated that the proposed integrated system achieved an overall accuracy of 92.57%, sensitivity of 88.31%, and specificity of 97.18%, indicating the high effectiveness of the segmentation-informed classification approach.
Conclusions: Overall, the findings suggest that anatomically informed deep learning models—particularly those integrating segmentation and classification—can serve as valuable decision-support tools in clinical dentistry.
Article number: e15
Keywords: Artificial Intelegance, Lower mandibular third molar, Inferior Alveolar nerve, panoramic radiograph
Article Type: Original Research |
Subject:
Police Medicine Related Technologies
Received: 2026/05/26 | Accepted: 2026/09/10 | Published: 2026/10/27
Received: 2026/05/26 | Accepted: 2026/09/10 | Published: 2026/10/27
Send email to the article author
| Rights and permissions | |
 |
This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License. |