Related Experiment Video
Updated: Jun 29, 2025
05:49
Author Spotlight: Advancing CBCT and Digital Dental Image Integration with AI-Assisted Digitization
Published on: February 23, 2024
834
Deep learning-based automatic segmentation of the mandibular canal on panoramic radiographs: A multi-device study
Moe Thu Zar Aung1,2, Sang-Heon Lim3, Jiyong Han3
1Department of Oral and Maxillofacial Radiology, School of Dentistry and Dental Research Institute, Seoul National University, Seoul, Korea.
Imaging Science in Dentistry
|April 4, 2024
Summary
A deep learning model effectively detects the mandibular canal in dental panoramic radiographs, achieving over 88% accuracy. This convolutional neural network approach shows promise for automated segmentation across different imaging machines.
Area of Science:
- Radiology
- Artificial Intelligence
- Medical Imaging
Background:
- Accurate identification of the mandibular canal is crucial for dental procedures.
- Current manual segmentation methods can be time-consuming and subjective.
- Deep learning offers potential for automated and precise analysis of dental radiographs.
Purpose of the Study:
- To develop and evaluate a deep learning model for automated mandibular canal detection.
- To assess the performance of convolutional neural networks (CNNs) using U-Net architecture for segmentation.
- To investigate the impact of data from multiple panoramic radiograph machines on model performance.
Main Methods:
- Collected 2,100 panoramic radiographs from three different machines (RAYSCAN Alpha, OP-100, CS8100).
- Used a U-Net based CNN architecture for automated segmentation of mandibular canals.
- Trained seven independent networks using various combinations of data from the three machines and evaluated on a hold-out test set.
Main Results:
- The network trained on all three machine groups achieved a Dice Similarity Coefficient (DSC) of 88.9%.
- Networks trained on two-group combinations also showed high performance, with DSCs ranging from 87.8% to 88.4%.
- The best performing network demonstrated high precision (90.6%) and recall (87.4%).
Conclusions:
- Deep learning models, specifically CNNs with U-Net, can achieve excellent performance for mandibular canal segmentation on panoramic radiographs.
- A DSC exceeding 88% indicates the robustness of the proposed deep learning approach.
- Model development should consider the characteristics of radiographs from different devices, not just dataset size, for optimal performance.
