AI auto-positioning of the dental arch in panoramic radiography
A Convolutional Neural Network-Based Auto-Positioning Method for Dental Arch in Rotational Panoramic Radiography
Authors: Xin Du, Yi Chen, Jun Zhao, Yan Xi
Summary
Dental panoramic radiography (DPR) is one of the most common diagnostic techniques in dentistry but is highly sensitive to patient positioning errors, especially in the anterior dental arch, which frequently causes image blurring and loss of diagnostic detail.
This study proposes a CNN-based auto-positioning method designed to estimate the deviation of the patient’s dental arch and correct it before image reconstruction.
The method operates in four main steps:
- Initial reconstruction of the DPR ...
- Extraction and normalization of the region of interest (ROI) for CNN processing.
- Estimation of the forward–backward translational deviation via CNN regression.
- Reconstruction of a new panoramic image after correcting the estimated deviation.
A simulated dataset containing 5,166 image–deviation pairs (±20 mm range) was used to train four CNN architectures (13–15 layers).
Models achieved mean absolute errors < 1.0 mm and maximum errors < 2.7 mm within the training deviation range, showing high stability across different dental-arch morphologies.
Iterative correction further improved image sharpness in cases of severe mis-positioning or morphological variation.
This approach substantially reduces blurring artifacts in the anterior region and produces consistent DPR quality suitable for automatic reconstruction and clinical diagnosis.
DOI: https://doi.org/10.1109/embc.2018.8512732
Key Words
Panoramic Radiography, Auto-Positioning, CNN, Image Reconstruction, Dental Arch Correction, Deep Learning, Image Quality Enhancement, Computer Vision, Radiographic Deblurring
Extracted Data
- Year: 2018
- Modality: Panoramic Radiography
- Dataset: 5,166 simulated pairs
- Dataset Split: 98 % training (5,062) / 1 % validation (52) / 1 % test (52)
- Network Architecture: CNN
- Metrics: Mean Squared Error (MSE); Mean Absolute Error (MAE); Max Error (Max AE)
- AP – Strategy: Not applicable
- AP – Professional Qty: Not applicable (simulated data only)
- AP – Supervisor Presence: Not applicable
- AP – Experience Level: Not applicable
- AP – Expertise Area: Not applicable
- AP – Tool or System: Custom panoramic simulation software
- Task: Regression (Estimation of Dental Arch Deviation)
- Project Objective: To automatically estimate and correct dental-arch positioning errors in rotational panoramic radiography to reduce blur and improve diagnostic quality
Clinical Relevance
- Clinical importance: Precise positioning in panoramic radiography is crucial for image clarity.
- Innovation: This AI-driven method introduces a fully automated solution for detecting and correcting positional errors before reconstruction, reducing blurring in the anterior region and enhancing diagnostic readability.
- Practical impact: By integrating CNN-based auto-positioning into panoramic systems, clinicians can achieve consistent image quality with minimal manual adjustment, improving workflow efficiency and reducing repeat exposures.
Related Video
▶️ Watch on YouTube
