Related Experiment Video
Updated: Jan 14, 2026
10:23
Author Spotlight: Three-Dimensional Cephalometric Landmark Annotation Demonstration on Human Cone Beam Computed Tomography Scans
Published on: September 8, 2023
3.6K
CeLR: A Transformer-Based Regression Network for Accurate Cephalometric Landmark Detection in High-Resolution X-Ray
IEEE Transactions on Medical Imaging
|January 12, 2026
Summary
A new Transformer-based network, Cephalometric Landmark Regression (CeLR), accurately locates landmarks on X-ray images for orthodontic analysis. This efficient method achieves state-of-the-art results with lower computational cost.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Orthodontics
Background:
- Accurate cephalometric landmark localization is crucial for automated orthodontic analysis.
- Current methods face challenges with high computational demands and complex pipelines, limiting end-to-end optimization.
Purpose of the Study:
- To introduce an end-to-end Transformer-based network, Cephalometric Landmark Regression (CeLR), for precise landmark localization on high-resolution X-ray images.
- To improve accuracy and efficiency in cephalometric analysis.
Main Methods:
- Developed CeLR, an end-to-end Transformer network utilizing a feature extractor, reference encoder, and finetune decoder with cross-attention.
- Incorporated a denoising module to enhance model robustness.
- Evaluated on public cephalometric datasets.
Main Results:
- CeLR achieved state-of-the-art performance, with a Mean Radial Error (MRE) of 0.98 mm and a 2 mm Success Detection Rate (SDR) of 89.82% on the ISBI 2015 Challenge Test1 dataset.
- Demonstrated a computational cost of 91.3 GFLOPs, balancing accuracy and efficiency.
- Showcased effectiveness and clinical potential.
Conclusions:
- The proposed CeLR network offers a highly effective and efficient solution for cephalometric landmark regression.
- CeLR presents significant clinical potential for automated orthodontic analysis.
- The Transformer-based approach enables end-to-end optimization, overcoming limitations of existing methods.
