Related Experiment Video
Updated: Jul 12, 2026
06:45
Automated Joint Space Detection Improves Bone Segmentation Accuracy
Published on: November 28, 2025
Automated multi-orientation (Plane-wise) segmentation of TMJ structures using deep learning: a dual-center study.
Wenqi Yang1, Yaqiong Zhang1, Yunying Zhao2
1Department of Radiology, Shanghai Stomatological Hospital, Fudan University, Shanghai, China; Shanghai Key Laboratory of Craniomaxillofacial Development and Diseases, Fudan University, Shanghai, China.
Summary
A novel deep learning framework accurately segments temporomandibular joint (TMJ) structures from multi-view MRI scans. This automated approach, particularly nnU-Net, enhances efficiency and reduces subjectivity in TMJ analysis.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Anatomy
Background:
- Temporomandibular joint (TMJ) disorders affect millions globally.
- Manual segmentation of TMJ structures in MRI is time-consuming and subjective.
- Quantitative TMJ analysis requires precise and reproducible segmentation.
Purpose of the Study:
- To develop and validate a deep learning framework for automated multi-view MRI segmentation of TMJ structures.
- To compare the performance of different deep learning models for this task.
- To assess the framework's accuracy, efficiency, and reproducibility.
Main Methods:
- Utilized a two-center dataset of 121 patients (242 TMJs) with oblique coronal/sagittal MRI sequences.
- Trained and tested four deep learning models: 2D U-Net, 3D U-Net, nnU-Net, and Swin-Unet.
- Evaluated segmentation performance using Dice similarity coefficient and intraclass correlation coefficients (ICCs).
Main Results:
- All tested models demonstrated high agreement with manual segmentations (ICC > 0.90).
- nnU-Net achieved superior performance, with ICC > 0.99 and high Dice scores for the condyle (0.949) and disc (0.865).
- Segmentation performance remained consistent across different imaging planes with minimal error (<5%).
Conclusions:
- The nnU-Net-based framework provides accurate, efficient, and robust multi-view TMJ segmentation.
- Dual-center validation supports the technical reproducibility of the framework.
- Further validation across multiple scanners and with clinical outcomes is needed to establish broader generalizability and utility.
