J Oral Med Pain.
2024 Dec;49(4):109-117. 10.14476/jomp.2024.49.4.109.
Temporomandibular Joint Segmentation Using Deep Learning for Automated Three-Dimensional Reconstruction
- Affiliations
-
- 1Chonnam National University School of Dentistry, Gwangju, Korea
- 2Department of Dental Materials, Dental Science Research Institute, Chonnam National University School of Dentistry, Gwangju, Korea
- 3Department of Oral and Maxillofacial Radiology, Dental Science Research Institute, Chonnam National University School of Dentistry, Gwangju, Korea
- 4Department of Oral Medicine, Dental Science Research Institute, Chonnam National University School of Dentistry, Gwangju, Korea
- 5Department of Oral Medicine, Chonnam National University Dental Hospital, Gwangju, Korea
- KMID: 2563842
- DOI: http://doi.org/10.14476/jomp.2024.49.4.109
Abstract
- Purpose
Cone beam computed tomography (CBCT) is widely used to evaluate the temporomandibular joint (TMJ). For the three-dimensional (3D) assessment of the TMJ, segmentation of the mandibular condyle and articular fossa is essential. This study aimed to perform deep learning-based 3D segmentation of the mandibular condyle on CBCT images and evaluate the performance of the segmentation.
Methods
CBCT scan data from 99 patients (mean age: 53.3±19.2 years) diagnosed with TMJ disorders were analyzed. From the CBCT images, sagittal, coronal, and axial planes showing the mandibular condyle were selected and combined to form two-dimensional (2D) images. The U-Net deep learning model was used to exclusively segment the mandibular condyle area from the 2D images. From these results, 3D images of the mandibular condyle were reconstructed. Accuracy, precision, recall, and the Dice coefficient were calculated to appraise segmentation performance in each plane.
Results
The average Dice coefficient was 0.92 for the coronal and axial planes and 0.82 for the sagittal plane. The CBCT image-based segmentation performance of the mandibular condyle in the coronal and axial planes exceeded that in the sagittal plane. The sharpness and uniformity of the 2D images affected segmentation performance, with segmentation errors more likely occurring in non-uniform images. Certain segmentation errors were corrected through software processing. Finally, the segmented mandibular condyle images were applied to the CBCT data to reconstruct a 3D TMJ model.
Conclusions
Mandibular condyle 3D segmentation on CBCT images using U-Net may help evaluate and diagnose TMJ disorders. The proposed segmentation method may assist clinicians in efficiently analyzing CBCT images, particularly in cases involving anatomical abnormalities.
- Full Text Links
-
- Actions
-
Cited
- CITED
-
- Close
- Share
-
- Similar articles
-
- Automated Segmentation of Left Ventricular Myocardium on Cardiac Computed Tomography Using Deep Learning
- Deep Learning in Dental Radiographic Imaging
- Deep Learning-Based Computed Tomography Image Standardization to Improve Generalizability of Deep Learning-Based Hepatic Segmentation
- Three-dimensional (3D) auto-segmentation of vascular structures and hepatic sectional parenchyme of living liver donors using computed tomographic angiography: a deep learning model for automatic 3D volumetry
- Inter-observer and intra-observer reliability between manual segmentation and semi-automated segmentation for carotid vessel wall volume measurements on three-dimensional ultrasonography
