STS-TransUNet: Semi-supervised Tooth Segmentation Transformer U-Net for dental panoramic image

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Dental panoramic image segmentation is vital for accurate diagnosis and treatment planning in dentistry.
• Traditional methods struggle with integrating global and local contextual information and utilizing unlabeled data.
• This limits the performance and generalizability of segmentation models in diverse clinical scenarios.

Purpose of the Study:

• To introduce a novel deep learning method for dental panoramic image segmentation.
• To address limitations of traditional methods in handling contextual information and unlabeled data.
• To enhance the accuracy and robustness of tooth segmentation for clinical applications.

Main Methods:

• Developed an advanced TransUNet architecture with direct input-output layer connections for improved feature utilization.
• Incorporated spatial and channel attention mechanisms in decoder segments for targeted region focus.
• Implemented deep supervision techniques for efficient training and a self-learning algorithm using unlabeled data to boost generalization.

Main Results:

• The proposed Semi-supervised Tooth Segmentation Transformer U-Net (STS-TransUNet) demonstrated superior performance.
• Achieved high effectiveness and robustness in tooth segmentation tasks on the MICCAI STS-2D dataset.
• The method successfully combined global and local context and leveraged unlabeled data.

Conclusions:

• The STS-TransUNet offers a significant advancement in automated dental image analysis.
• The model's ability to utilize unlabeled data enhances its applicability in real-world clinical settings.
• This deep learning approach holds promise for improving diagnostic accuracy and treatment planning in orthodontics and oral medicine.