Daily Arxiv

This page organizes papers related to artificial intelligence published around the world.
This page is summarized using Google Gemini and is operated on a non-profit basis.
The copyright of the paper belongs to the author and the relevant institution. When sharing, simply cite the source.

U-Mamba2-SSL for Semi-Supervised Tooth and Pulp Segmentation in CBCT

Created by
  • Haebom

Author

Zhi Qin Tan, Xiatian Zhu, Owen Addison, Yunpeng Li

Outline

This paper proposes U-Mamba2-SSL, an automated algorithm for accurate tooth and pulp segmentation in Cone-Beam Computed Tomography (CBCT) images. This framework is based on the U-Mamba2 model and employs a multi-stage training strategy to effectively utilize unsupervised training data. Specifically, it improves performance by combining self-supervised pretraining, consistency regularization, and pseudo-labeling strategies. It achieved first place in Task 1 of the STSR 2025 Challenge, achieving a DSC of 0.917 and an average score of 0.789.

Takeaways, Limitations

•
Takeaways:
◦
Development of an automated CBCT image segmentation algorithm to increase clinical applicability.
◦
Proposing a novel semi-supervised learning framework based on the U-Mamba2 model.
◦
Utilize effective learning strategies such as self-supervised pre-training, consistency regularization, and pseudo-labeling.
◦
Proven performance by achieving first place in STSR 2025 Challenge Task 1.
◦
Increase reproducibility and usability of research by making code open via GitHub.
•
Limitations:
◦
Lack of detailed information about the specific dataset composition and learning environment.
◦
Further research is needed on the generalization performance of the algorithm.
◦
Possibility of error propagation due to pseudo-labeling strategy.
👍
Made with Slashpage