Semi-supervised segmentation of teeth and root canals for bitewing X-ray images incorporating one-dimensional anatomical priors

Publication Type

Conference Proceeding Article

Publication Date

5-2026

Abstract

Segmentation of teeth and root canals in X-ray images is essential for evaluating Root Canal Treatments (RCT). However, it remains challenging due to occlusion, blurred boundaries, and limited labeled samples. To address this, we propose a semi-supervised method based on a Dual Teacher-Student model with One-dimensional Anatomical Priors (DTS-OAP). It effectively leverages unlabeled images by uti-lizing dual teachers to supervise dual-perturbation students for consistency learning and by incorporating a loss term that uses 1D anatomical priors to improve accuracy. Labeled prior consistency is achieved by extracting horizontal and vertical features through row and column transformations. We evaluated DTS-OAP on a dataset of X-ray images of RCT. With only 20% labeled data, it outperforms six state-of-the-art semi-supervised methods. In addition, it achieves the best performance in predicting the status of RCT.

Discipline

Artificial Intelligence and Robotics | Dentistry

Research Areas

Intelligent Systems and Optimization

Publication

Proceedings of the 2026 IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP), Barcelona, Spain, May 3-8

Identifier

10.1109/ICASSP55912.2026.11462988

Publisher

IEEE

City or Country

Piscataway, NJ

Additional URL

http://doi.org/10.1109/ICASSP55912.2026.11462988

This document is currently not available here.

Share

COinS