Cross-center Model Adaptive Tooth segmentation.

Area of Science:

• Digital Dentistry
• Medical Image Analysis
• Artificial Intelligence in Healthcare

Background:

• Automatic 3D tooth segmentation from intraoral scans (IOS) is crucial for computer-aided orthodontic treatments.
• Deploying AI models across medical centers faces challenges due to data distribution shifts and data privacy concerns, hindering model re-training or fine-tuning.

Purpose of the Study:

• To propose a novel framework, Cross-center Model Adaptive Tooth segmentation (CMAT), for adapting pre-trained tooth segmentation models to new centers without data sharing or additional annotations.
• To address performance degradation caused by data distribution shifts in cross-center scenarios.

Main Methods:

• CMAT adapts source-center models to target centers using a tooth-level prototype alignment module, a progressive pseudo-labeling transfer module, and a tooth-prior regularized information maximization module.
• The framework is designed for source-data-free, multi-source-data-free, and test-time cross-center adaptation scenarios.

Main Results:

• CMAT consistently outperformed existing baseline methods across three cross-center scenarios on two datasets.
• Extensive ablation studies and statistical analysis validated the effectiveness of the proposed approach.

Conclusions:

• CMAT offers an effective and privacy-preserving solution for model adaptive tooth segmentation in digital dentistry.
• The framework demonstrates significant potential for real-world deployment in clinical settings with diverse data distributions.