Pose-aware instance segmentation framework from cone beam CT images for tooth segmentation.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Dental Radiology

Background:

• Accurate individual tooth segmentation in cone beam computed tomography (CBCT) is crucial for dental applications.
• Metal artifacts in CBCT images present a significant challenge to precise tooth segmentation.
• Existing methods struggle with the accuracy of tooth segmentation in the presence of artifacts.

Purpose of the Study:

• To develop a novel neural network framework for robust individual tooth segmentation in CBCT images.
• To address the challenge of metal artifacts impacting segmentation accuracy.
• To improve the precision of tooth segmentation for orthodontic and implantology applications.

Main Methods:

• A three-step approach involving pose-aware image cropping and realignment using pose regression neural networks.
• Metal-robust individual tooth detection utilizing a convolutional detector with enhanced region proposal network.
• Instance segmentation via a convolutional neural network (CNN) converting pixel-wise labeling to distance regression, augmented with metal-intensive image processing.

Main Results:

• The proposed method demonstrated superior performance compared to state-of-the-art algorithms, particularly for teeth affected by metal artifacts.
• Achieved 5.68% higher F1 score and 30.30% higher aggregated Jaccard index compared to existing methods.
• Successfully improved the accuracy and robustness of individual tooth segmentation in challenging CBCT datasets.

Conclusions:

• The developed method offers a significant advancement in accurate tooth segmentation from CBCT images, especially in the presence of metal artifacts.
• The pose-aware realignment and metal-robust CNN framework provide a reliable solution for dental imaging analysis.
• This technique has substantial implications for improving the accuracy of dental treatment planning and simulations.