Mask-Transformer-Based Networks for Teeth Segmentation in Panoramic Radiographs

Area of Science:

• Dentistry
• Medical Imaging
• Computer Vision

Background:

• Accurate teeth segmentation is crucial for dental diagnosis and treatment planning.
• Traditional methods often neglect the broader oral tissue context, limiting segmentation accuracy.
• Existing segmentation techniques require improvement in performance and robustness.

Purpose of the Study:

• To propose a novel panoptic-segmentation-based method for teeth segmentation.
• To develop a dual-path transformer-based network architecture for instance teeth segmentation.
• To enhance the consideration of oral tissue context in dental image analysis.

Main Methods:

• A panoptic segmentation approach combining instance and semantic segmentation.
• A novel dual-path transformer network architecture for instance teeth segmentation.
• Integration of a panoptic quality (PQ) loss function for streamlined training.
• Utilizing pixel-to-memory feedback, pixel-to-pixel self-attention, and memory-to-pixel/memory-to-memory self-attention mechanisms.

Main Results:

• The proposed model achieved substantial improvements in teeth segmentation performance on the UFBA-UESC Dental Image dataset.
• The method surpassed existing state-of-the-art techniques in segmentation accuracy and robustness.
• The dual-path transformer architecture effectively integrated multi-scale features and facilitated bi-directional communication.

Conclusions:

• The developed panoptic segmentation method represents a significant advancement in teeth segmentation technology.
• This approach offers a more comprehensive understanding of oral structures by including surrounding tissues.
• The findings contribute to improved diagnostic capabilities and treatment planning in dentistry.