Semantic Decomposition Network With Contrastive and Structural Constraints for Dental Plaque Segmentation

Area of Science:

• Medical Imaging
• Computer Vision
• Biomedical Engineering

Background:

• Dental plaque segmentation is crucial for diagnosis and treatment planning.
• Existing methods struggle with semantic-blur regions and complex shape variations.
• Accurate segmentation of teeth and dental plaque remains a significant challenge.

Purpose of the Study:

• To develop an advanced method for accurate dental plaque segmentation.
• To address limitations of current segmentation techniques, particularly in handling ambiguous boundaries.
• To improve diagnostic information derived from medical reagent-stained dental images.

Main Methods:

• Proposed a Semantic Decomposition Network (SDNet) with two single-task branches for teeth and dental plaque segmentation.
• Introduced a contrastive constraint module (CCM) to learn discriminative features and reduce semantic-blur impact.
• Incorporated a structural constraint module (SCM) for improved segmentation of varied plaque shapes.
• Developed and utilized the large-scale Stained Dental Plaque Segmentation dataset (SDPSeg).

Main Results:

• SDNet demonstrated superior performance in segmenting dental plaque compared to existing methods.
• The proposed constraint modules effectively enhanced category-specific feature learning.
• The divide-and-conquer approach successfully decoupled the relationship between teeth and plaque segmentation.
• Achieved state-of-the-art results on the SDPSeg dataset.

Conclusions:

• SDNet offers a robust solution for accurate dental plaque segmentation.
• The method effectively handles challenges posed by semantic-blur regions and complex shapes.
• This advancement holds significant potential for improving dental diagnostics and treatment planning.