Precise dental caries segmentation in X-rays with an attention and edge dual-decoder network.

Area of Science:

• Medical Image Analysis
• Deep Learning
• Computational Dentistry

Background:

• Accurate segmentation of dental caries is crucial for early detection and treatment.
• Existing deep learning methods face challenges in precisely segmenting complex caries boundaries.

Purpose of the Study:

• To propose a novel deep learning network, AEDD-Net, for enhanced caries boundary segmentation.
• To improve the accuracy and clinical applicability of automated caries detection.

Main Methods:

• AEDD-Net integrates atrous spatial pyramid pooling and cross-coordinate attention for feature fusion.
• A dedicated boundary generation module and an innovative boundary loss function are introduced.
• The network combines an attention mechanism with a dual-decoder structure.

Main Results:

• AEDD-Net significantly outperforms existing networks in Dice coefficient, Jaccard similarity, precision, and sensitivity.
• The proposed network demonstrates superior performance specifically in segmenting caries boundaries.
• Experimental results validate the effectiveness of the boundary generation module and loss function.

Conclusions:

• AEDD-Net offers an innovative and effective approach for automated caries segmentation.
• The method shows significant potential for improving clinical applications in dental diagnostics.
• Enhanced boundary segmentation accuracy can lead to better treatment outcomes for dental caries.