ADSeg: A flap-attention-based deep learning approach for aortic dissection segmentation.

Area of Science:

• Medical imaging analysis
• Computational pathology
• Cardiovascular imaging

Background:

• Accurate segmentation of aortic dissection (AD) is crucial for patient risk stratification and treatment planning.
• Existing methods often overlook the intimal flap, a key structure separating true and false lumens in AD.
• Incorporating long-range spatial information, particularly along the z-axis, can enhance segmentation accuracy.

Purpose of the Study:

• To develop an advanced method for accurate and rapid segmentation of aortic dissection (AD).
• To specifically address the challenge of segmenting the intimal flap in AD.
• To improve segmentation accuracy by leveraging long-distance z-axis information interaction.

Main Methods:

• Proposed a novel flap attention module to focus on critical intimal flap voxels using long-distance attention.
• Implemented a cascaded network architecture with feature reuse for enhanced representation power.
• Employed a two-step training strategy to optimize network performance.
• Evaluated the method on a multicenter dataset of 108 aortic dissection cases.

Main Results:

• The proposed ADSeg method achieved superior performance compared to state-of-the-art techniques for AD segmentation.
• Demonstrated significant improvements in segmentation accuracy and robustness across different imaging centers.
• The flap attention module effectively utilized long-distance z-axis information for better lumen and flap identification.

Conclusions:

• ADSeg offers a significant advancement in the automated segmentation of aortic dissection.
• The incorporation of intimal flap attention and long-distance interactions improves segmentation accuracy and clinical utility.
• The method shows promise for widespread adoption in clinical practice for AD management.