ACE-ProtoNet: Adaptive covariance eigen-gate and uncertainty-aware prototype learning for coronary artery...

Area of Science:

• Medical Imaging
• Computer Vision
• Artificial Intelligence

Background:

• Accurate segmentation of coronary arteries from Coronary CT Angiography (CCTA) is crucial for diagnosing cardiovascular diseases.
• Challenges include low contrast, anatomical variability, and complex vessel structures, hindering automated segmentation.

Purpose of the Study:

• To introduce ACE-ProtoNet, a novel framework for robust and accurate coronary artery segmentation.
• To address limitations of existing automated segmentation methods in CCTA.

Main Methods:

• A parallel dual-encoder backbone combining a Vision Foundation Model (VFM) and a CNN.
• An Adaptive Covariance Eigen-Gate (ACE-Gate) for feature integration.
• An Uncertainty-aware Prototype Learning Head (UPL-Head) for enhanced representation.

Main Results:

• ACE-ProtoNet achieved superior performance compared to twelve state-of-the-art methods across multiple metrics.
• Demonstrated strong cross-domain generalization, cross-modality, and cross-anatomy transferability.
• Significantly improved segmentation accuracy in challenging regions.

Conclusions:

• ACE-ProtoNet offers a robust and accurate solution for coronary artery segmentation in CCTA.
• The framework shows promise for advancing quantitative stenosis evaluation and surgical planning.
• The proposed methods enhance feature integration and representation learning for medical image analysis.