Interference-Free Causality Learning Promotes Cross-Level, Fine-Grained Diagnosis of Coronary Artery Disease in...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Cardiovascular Disease Research

Background:

• Coronary artery disease (CAD) poses a significant global health challenge.
• Automated diagnosis using coronary CT angiography (CCTA) shows promise but is limited by heterogeneous plaque/stenosis features and confounding factors.
• Existing methods struggle with fine-grained, cross-level CAD diagnosis in real-world clinical settings.

Purpose of the Study:

• To introduce the Attribute-Decoupled Intervention Network (ADI-Net), a confounder-free framework for fine-grained CAD diagnosis.
• To enable precise analysis at both artery and patient levels, enhancing clinical applicability.
• To develop a robust CAD diagnostic tool that overcomes limitations of current automated techniques.

Main Methods:

• Developed ADI-Net, a novel framework utilizing an attribute-decoupled representation with differential constraints for heterogeneous feature capture.
• Implemented dynamic-updating causal intervention with Do-expression for refined confounder management and cross-level assessments.
• Validated the framework on CCTA datasets from three clinical centers.

Main Results:

• ADI-Net demonstrated superior performance compared to state-of-the-art methods in cross-level, fine-grained CAD diagnosis.
• The framework exhibited enhanced robustness, domain adaptability, and data efficiency in experimental evaluations.
• ADI-Net successfully addressed challenges posed by stenosis heterogeneity and confounding variables in CCTA analysis.

Conclusions:

• ADI-Net offers a robust and effective solution for confounder-free, fine-grained CAD diagnosis from CCTA.
• The proposed approach significantly improves diagnostic accuracy and clinical applicability for cardiovascular disease.
• ADI-Net represents a advancement in automated medical image analysis for complex diseases like CAD.