Future cardiovascular events prediction from invasive coronary angiography: A graph representation learning...

Area of Science:

• Cardiovascular Medicine
• Artificial Intelligence in Healthcare
• Medical Imaging Analysis

Background:

• Coronary artery disease (CAD) is a leading cause of death globally, necessitating improved risk stratification.
• Current methods for predicting future cardiovascular events from invasive coronary angiography (ICA) are insufficient.
• There is an urgent need for innovative approaches to leverage ICA data for early event prediction.

Purpose of the Study:

• To develop and evaluate AngioGraphCAD, a novel graph neural network framework for patient-level prediction of future cardiovascular events using ICA.
• To utilize coronary artery geometry and clinical data for enhanced prediction accuracy.
• To create a flexible and interpretable model for personalized CAD management.

Main Methods:

• AngioGraphCAD constructs geometric graph representations for individual coronary lesions.
• It learns lesion-level embeddings by integrating coronary artery geometry with clinical data.
• A masked-attention mechanism fuses lesion-level embeddings into a unified patient-level representation.

Main Results:

• AngioGraphCAD demonstrated strong lesion-level performance with AUCs of 0.71 (FAME2) and 0.73 (FCL), outperforming existing methods.
• The framework achieved an AUC of 0.70 for patient-level prediction in the FAME2 cohort.
• The study highlights the predictive power of coronary artery geometry in ICA.

Conclusions:

• Coronary artery geometry from ICA holds significant predictive value for future cardiovascular events.
• Geometry-aware representations offer a promising avenue for personalized CAD management.
• AngioGraphCAD provides a robust framework for improving risk stratification in CAD patients.