Geometric deep learning-based coronary wall shear stress estimation from real-world patients.

Area of Science:

• Cardiovascular Imaging
• Artificial Intelligence in Medicine
• Biomedical Engineering

Background:

• Coronary wall shear stress (WSS) offers prognostic value but is computationally intensive.
• Current methods hinder clinical translation due to complexity and long computation times.
• Geometric deep learning (GEM-GCN) offers a potential solution for rapid WSS estimation.

Purpose of the Study:

• To evaluate a GEM-GCN framework for estimating coronary WSS directly from patient angiography.
• To compare GEM-GCN WSS estimation accuracy against traditional CFD methods.
• To assess if GEM-GCN WSS can predict myocardial infarction (MI) risk.

Main Methods:

• Reconstructed 1078 coronary arteries from 748 patients' invasive angiography.
• Trained and tested GEM-GCN using CFD-derived time-averaged WSS as reference labels.
• Conducted random and clinical data splits to evaluate predictive performance for MI.

Main Results:

• GEM-GCN generated patient-specific WSS maps in under 5 seconds.
• Achieved high spatial agreement in high-WSS regions (Dice 0.88) and comparable MI prediction performance.
• Normalized GEM-GCN WSS showed strong correlation (R=0.89) with CFD-derived WSS.

Conclusions:

• Geometric deep learning enables rapid, CFD-free coronary WSS estimation.
• This approach facilitates large-scale, real-world cardiovascular risk stratification.
• Supports clinical translation of WSS analysis for improved patient outcomes.