Deep vectorised operators for pulsatile hemodynamics estimation in coronary arteries from a steady-state prior

Area of Science:

• Cardiovascular hemodynamics
• Computational fluid dynamics
• Machine learning in medicine

Background:

• Cardiovascular hemodynamic fields are crucial for diagnosing coronary artery disease.
• Computational fluid dynamics (CFD) offers accurate, non-invasive in silico evaluation.
• Current methods require time-consuming simulations.

Purpose of the Study:

• To develop a time-efficient machine learning surrogate model for estimating pulsatile hemodynamics.
• To leverage steady-state priors for improved computational efficiency.
• To enable faster, accurate hemodynamic analysis in coronary arteries.

Main Methods:

• Introduction of deep vectorised operators, a novel modeling framework.
• Utilizing a neural field architecture conditioned on hemodynamic boundary conditions.
• Employing permutation-equivariance, message passing, and self-attention mechanisms for parameterization.
• Validation on 74 stenotic coronary arteries using patient-specific CFD simulations.

Main Results:

• The proposed model accurately estimates pulsatile velocity and pressure with a low approximation disparity (0.368 ± 0.079).
• The model demonstrates discretization independence, being agnostic to source domain re-sampling (p<0.05).
• Achieved significant time efficiency compared to traditional CFD.

Conclusions:

• Deep vectorised operators represent a powerful tool for cardiovascular hemodynamics estimation.
• The approach is applicable to coronary arteries and potentially other vascular regions.
• Enables rapid and accurate hemodynamic analysis, supporting clinical decision-making.