Physics-informed neural network-based simulation of pulmonary arterial hemodynamics abnormalities

Area of Science:

• Cardiovascular Engineering
• Computational Fluid Dynamics
• Artificial Intelligence in Medicine

Background:

• Ventricular septal defect (VSD) can cause abnormal pulmonary artery hemodynamics.
• Traditional Computational Fluid Dynamics (CFD) methods are computationally expensive.
• Acquiring comprehensive hemodynamic measurement data is challenging.

Purpose of the Study:

• To develop a Physics-Informed Neural Network (PINN) model for predicting pulmonary artery hemodynamics in VSD.
• To address the limitations of high computational cost and sparse data in traditional methods.
• To offer a novel, non-invasive approach for assessing pulmonary artery pressure.

Main Methods:

• A PINN model was trained using boundary conditions and limited clinical CFD data.
• Dynamic weighting factors were implemented to improve training efficiency.
• Model predictions of blood flow velocity and pressure were validated against CFD simulations.

Main Results:

• The PINN model accurately reproduced velocity and pressure fields with boundary conditions and sparse data.
• Velocity prediction errors (RMSE, MAE, MRE) were below 1.341%.
• Pressure prediction errors (RMSE, MAE, MRE) were below 5.679%, demonstrating reliable hemodynamic estimation.

Conclusions:

• PINN models leverage physical constraints to compensate for incomplete measurement data.
• This approach enables rapid and accurate prediction of pulmonary artery hemodynamics.
• The PINN model presents a promising non-invasive alternative for pulmonary artery pressure measurement.