Physics-Informed Neural Network for Modeling the Pulmonary Artery Blood Pressure from Magnetic Resonance Images: A...

Area of Science:

• Cardiovascular imaging and modeling
• Computational fluid dynamics
• Artificial intelligence in medicine

Background:

• Pulmonary arterial pressure is crucial for diagnosing cardiovascular and pulmonary diseases.
• Current gold standard measurement via right heart catheterization is invasive and risky.
• Non-invasive techniques using patient-specific imaging are needed.

Purpose of the Study:

• To develop and implement a Physics-Informed Neural Network (PINN) model for estimating pulmonary arterial pressure.
• To predict pressure, velocity, and area variations in the pulmonary artery bifurcation.
• To validate the non-invasive approach using 4D Flow MRI data.

Main Methods:

• A PINN model was implemented integrating 1D reduced Navier-Stokes physics.
• The model utilized velocity and area measurements from 4D Flow MRI in the pulmonary artery bifurcation.
• Training incorporated penalties for flow and momentum conservation.

Main Results:

• The PINN model successfully estimated pulmonary arterial pressure in a healthy patient.
• Pressure estimates were consistent with literature values, showing a mean arterial pressure of 21.5 mmHg.
• The model predicted pressure, velocity, and area variations throughout the bifurcation.

Conclusions:

• PINNs offer an innovative, non-invasive alternative to catheterization for pulmonary arterial pressure estimation.
• This is the first study applying PINNs to estimate pressure in pulmonary artery bifurcations.
• Future work includes validation in larger populations and pulmonary hypertension cases.