Investigation on aortic hemodynamics based on physics-informed neural network.

Area of Science:

• Biomedical Engineering
• Computational Fluid Dynamics
• Machine Learning

Background:

• Non-invasive arterial pressure measurement remains challenging.
• Computational fluid dynamics (CFD) offers precision but lacks real-time application due to complex preprocessing and boundary condition requirements.
• Machine learning (ML) shows promise in hemodynamics due to its learning capacity and speed.

Purpose of the Study:

• To propose a novel method for non-invasive intra-vascular pressure estimation using a physics-informed neural network (PINN).
• To validate the PINN method against CFD simulations in an idealized aortic arch model.
• To investigate the sensitivity of the PINN model to various physiological parameters.

Main Methods:

• Developed an idealized aortic arch model and performed CFD simulations with two-way fluid-solid coupling.
• Utilized CFD-generated data (space-time coordinates, velocity, pressure) for PINN training and validation.
• Integrated Navier-Stokes and continuity equations into the PINN loss function for velocity and relative pressure calculation.
• Implemented post-processing to derive absolute pressure based on relative pressure, elastic modulus, and vessel wall displacement.

Main Results:

• PINN accurately predicted velocity and pressure fields, showing good consistency with CFD simulation results.
• Relative errors for maximum and average absolute pressure in the aorta were 7.33% and 5.71%, respectively.
• The relative pressure field was most sensitive to blood velocity, followed by blood viscosity and vascular elasticity.

Conclusions:

• The proposed PINN method offers a viable approach for non-invasive intra-vascular pressure estimation.
• This technique has significant potential for the diagnosis and monitoring of cardiovascular diseases.
• PINN's sensitivity analysis provides valuable insights into hemodynamic parameter influences on pressure estimation.