Full 3D blood flow simulation in curved deformable vessels using physics-informed neural networks

Abstract

Investigating blood flow in the cardiovascular system is crucial for assessing cardiovascular health. Numerical simulation is widely used to replace invasive measurements and reduce harnesses for patients. Traditional numerical methods, such as finite-element methods, have yielded excellent results. But adapting these methods to real-life simulations remains expensive. In this study, we extend our previous research by introducing a flexible and efficient approach for real-world simulations in a fully 3D framework. We employ physics-informed neural networks (PINNs) to solve the Navier-Stokes equations in a dynamic, deformable domain, focusing on simulating blood flow through elastic vessels with various bending degrees. The structure mechanics modeling is also extended into a full 3D sense, which can give more precise simulations for the fluid mechanics within cardiovascular systems. Our mesh-free approach circumvents the need for discretization and meshing, thus enhancing computational efficiency for complex geometries. Experiments on curved vessels in different degrees are included. The blood flow mechanics are analyzed, indicating that highly curved vessels significantly reduce fluid velocity and exhibit less activity during the diastolic phase in a non-linear fashion.