Related Experiment Video
Updated: Jun 23, 2025
Assessing Cerebral Autoregulation via Oscillatory Lower Body Negative Pressure and Projection Pursuit Regression
Published on: December 10, 2014
Physics-informed neural networks for parameter estimation in blood flow models
Jeremías Garay1, Jocelyn Dunstan2, Sergio Uribe3
1Department of Mechanical and Metallurgical Engineering, Pontificia Universidad Católica de Chile, Chile; Center of Biomedical Imaging, Pontificia Universidad Católica de Chile, Chile; Millennium Institute for Intelligent Healthcare Engineering (iHealth), Chile.
Physics-informed neural networks (PINNs) effectively estimate parameters and velocity fields from limited hemodynamic data. This deep learning approach shows promise for complex physical system simulations, outperforming traditional methods with more parameters.
Area of Science:
- Computational fluid dynamics
- Biomedical engineering
- Machine learning
Background:
- Physics-informed neural networks (PINNs) are valuable for inverse problems with incomplete data.
- Hemodynamics presents challenges due to difficult boundary modeling and scarce high-quality measurements.
Purpose of the Study:
- To apply PINNs for estimating reduced-order model parameters and velocity fields in the aorta.
- To analyze performance in both stationary and transient flow regimes using noisy scatter measurements.
Main Methods:
- Utilized PINNs methodology for parameter estimation and velocity field reconstruction.
- Investigated two distinct flow regimes: stationary and transient.
- Compared PINNs performance against a Kalman filter approach.
Main Results:
- Achieved robust and accurate parameter estimations from simulated data.
- Velocity reconstruction accuracy depends on measurement quality and flow complexity.
- PINNs outperformed Kalman filters when estimating a higher number of parameters.
Conclusions:
- PINNs offer a powerful deep learning-driven approach for simulating complex coupled physical systems.
- The method demonstrates significant potential for advancing hemodynamic modeling and analysis.
