Estimation of passive and active biomechanical parameters in cardiac models using physics-informed neural networks

Federica Caforio

University of Graz, Medical University of Graz, BioTechMed-Graz, Austria

Wednesday 6th November 2024

14:30:00

Aula Saleri

Biophysical models of cardiac function are gaining traction due to their ability to predict patient outcomes and optimise treatment strategies. Accurately calibrating the parameters of these models is essential for a detailed understanding of myocardial function, but it remains challenging in clinical practice. This seminar investigates a novel approach [1] that combines physics-informed neural networks [2] with advanced three-dimensional nonlinear cardiac biomechanical models to reconstruct displacement fields and estimate patient-specific biophysical properties, such as passive stiffness and active contractility. A key feature of the proposed learning algorithm is that it solely relies on displacement and strain data routinely obtained in clinical settings. Benchmark tests demonstrate the method’s accuracy, robustness, and its potential to efficiently estimate patient-specific biophysical parameters in nonlinear, time-dependent biomechanical models. This approach opens up new possibilities for detecting and characterising tissue inhomogeneities, such as fibrotic regions, and could drastically improve the diagnosis and treatment planning of cardiac conditions. [1] Caforio, F., Regazzoni, F, Pagani, S., Karabelas, E., Augustin, C., Haase, G., Plank, G. and Quarteroni, A. (2023). Physics-informed neural network estimation of material properties in soft tissue nonlinear biomechanical models. Computational Mechanics, pp.1-27. [2] Raissi, M., Perdikaris, P., and Karniadakis, G. E. (2019). Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations. Journal of Computational Physics, 378:686-707. Contatti: stefano.pagani@polimi.it francesco.regazzoni@polimi.it