Machine learning of multiscale active force generation models for the efficient simulation of cardiac electromechanics

33/2019

Machine learning of multiscale active force generation models for the efficient simulation of cardiac electromechanics

Wednesday 4th September 2019

Regazzoni, F.; Dede', L.; Quarteroni, A.

High fidelity (HF) mathematical models describing the generation of active force in the cardiac muscle tissue typically feature a large number of state variables to capture the intrinsically complex underlying subcellular mechanisms. With the aim of drastically reducing the computational burden associated with the numerical solution of these models, we propose a machine learning technique that builds a reduced order model (ROM). In our approach, the latter is obtained as the best-approximation of the HF model within a class of candidate models represented by means of Artificial Neural Networks (ANNs). The ANN is trained to learn the dynamics of the HF model from input-output pairs generated by the HF model itself from which the ROM is built in a non-intrusive (black-box) way. Moreover, the learning machine is informed with some a priori knowledge on the HF model, in a semi-physical (gray-box) way. A drastic reduction in both computational cost and memory storage is achieved with respect to the HF model. This is crucial when performing numerical simulations of the cardiac function, that is when active force models are exploited in the multiscale problem of cardiac electromechanics. As a matter of fact, we achieve a computational speedup of about one order of magnitude, while preserving almost the same accuracy of the HF solution.

This report, or a modified version of it, has been also submitted to, or published on
Computer Methods in Applied Mechanics and Engineering