Supervisors

Suitable for

Abstract

Cardiac electromechanical simulations of the human heart are computationally expensive, with one single beat taking hours to simulate in a supercomputer. This has prevented realising the vision of the cardiac digital twins for real-time therapy testing in clinical practice.

Cardiac electromechanical simulations of the human heart are computationally expensive, with one single beat taking hours to simulate in a supercomputer. This has prevented realising the vision of the cardiac digital twins for real-time therapy testing in clinical practice.

Graph Neural Networks have recently demonstrated great promise in replacing finite-element methods for solving mesh-based PDEs. Several works have been presented demonstrating the emulation of complex dynamics in physical systems to a high level of fidelity.

Generate training data through simulations of cardiac mechanics, design and train the machine learning model, and test the machine learning models under different conditions. Paper on Graph Neural Networks for emulating passive left ventricle mechanics https://www.sciencedirect.com/science/article/pii/S0045782522006004 (source code available). This project would consider a biventricular geometry of the heart. Paper on modelling and simulation of cardiac electromechanics for investigating mechanisms of disease in post-myocardial infarction https://www.biorxiv.org/content/10.1101/2022.02.15.480392v1.abstract

Prerequisites: How accurately can machine learning emulate patient-specific cardiac mechanics? The expected contribution would be to train a machine learning algorithm on simulation data to provide a fast alternative to existing simulation frameworks.