Supervisor

Suitable for

Abstract

Co-supervised by Vicente Grau (Biomedical Engineering, Oxford) vicente.grau@eng.ox.ac.uk

Cardiac anatomy and function vary considerably across the human population with important implications for clinical diagnosis, treatment planning, and prognosis of disease. Consequently, many computer-based approaches have been developed to capture this variability for a wide range of applications, including explainable cardiac disease detection and prediction, dimensionality reduction, cardiac shape analysis, and the generation of virtual heart populations. Here, we will leverage on these technologies to further investigate connections between cardiac anatomy and electrocardiographic (ECG) signals in patients with Hypertrophic Cardiomyopathy (HCM), the most common hereditary heart disease and leading cause of sudden cardiac death in the young and competitive athletes.

To investigate connections between patterns of cardiac hypertrophic and their manifestation in the ECG in HCM patients. Generation of virtual populations of HCM hearts, capturing the variability in cardiac shape and ECG signals in this group of high-risk patients.

We have recently proposed novel variational mesh autoencoder (mesh VAE) methods as a novel geometric deep learning approach to model population-wide variations in cardiac anatomy [1], which enable direct processing of surface mesh representations of the cardiac anatomy in an efficient manner. These methods can also be extended to simultaneously learn the ECG signal of a given patient [2]. Exploiting cardiac Magnetic Resonance Imaging (MRI) and ECG resources from the UK Biobank, this project will explore the quality and interpretability of the mesh VAE's latent space for the reconstruction and synthesis of multi-domain cardiac signals in patients with HCM. It will also investigate the method's ability to generate realistic virtual populations of cardiac anatomies and ECG signals in terms of multiple clinical metrics in this group of patients.

(1) Interpretable Cardiac Anatomy Modeling Using Variational Mesh Autoencoders.https://doi.org/10.3389/fcvm.2022.983868 (2) Multi-Domain variational Autoencoders for Combined Modeling of MRS-Based Biventricular Anatomy and ECG-Based Cardiac Electrophysiology. https://doi.org/10.3389/fphys.2022.886723