Artificial intelligence-enhanced electrocardiography for cardiovascular risk prediction
File(s)
Author(s)
Sau, Arunashis
Type
Thesis or dissertation
Abstract
The electrocardiogram (ECG) is an ubiquitous and low-cost investigation that records the electrical activity of the heart non-invasively. Through the advent of artificial intelligence (AI), previously unheralded insights are being discovered through the application of AI to the ECG (AI-ECG). In this thesis I explore the application of AI-ECG for cardiovascular risk prediction. Additionally, I explore the application of detailed phenotyping and genotyping methods to AI-ECG derived traits, including Phenome- and Genome-wide association studies. First, using an unsupervised machine learning approach, I demonstrated how neural-network derived ECG features derived from a simple rhythm identification task could be used to identify phenogroups with prognostic significance. Second, I developed a novel supervised machine learning based mortality prediction model, using a discrete-time survival approach. This mortality prediction model forms the basis of an AI-ECG risk estimation platform (AIRE), capable of predicting not only mortality, but also the future occurrence of important cardiovascular diseases: atherosclerotic cardiovascular disease, heart failure, ventricular arrhythmia, complete heart block and atrial fibrillation. I demonstrate the exceptional performance of AIRE and compare it to existing clinical, imaging, biochemical and genetic risk factors. Finally, I explore how AI-ECG traits can be thought of as digital biomarkers, and describe the finding that AI-ECG sex prediction can be used to develop a continuous sex score that can capture the continuum of risk and sex phenotypes within each sex. I found sex score discordance was a novel AI-ECG biomarker capable of identifying females with disproportionately elevated cardiovascular risk.
Version
Open Access
Date Issued
2024-05-02
Date Awarded
2024-08-01
Copyright Statement
Creative Commons Attribution NonCommercial Licence
License URL
Advisor
Ng, Fu Siong
Peters, Nicholas
Mandic, Danilo
Sponsor
British Heart Foundation
Grant Number
FS/CRTF/21/24183
Publisher Department
National Heart & Lung Institute
Publisher Institution
Imperial College London
Qualification Level
Doctoral
Qualification Name
Doctor of Philosophy (PhD)