Automated ECG analysis to screen for occult cardiac rhythm disorders
Project ID: PSM19
Supervisor: Dr. James Fraser
Second Supervisor: Dr. Vadim Alexeenko
Second Supervisor DepartmentOther
Introduction and background.
Disorders of cardiac rhythm including paroxysmal atrial fibrillation (PAF) and sudden cardiac death are important causes of poor health and premature death. They are treatable if detected sufficiently early, but may show no human-detectable ECG abnormalities. However, our recent work has revealed that ECG complexity changes allow automated detection of PAF from ECGs that appear normal even to expert eyes. This might allow screening for treatable, electrically occult cardiac rhythm disorders.
Our work to date has consisted of a small pilot study (25 cases, 25 controls). The proposed PhD project will take this work much further with the aim of translation to a clinical screening programme.
Outline of Method.
The project consists of three linked studies:
1) Analysis of ECGs from the UK Biobank to investigate correlations between ECG complexity measurements and cardiac rhythm disorders in a normal population case mix. This will initially employ Lempel-Ziv complexity analysis to detect PAF, following the success of the pilot study. It will then investigate the use of complexity analysis and other ECG analysis techniques, such as restitution analysis of exercise ECGs, to investigate risk of ventricular arrhythmia and sudden death, including, but not limited to, post-MI patients. The UK Biobank contains ECGs from over 100,000 subjects alongside up to 10 years of clinical diagnosis and outcome data.
2) Development, with a collaborator, of a novel machine learning approach to combine automated ECG analysis measurements with other clinical measurements in the UK Biobank, such as body weight, smoking status etc., to produce a combined predictor of cardiac arrhythmic risk.
3) Testing, with a clinical collaborator, of the developed cardiac arrhythmia risk score in a pilot prospective clinical study. This may involve the development of a hand-held device for simple automated ECG analysis.
This project involves computational analysis of large databases. Thus, the ideal PhD student would be interested in coding, statistics and machine learning. However, these skills can be taught during the project and so it will not be necessary to have prior skills in these areas.
The central aim of this project is translational, to develop clinical screening test for occult cardiac rhythm disorders. This will first investigate automated ECG screening for paroxysmal atrial fibrillation, continuing from existing pilot work. It will then further extend the automated ECG analysis with the aim of screening for arrhythmia and sudden death risk.
We have discussed this project with Dr Patrick Heck, a cardiologist at Papworth Hospital, who has kindly offered to collaborate with us subject to the success of the Biobank study.
The machine learning aspect of the project will be undertaken in collaboration with Dr Gareth Conduit, Department of Physics, University of Cambridge.
On success of the project, we will explore the production of devices for automated ECG analysis. We have made preliminary enquiries of Plessey Semiconductors, who have indicated their support for this approach in principle.