A project led by Dr Tingting Zhu at the University of Oxford.
Cardiovascular diseases (CVD) cause 17.7 million of deaths annually across the globe, according to the World Health Organisation. CVD is especially prevalent in developing countries (DCs) such as China, accounting for 45% of all deaths annually, with myocardial infarction (MI) as the most common cause. This disease also has an associated financial burden of $558 billion.
Conventional clinical practice for diagnosing MI includes an interpretation of ECG readings and a blood test. The former uses ECG-based features to provide an indication of risk of MI, typically from 12 leads of ECG recording. This interpretation of the ECG is highly dependent on the expertise and training of the clinician. In DCs, this poses a challenge as the scarcity of expertise results in inequities in access to clinicians with appropriate training. The difficulty involved in interpreting ECGs results in inaccurate diagnoses, delayed treatment, and increased risk of mortality.
Additionally, ECGs are frequently noisy or artefactual (even in developed healthcare systems), where the quality of the data is dependent on the skill of the healthcare worker. This situation is exacerbated in DCs, where such expertise is scarce, and where the use of low-cost ECG equipment further reduces data quality.
There is a need for an automated system that can reduce inter- and intra-expert variability in diagnosis, and where it would be required, to improve care in resource-constrained settings. This program aims to produce a proof-of-principle using a novel machine learning system based on deep learning, constructed using data acquired from collaborators in China during this proposed study, and exploiting data already available from the collaborating China Kadoorie Biobank.