Predicting and monitoring cardiovascular outcomes using wearable devices and novel machine learning techniques

1) Brief description of the context of the research including potential impact

Diagnosis and monitoring of cardiovascular disease(CVD) requires intrusive, in-clinic testing. This requires patients to take time out of their daily lives, as well as clinical resources to perform tests, and only reflects a small snapshot of time. Wearable devices offer a solution to this problem - patients can wear them freely in their daily lives, whilst having relevant clinical data collected regularly or even continuously.
Wearables have been massively adopted by consumers, with over 1 billion connected devices as of 2022 [1]. This adoption shows that people are comfortable with electronics that monitor their health, and many products such as the Apple Watch use health awareness as a large part of their product offering through heart rate monitoring, SpO2 max., and other fitness measurements.
By combining this wealth of data with novel time series methods borne out of advancements in machine learning/artificial intelligence, wearables can help move cardiovascular monitoring from the clinic to the home, as well as help at-risk users seek medical attention that could stop or slow the progression of their conditions. The potential impact for this technology is huge - cardiovascular disease is responsible for 25% of deaths in the UK, and clinical resources are already stretched. Moving the burden of diagnosis from the clinic to an algorithm, or simply offering clinicians another metric for prioritising patient care, could have a massive effect on patient wellbeing.

2) Aims and Objectives

The key aim of this research is to investigate whether advanced ML/AI based time series analysis techniques can predict cardiac outcomes using data collected from wearable devices.

3) Novelty of Research Methodology

This research aims to use previously unused features in wearable signals (e.g. photoplethysmography) to predict cardiac outcomes, as well as applying new methods to known signals to...