Abstract
Mobile systems and wearable technology have developed substantially over the last decade and provide a unique long-term and continuous insight and monitoring into medical condi- tions in health research. The opportunities afforded by mobile health in access, scale, and round-the-clock recording are counterbalanced by pronounced issues in areas like participant engagement, labelling, and dataset size. Throughout this thesis the different aspects of an mHealth study are addressed, from software development and study design to data collection and analysis. Three medically relevant fields are investigated: detection of stress from physiological signals, seizure detection in epilepsy and the characterisation and monitoring of COVID-19 through mobile health techniques.The first two analytical chapters of the thesis focus on models for acute stress and epileptic seizure detection, two conditions with autonomic and physiological manifestations. Firstly, a multi-modal machine learning pipeline is developed targetting focal and general motor seizures in patients with epilepsy. The heterogenity and inter-individual differences present in this study motivated the investigation of methods to personalise models with relatively little data. I subsequently consider meta-learning for few-shot model personalisation within acute stress classification, finding increased performance compared to standard methods.
As the COVID-19 pandemic gripped the world the work of this thesis reoriented around using mHealth to understand the disease. Firstly, the study design and software development of Covid Collab, a crowdsourced, remote-enrollment COVID-19 study, are examined. Within these chapters, the patterns of participant enrolment and adherence in Covid Col- lab are also considered. Adherence could impact scientific interpretations if not properly accounted for. While basic drop-out and percent completion are often considered, a more dynamic view of a participant’s behaviour can also be important. A hidden Markov model approach is used to compare participant engagement over time.
Secondly, the long-term effects of COVID are investigated through data collected in the Covid Collab study, giving insight into prevalence, risk factors, and symptom manifestation with respect to wearable-recorded physiological signals. Long-term and historical data accessed retrospectively facilitated the findings of significant correlations between development of long-COVID and mHealth-derived fitness and behaviour.
| Date of Award | 1 Jan 2024 |
|---|---|
| Original language | English |
| Awarding Institution |
|
| Supervisor | Richard Dobson (Supervisor) & Amos Folarin (Supervisor) |