Abstract
dc:description.abstractWearable devices are revolutionising personal health and fitness monitoring by enabling real-time, continuous, and non-invasive tracking of various physiological parameters. Among wearables, earables (sensor-equipped earbuds) have emerged as a promising platform for physiological sensing due to their widespread use, stable position on the body, and proximity to key organs and vasculature. Research has investigated embedding specialised sensors into earables for physiological monitoring. However, earables face size and shape constraints due to their need to fit comfortably in the ears, limiting the feasibility of including such sensors. Nonetheless, commodity noise-cancelling earbuds like the Apple AirPods Pro natively contain a microphone that faces inside the ear canal. In this thesis, we explore the potential of using this in-ear microphone for monitoring physiological parameters. Specifically, we focus on three key vital signs: heart rate, respiratory rate, and stroke volume. For each of these, we develop novel sensing pipelines that leverage the unique properties of in-ear audio, including its ability to capture both heart sounds and footstep sounds. First, we study heart rate monitoring under conditions of daily life, encompassing when the user is both sedentary and active. We propose a pipeline to perform supervised denoising of in-ear audio to emphasise heart sounds within motion-corrupted signals from which heart rate can be determined. We thus show the feasibility of accurately estimating heart rate under various conditions, demonstrating the potential for a new sensing modality for heart rate on earables. Our second contribution explores the possibility of measuring respiratory rate, another key physiological parameter, under daily life conditions using only in-ear audio. We use physiological couplings between cardiovascular activity, gait, and respiration to indirectly estimate respiratory rate from heart sounds and footstep sounds captured by the in-ear microphone. In doing so, we effectively overcome the shortcomings of requiring reliable respiratory sounds for respiratory rate measurements. This contribution proves the possibility of achieving robust respiratory rate estimation under various conditions using earables. Finally, we investigate the potential for monitoring stroke volume, a clinical vital sign typically measured only in medical settings, using in-ear audio. We employ self-supervised learning and transfer learning to demonstrate the feasibility of estimating average stroke volume using only sensors available in commodity earbuds. This thesis contributes to the growing field of earable computing and physiological sensing, advancing the understanding and application of in-ear audio for physiological monitoring. Our research demonstrates that in-ear microphones can effectively monitor key physiological parameters in real-life conditions, paving the way for widespread, continuous health moni- toring. Our findings have significant implications for personal health monitoring, fitness tracking, and potential clinical applications, demonstrating the potential of earables in transforming how we monitor and understand human physiology in daily life.
Degree
thesis:*- Name dc:type.qualificationname
- Doctor of Philosophy (PhD)
- Level dc:type.qualificationlevel
- Doctoral
- Grantor dc:publisher.institution
- University of Cambridge
- Year dc:date.issued
- 2024
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Butkow, Kayla-Jade
- Advisor dc:contributor.advisor
-
- Mascolo, cecilia
Subjects
dc:subject × 3Rights
dc:rightsIdentifiers
dc:identifier.*- DOI dc:identifier.doi
- https://doi.org/10.17863/CAM.116369
- OAI identifier oai:identifier
- oai:www.repository.cam.ac.uk:1810/380952