Machine learning for vision based patient monitoring

Lead Research Organisation: University of Oxford
Department Name: Engineering Science


Brief description of the context of the research including potential impact
Vision based patient monitoring (VBM) currently encompasses a range of existing technologies such as the remote monitoring of vital signs such as heart and breathing rate [1]. It has already had a positive impact in
many clinical settings, such as mental health care, acute care and assisted living, giving clinicians a better understanding of a patient's physiological state and patients a better night's sleep [2]. More intelligent visual
systems would open up a wider range of useful downstream applications and, ultimately, data which can better inform clinicians.
Aims and Objectives
The "gold standard" tool for sleep monitoring is polysomnography (PSG). However, the data produced by PSG requires interpretation by a trained expert. Additionally, there can often be subjective disagreement between
experts [3]. This work aims to show that a machine learning approach using VBM can achieve a similar level of agreement, or better, than independent experts achieve when analysing sleep.
This work also aims to identify patient activity that is indicative of underlying health concerns. In particular: restlessness, self-harm and excessive exercise are three examples of activity that could be identified through
VBM using a machine learning approach.
One of the key pillars to machine learning success has traditionally been large, annotated datasets. However, in healthcare, data is often scarce, unlabelled and sensitive in nature. I will explore how these challenges can
be overcome, to develop machine learning systems that can be used to tackle the downstream tasks detailed above.
Novelty of the research methodology
Sleep staging/quantification remains an active area of research. No single process can currently claim to provide an objective measure of sleep. Using state-of-the-art computer vision methods in combination with
existing biomarker data, is a novel approach to achieve this.
Alignment to EPSRC's strategies and research areas
The proposed research strongly aligns with the EPSRC areas of 'Artificial intelligence technologies' and
'Medical imaging'.
Any companies or collaborators involved
My research is kindly sponsored by Oxehealth, a vision based patient monitoring company spun out of the
Oxford Institute of Biomedical Engineering in 2012.
[1] M. Villarroel et al., 'Non-contact physiological monitoring of preterm infants in the Neonatal Intensive Care Unit', npj
Digital Medicine, vol. 2, no. 1, Art. no. 1, Dec. 2019, doi: 10.1038/s41746-019-0199-5.
[2] H. Lloyd-Jukes, O. J. Gibson, T. Wrench, A. Odunlade, and L. Tarassenko, 'Vision-Based Patient Monitoring and
Management in Mental Health Settings', Journal of Clinical Engineering, vol. 46, no. 1, pp. 36-43, Mar. 2021, doi:
[3] H. Danker-Hopfe et al., 'Interrater reliability for sleep scoring according to the Rechtschaffen & Kales and the new AASM
standard', Journal of Sleep Research, vol. 18, no. 1, pp. 74-84, 2009, doi: 10.1111/j.1365-2869.2008.00700.x


10 25 50

Studentship Projects

Project Reference Relationship Related To Start End Student Name
EP/S024050/1 30/09/2019 30/03/2028
2416606 Studentship EP/S024050/1 30/09/2020 29/09/2024 Jonathan Carter