Vision-Based Patient Monitoring in Inpatient Mental Health Care

Inpatient mental health services care for people who can no longer be supported at home and need to
be admitted to hospital due to severe mental illness. The acute mental health setting is particularly
challenging due to the risks of harm to patients and staff. Vision-based patient monitoring is the use of
cameras and computer vision software to support staff in maintaining patient safety by providing
physiological measurements, such as pulse and breathing rate, and information about patient
behaviour and activity. Vision-based patient monitoring is particularly well-suited to the inpatient
mental health setting as contactless monitoring minimises disturbance, discomfort, self-harm risk and
logistical challenges that may be introduced by wearable technologies, without sacrificing the
immediacy of the monitoring, which is often vital to support nurses in responding to emergencies
quickly.
The central aim of the research will be to develop novel vision-based patient monitoring algorithms
that can support healthcare professionals in inpatient mental health care. These algorithms could be
in the form of alerts to immediate risks to patient health, such as rapid deterioration of vital signs.
Alternatively, algorithms could help with long-term health monitoring of patients, such as screening for
sleep disorders or tracking longer-term changes in patients' mental health or physical health during
their stay on a ward.
The novelty of the proposed research comes from access to large-scale high-quality datasets and the
development of state-of-the-art machine learning algorithms, such as transformers and LLMs, to
analyse multimodal data from those datasets. Access to these data is granted through collaboration
with the vision-based patient monitoring company Oxehealth. Oxehealth will provide access to
proprietary datasets, under the appropriate research governance, offering a unique opportunity for
algorithmic development and evaluation. Additionally, Oxehealth has many...