Post on 01-Apr-2015
transcript
Early Warning Systems
in
Biomedical Signal Processing
davidc@robots.ox.ac.uk
Dr. David A. Clifton, College LecturerInstitute of Biomedical EngineeringUniversity of Oxford
I have a neural network
processor.
The problem 23,000 preventable cardiac
arrests occur every year in UK hospitals
20,000 readmissions into ICU every year – mortality 50%
The majority of these occur because physiological deterioration goes undetected – why?
Primitive warning systems
Level 3:ICU 1 : 1
Level 2: Step-down 1 : 4
Level 1: Acute wards 1 : 4
Level 0: General wards 1 : 10
Level -1: Home 1 : ?
Patient monitors generate very high numbers of false alerts (e.g. 86% of alerts)
The NHS response
Conventional univariate analysis
Existing methods apply simple thresholds to each parameter
Intolerant to transient noise Possibly not the appropriate domain (time ,
frequency) Where do we set these thresholds in a principled,
reliable manner?
Nurses & junior doctors trained to ignore alarms Rolls-Royce has deactivated conventional
automated methods
Intelligent early warning systems
Intelligent early warning systems
Available biosignals
EEG / GCSHeart rateBreathing rateSpO2Blood pressureTemperature
On a “good” day... Obvious
tachycardia Obvious
tachypnea Obvious
desaturations Obvious
hypotension Obviously
unconscious
Abnormalities were detected by clinicians,patient escalated.
Note the difficulties: Incomplete data Noisy data Varying sample
rates
On a “not-so-good” day...
Gradual deterioration
Is this patient gettingworse?
Should we make a call to emergency teams?
Patient unescalated,died soon after.
Intelligent early warning systems
How can we detect abnormality in patient biomedical signals?
How can we do it in a reliable way?
What are the pitfalls that we have to avoid?
How can we evaluate it?
In Hilary term... Plenty more to look forward to:
machine learning in biomedical engineering
In Hilary term...
Hardware Devices& Comms
Physiology & Clinical Issues
Commercial Solutions & Regulatory Issues
Signal Processing & Machine Learning
Projectroles