Monitoring the sensor signals of intensive-care patients can catch a fatal emergency early. Automating that monitoring runs into a hard problem: the system has to know what is currently happening in the room. A sudden spike in heart rate means one thing if the patient is being moved by staff and something very different if they are lying still.
ZHAW’s Computer Vision, Perception and Cognition group worked with the intensive care unit at University Hospital Zurich, under Prof. Emanuela Keller, to extend their ICU Cockpit platform with video analysis that detects movements of the patient and of the medical staff, and uses them to annotate the sensor record automatically.
The clinical value is in the negative cases: correctly explaining away the alarms that are not emergencies is what makes an automated monitor usable on a real ward.