If you end up in a hospital in Europe, you have a one in 20 chance of acquiring an infection while there. One of the leading causes is a lack of hand hygiene. Even though hospitals are filled with alcohol-based gel dispensers and advisory posters, they’re not working. The conclusion of a recent pilot study may just have found the solution. Using a combination of depth cameras and computer-vision algorithms, a research team has tracked people around two hospital wards and automatically identified when they used gel dispensers. The trial was so successful that the group is now going to fully kit out three hospitals for a whole year, to see if it puts a dent in the stubborn acquired infections statistics. “We’re trying to shed light on the dark spaces of healthcare. Understanding the problem is just the first step,” says Alexandre Alahi at the Swiss Federal Institute of Technology (EPFL) in Lausanne. In the initial study, during a busy Friday lunch time they collected images from cameras installed overlooking corridors, patient rooms and alcohol-based gel dispensers, among other places. Of the 170 people they recorded entering a patient’s room, only 30 people correctly used the gel dispensers. The team then used 80 per cent of the images to train their algorithms to detect healthcare staff, track them as they move from one spot to another across different cameras, and monitor their hand hygiene behaviour. Once trained up, they then tested the system on the remaining 20 per cent, and achieved an accuracy of 75 per cent in telling whether people had used the dispensers. Secret shoppers Normally hospitals have a person with a clipboard to collect this data. Across the same period, a human trying to do comparable monitoring only had an accuracy of 63 per cent. But accuracy is only one aspect. A “secret shopper” can only note how often gel dispensers are correctly used, they can’t provide a continuous map of every person moving in and out of different locations, 24 hours a day – cameras can. “You can’t get the necessary resolution with human observers to really understand the problem,” says Philip Polgreen at the University of Iowa. He has used wearable technology to monitor hand hygiene and made discoveries that wouldn’t have been possible with observers alone. For example, in larger groups, people were more likely to adhere to the correct procedures, and there were some individuals who were far more important than others for controlling the problem. “We found that if you can improve the behaviour of a few very well connected people, you end up having a much bigger effect than when trying to target the overall average. This can tell us how to stop outbreaks,” says Polgreen. There are clear privacy concerns related to cameras constantly monitoring hospitals. However, the depth cameras used capture more information about the position of a person than about how they look. The resulting images consist of unidentifiable human blobs. Cameras dotted around hospitals could help with other things too. Artificial intelligence has already demonstrated the ability to detect falls and monitor vital signs. It could do this throughout a hospital. “We can’t afford to have a doctor in a room 24/7, but we could afford an AI doctor every room, and every corridor too, leaving humans to do the most important jobs,” says Alahi. Source
