An artificial intelligence that can detect diarrhoea with 98 per cent accuracy from recordings of toilet sounds could help track outbreaks of diseases, such as cholera
An artificial intelligence can detect diarrhoea with up to 98 per cent accuracy by analysing the sounds emanating from toilets. This skill could help us track outbreaks of diseases such as cholera.
Maia Gatlin at the Georgia Institute of Technology and her colleagues collected 350 recordings of toilet-based sounds from YouTube and sound database Soundsnap – covering standard defecation, diarrhoea, urination and flatulence.
The researchers then used 70 per cent of the recordings to train an AI to recognise audible differences between the four types of excretion. Once they confirmed that the AI could consistently do this with another 10 per cent of the data, they tested the AI’s performance using the last 20 per cent of the recordings.
This revealed that the AI could correctly class an excretion event as diarrhoeal or non-diarrhoeal with 98 per cent accuracy, if background noise – such as people talking – was filtered out, and with 96 per cent accuracy if background noise was kept in.
Using this approach to track outbreaks of disease would involve placing microphones by public toilets and feeding the data to the AI, says Gatlin.
The researchers have created a set-up that could be mounted in toilets. A microphone picks up the noise from lavatory use, which is recorded on a microprocessor in a nearby “Diarrhea Detector” box (pictured above) that has a machine learning AI model onboard. The signal is processed and evaluated before being classified as being diarrhoea or not.
Diarrheal diseases, such as cholera, can lead to death if left untreated, and automatically detecting levels of diarrhoea in the community could help track outbreaks and reduce disease spread.
However, the sound of excretion events depends on the type of toilet used.
“Many areas where cholera is rife do not have the same types of toilets we have in the US or UK, so we would need to develop an AI for sounds made in different types of toilets,” says Gatlin, who announced the results at a Meeting of the Acoustical Society of America on 5 December.
The reliance on online recordings in developing the AI also meant that the researchers had to manually listen to recordings and decide whether the audio labels accurately described the type of excretion event – without knowing for certain what type was recorded.
“Going forwards, we would like to collect real-world excretion recordings and develop the AI on those,” says Gatlin.
Sign up to our free Health Check newsletter that gives you the health, diet and fitness news you can trust, every Saturday
More on these topics: