© IE Online Media Services Pvt Ltd
Tags:
Journalism of Courage
The Diarrhoea Detector’s name is pretty self-explanatory—the machine is a non-invasive sensor that uses a Raspberry Pi computer, a microphone and a machine learning algorithm to listen to people while on the toilet and ascertain whether they have diarrhoea or not.
“The sensor consists of a cheap, off-the-shelf microphone connected to a Raspberry Pi. The Raspberry Pi model could be downsized to cut costs in the future while still serving the same functions of data collection and model inference. At this stage, the device still needs to be connected to power, however, this may be something we can update in future stages,” said Maia Gatlin of the Georgia Institute of Technology to indianexpress.com over email. Gatlin led the team that developed the device and presented the research at the 183rd Meeting of the Acoustical Society of America.
The device is designed to identify bowel diseases without collecting any identifiable information from the user. To develop the device, Gatlin and her team collected data from online sources. The audio samples of excretion events were trimmed and turned into spectrograms, which graphs out the audio visually.
These spectrograms were labelled into four categories— “defecation,” “urination,” “flatulence” and “diarrhoea” — and used to train a neural network. 70 per cent of the data collected online was used to train the network while 10 per cent was used for validation and 20 per cent was used for testing.
They also tested the algorithm’s performance for background noises to make sure that it was looking for the right sound features, irrespective of the sound environment around the sensor. According to the researchers, the device was able to predict diarrhoea accurately 98 per cent of the time.
The researchers then developed a device that they call Synthetic Human Acoustic Reproduction Testing Machine (SHART), which simulates excretion events. This machine was found to be able to trick the machine learning model into “detecting” an excretion event 72 per cent of the time.
“The insights we look for at this stage of the research is a community-wide tracking system. So, not just that an individual is experiencing diarrhoea, but that a significant amount of these events are occurring in a community. This would allow us to alert health professionals early on to stop the spread of water-borne illnesses such as Cholera,” said Gatlin, when asked about the use case for such a device.
According to the United States Centers for Disease Control and Prevention (CDC), as many as 4 million people get affected by Cholera every year, with the disease claiming as many as 143,000 lives annually. Diarrhoea is one of the key symptoms of the disease. As the next step for the research, the scientists aim to gather real-world acoustic data to improve the machine learning model. They will also try to improve the SHART device to make it simulate more realistic sounds so that it can also be used to train the model.
