Engineering researchers combine biochemistry and smartphone AI software in new system
Sushanta Mitra was doing field trials in remote areas of India when he was struck by how many people with no reliable access to clean water were connected to the internet via smartphones.
The contrast amazed him. It also gave him a good idea.
Six years later, the University of Waterloo engineering professor and his colleagues are refining new technology that puts the power to monitor water quality in the palms of people in the developing world.
The system combines artificial intelligence (AI) software on smartphones with low-cost tools the researchers previously developed to test water for potentially deadly E. coli bacteria.
Their ultimate aim is an affordable, comprehensive system to routinely monitor water for contamination without the delays and high costs of off-site laboratory tests.
“In remote locations, you’re talking about two days to take a sample, culture the bacteria and get results,” says Mitra, executive director of the Waterloo Institute for Nanotechnology. “That is a major problem because people will keep drinking the water and fall sick or even die.”
The new addition to their existing testing tools — which use paper strips, gel and plastic filters that change colour when E. coli bacteria are present in water — is AI trained to read and interpret the results.
Users simply take a smartphone photograph of the strip, gel or filter and the AI application determines, based on colour change, whether there is contamination or not. It has proven 99.99 per cent accurate, all but eliminating errors made by human interpreters.
Work is ongoing to refine the system through training with more data to also predict E. coli concentration, and therefore the level of contamination, by interpreting the intensity of the colour change.
“We brought two disciplines together,” says Mitra, a professor of mechanical and mechatronics engineering. “One is biochemistry and the other is a mathematical tool, a convolutional neural network.”
Tests using mobile water kits that were first developed by the researchers cost about $15 each. A subsequent, simpler system using paper strips costs about $1 a test. The AI app will be publicly available for free downloading.
Lab tests to determine the degree of contamination, by contrast, cost about $60 apiece, an enormous expense in developing countries that lack modern water infrastructure.
“This is a very empowering tool,” says Mitra, whose primary collaborator is post-doctoral fellow Naga Siva Kumar Gunda. “Even people who don’t have access to clean water have access to cellphone networks.”