University of Calgary researcher helps create AI tool to detect whirling disease

Whirling disease was first found in Banff National Park in the summer of 2016. It can spread easily throughout a watershed.

A study led by a University of Calgary researcher has created a way to detect river-borne pathogens like whirling disease with an AI-based system.

Pouria Ramazi, an assistant professor in the faculty of science, is the lead author of a study published in Methods in Ecology and Evolution about finding a way to predict waterborne diseases with little-to-no data using artificial intelligence.

Ramazi and his team used an AI model called “hidden Markov models.”

• The information you need to know, sent directly to you: Download the CTV News App

Team members inputted easily obtained data like river flow and temperature to predict the presence of a microscopic parasite (Myxobolus cerebralis) that causes whirling disease in salmon, trout and whitefish.

“This project was funded and initiated by Alberta Environment and Parks, so we worked hand in hand with them to develop this model,” Ramazi said. “Some of it inevitably will spread anyways, but probably we could still prevent it to a larger extent and maybe also slow it down.”

Ramazi’s work began in 2018, when the province reached out to him to help track the spread of whirling disease in the Old Man River basin.

“At that time, they had less than a hundred sample points – because they need to send a crew out to take samples, take [the samples] to the labs or universities to test whether the parasite exists there or not – so it’s very costly,” he said. “And you cannot do that for everywhere around the watershed.”

The province needed a cost-effective solution to obtain data without sending people out to collect samples.

Ramazi immediately thought of AI, but says it typically needs vast amounts of data to be accurate.

“What we thought of doing was to use some other available data, environmental measurements that are readily available everywhere, so we don’t need to send crews. If we have weather stations, they already have good data.”

Once Ramazi and the team started inputting data in the AI model, the results came in and kept getting better and more accurate in its predictions.

“What was surprising was that even with very few samples, like just one sample, we were still it got like 70 per cent AUC – which is a measure for accuracy,” he said.

Whirling disease was confirmed for the first time in Canada in Banff National Park and the Bow River in August 2016.

It has since been detected in British Columbia.

It can be extremely harmful to the environment and economy by killing entire fish populations.

Ramazi says the AI model has the potential to help fishery officials manage the disease here and around the world.

“If more effort and funding is put into it, it can be also extended to [detect] other diseases in Canada – it doesn’t have to be just in the Oldman River,” he said.