Toronto (Canada) (AFP)
On the shores of Lake Ontario, a Canadian start-up was one of the first to warn of the risk of an epidemic of pneumonia, which started in Wuhan, China. His secret? Artificial intelligence.
Based in Toronto, BlueDot has developed an algorithm that scans hundreds of thousands of press articles every day and air traffic data to detect and track the risks of the spread of infectious diseases.
In the case of the new Chinese coronavirus, BlueDot sent its customers alerts as early as December 31, a few days before the first official communications from major public health agencies. It also correctly predicted in which countries the virus was likely to spread.
"What we are trying to do is push the limits of the use and analysis of data and technology in order to go faster," said Kamran Khan in an interview with AFP. the founder and CEO of BlueDot. "Facing an epidemic, time management is essential."
The idea for BlueDot germinated in the mind of this 49-year-old epidemiologist following the SARS epidemic in 2003. He was then a doctor specializing in infectious diseases in a Toronto hospital and helplessly witnessed the ravages of the virus that killed 44 people in the Canadian metropolis.
"Health workers had been infected, including one of my colleagues. Some died. It really opened my eyes," he says. "Once the epidemic was stopped, I said to myself that it should not start again."
- 65 languages, 150 diseases -
In 2013, he created BlueDot, which today has 40 employees, a "fairly unique team" made up notably of doctors, veterinarians, epidemiologists, "data scientists" and software developers.
They have developed an early warning system based on automatic language processing and machine learning methods.
Every 15 minutes, 24 hours a day, the algorithm sifts through official reports, professional forums and thousands of articles online and scans texts for keywords and phrases. He can read 65 languages and is able to track more than 150 types of disease.
"The machine searches for needles in a haystack and presents them to human experts," he said. The BlueDot team then trains the machine to recognize whether the information detected corresponds to a real threat or epidemic or not.
If it is credible, it is entered into a database which analyzes the location of the outbreak, the surrounding airports and the anonymous routes of air passengers around the world. Climatic data, as well as on the health system of each country or the presence of mosquitoes and animals causing human diseases are also examined.
Once the analysis is completed, BlueDot sends an alert to its customers - government agencies, airlines or hospitals - in the places that will receive the greatest number of these travelers. The objective: that these actors can prepare for the worst and anticipate.
Thus on the morning of December 31, the BlueDot algorithm spotted an article in Chinese referring to cases of pneumonia linked to an animal market in Wuhan. The virus was not yet identified but the machine recognized two expressions which made tilt: "Pneumonia" and "unknown cause".
At 10:00 am, a first alert was sent to its customers, particularly in Asia. "We didn't know it was going to become a global epidemic but we recognized certain ingredients similar to those we saw during the SARS," said Kamran Khan.
Thanks to its method, BlueDot also succeeded in predicting that the virus might spread from Wuhan to Bangkok, Taipei, Singapore, Tokyo and Hong Kong.
This is not BlueDot's first feat: in 2016, the company also rightly predicted that the Zika virus in Brazil would also affect southern Florida.
"These viruses are complex. These diseases are complex. But we are constantly pushing the limits of our knowledge after each of these epidemics," he said.
© 2020 AFP