Catalan researcher leading team to build Covid-19 detecting algorithm

Professor Brian Subirana, a Director at MIT, is leading a project to devise an algorithm that can detect Covid-19 in people’s coughs, recorded directly into a microphone connected to a phone or laptop.

On the web page opensigma.mit.edu, anybody can record their cough and provide a sample to the researchers. People taking part also answer a couple of short questions on Covid-19, whether they have it, suspect they may have it, or know they don’t have it.

In fact, even coughs from people who know they don’t have the coronavirus can provide plenty of value to the project. Most of the samples collected already are from Covid-19 negative people, and Subirana explains that "really helps us because that builds up a reference cough and that helps algorithms a lot. The algorithm works on Artificial Intelligence to learn from a library of coughs how to distinguish them."

The Catalan professor compares the algorithm with a Ferrari, and says "samples are like gasoline to AI," meaning the more samples collected, the better the machine will work. "We just started collecting samples very recently," he says. "We already have the largest cough database in the world as far as I know, we have around 150,000," but the research team is aiming for one million samples. 

Professor Subirana is also pleased with the relatively recent breakthrough in transfer learning - transferring knowledge learned in solving one problem in AI to another related matter. He says that even by using regular coughs, this still provides valuable information to finding Covid-19 coughs. 

Collaborations with hospitals 

"The end goal [of the project] is to provide tools to doctors," Subirana says. While the AI will not be able to medically confirm whether or not somebody has the coronavirus, it may be able to provide people with valuable information, and the MIT director likens it with a "pre-screening test." 

The project is working in conjunction with many hospitals across the world, including Barcelona's Hospital Clinic. These collaborations provide data to the researchers about many patients who are confirmed Covid-19 positive and the evolution of their cough "as people go through the symptoms."

"Another possibility," Subirana adds, "is 'you seem to be in danger so we’re going to put you in the ICU because we don’t like this cough and the algorithm is signalling something that we didn’t know before.'"

Benefits of AI programme

The algorithm being designed holds varying advantages. When it has enough samples and can function as a reliable pre-screen test, it will be able to give people an indication of their health levels for free, any time, any place. 

With it, nobody has to leave their home, and nobody has to be seen by a healthcare professional, eliminating the potential danger of spreading the coronavirus. 

Subirana is also excited about the possibilities that AI can bring to the world of healthcare. He says that "for the first time, we can help doctors" from our own homes. 

"That may show us how AI may evolve in the future. Maybe we can all crowdsource AI, because the more [samples] AI gets, the more we can pull out and we can apply it to other areas, not just covid-19."