Artificial intelligence applications will recognize from your cough that will be recorded on your mobile phone, if you are infected with COVID-19.
The new machine learning models will be based on how your cough sounds will be recorded by the mobile phone, so that they can accurately detect if you have coronavirus, even if you are asymptomatic.
Research teams around the world are working feverishly to find a cheap, fast and effective way to detect coronavirus. Some have turned their attention to artificial intelligence to build an application that could detect the presence of a coronavirus from coughing.
Not at all imaginative, since in June, one team at the University of Oklahoma showed that It was possible to distinguish a COVID-19 cough from a cough caused by other infections. So now a document from MIT states that Using a huge data set from coughing sounds, it can and does identify asymptomatic people with a 100 percent detection success rate.
He states that these applications could record, at the user's command, his cough, to alert him if he has a coronavirus and could eventually be used for free, large-scale population protection.
The fact that artificial intelligence models can detect COVID-19 through coughing suggests that there is no real asymptomatic infection, and if you get stuck then there are always natural changes in your body that change the way a person makes sound.
"There are not many situations that do not give you any symptoms," he says Brian Subirana, director of the MIT laboratory and co-author of a recent study published in the IEEE Open Journal of Engineering in Medicine and Biology.
While the human ear cannot discern these changes, AI can. Ali Imran, who led the previous investigation in AI4Neworks Research Center of the University of Oklahoma, compares the idea to a guitar: If you put objects of different shapes or materials inside a guitar but play the same notes, it will result in subtly different sounds. "The human ear is capable of distinguishing perhaps five to ten different characteristics of a cough," says Imran. “With processing signaland machine learning, we can extract up to 300 different features."
When the pandemic broke out around the world, Subirana's team at MIT was working on machine learning algorithms to detect Alzheimer's disease on recordings, using biomarkers such as vocal cord strength, emotion, lung performance and muscle collapse. When it became clear that cough was a key feature of COVID-19, they quickly turned to see if it was possible for AI to detect coronavirus infections.
So the team collected cough recordings through a website, around April and May, developing what the team claims is the largest sound data set for COVID-19 to date, with 70.000 recordings, of which 2.680 were submitted by verified individuals. .
Initially, the MIT team developed AI models with zero output, but soon reached the upper limit of accuracy, around 70%. As a test over a weekend, the researchers trained the existing AI model of Alzheimer's disease with the data for the cough for COVID-19 and it worked. The model was 98,5 percent accurate in detecting people who had received a positive test result. When identifying positive people with no symptoms at all, this accuracy increased to 100 percent, while in negative cases the success rate rose to 83,2 percent.
"It was a bit strange" that asymptomatic patients were easier to detect than symptomatic patients, says Subirana, but it makes sense as the confounding agents of other infections would make it difficult to detect the characteristics of COVID-19 cough.
In June, Imran and his colleagues managed to develop an artificial intelligence model for detecting asymptomatic coughs and overcome these confusing factors to distinguish COVID-19 cough from the coughing sounds of bronchitis, pertussis and asthma with an overall accuracy of 90%. "Our goal was to ensure that someone who just had asthma would not be misdiagnosed as having COVID," says Imran.
Most groups currently running this type of work are currently collecting more cough recordings: in workplaces, hospitals, online and elsewhere. The researchers hope that cough apps will one day be used for everyday screenings, such as for students or workers who cough into their phones before heading to the bathroom. school or at work. Kind of like the thermometer going on right now.
Ultimately, says Subirana, the tool θα μπορούσε να είναι μέρος ενός πραγματικού διαγνωστικού COVID-19, ίσως όταν χρησιμοποιείται σε συνδυασμό με άλλους βιοδείκτες, όπως ο πυρετός.
