- Many people infected with the novel coronavirus are asymptomatic or presymptomatic, but they can still be contagious.
- Researchers from MIT developed what could become a brilliant screening tool that offers an instant COVID-19 test every time you leave the house.
- Using machine learning algorithms trained to detect Alzheimer’s via forced coughs, the researchers think they can use the same technology to detect asymptomatic COVID-19 cases.
A combination of factors makes the novel coronavirus particularly difficult to tame. First of all, it spreads with incredible ease, especially indoors. Secondly, the symptoms need anywhere from 2 to 14 days to show up, and that’s if they even appear at all. Many people get an asymptomatic form of the illness but they can still spread it to others. This brings us to the third feature of the virus: infected people become contagious before the onset of symptoms even if they’re not asymptomatic, which means people who appear to be perfectly healthy can pass on the illness to others.
Catching the asymptomatics, and halting the asymptomatic spread (which includes presymptomatic transmission), might help governments reduce transmission and contain the spread. And in the future, a simple app running on phones or smart speakers might give asymptomatic people an instant COVID-19 test after just one cough.
Coughing is a symptom, so a person who coughs is no longer asymptomatic. But what MIT researchers were able to do relies on forced coughing. A healthy person can cough on command, and that might be enough for artificial intelligence (AI) to tell you whether you’ve been infected with the novel coronavirus. Such an app could deliver instant results without requiring a professionally administered test. The results might not always be perfect, but a positive diagnosis would mean the person should immediately seek out confirmation from a PCR test.
The app hinges on research that MIT researchers perfected for Alzheimer’s. They used three machine learning algorithms to train the AI to recognize signs of Alzheimer’s, which causes neuromuscular degradation that impairs the vocal cords, not just memory decline. The vocal cords are involved in coughing, and while people might not be able to distinguish an Alzheimer’s cough by ear, the AI can.
One of the neural networks the solution uses was taught to distinguish sounds and associate them with vocal cord strength. The AI learned how to interpret the sound “mmmm” from 1,000 hours of audiobook speech until it could pick up the words “them,” “the,” and “then.” The second neural network was trained to pick up emotional states, since people with Alzheimer’s display sentiments like frustration or a flat affect more frequently than others. The third machine learning algorithm worked on a cough database, gauging changes in lung and respiratory performance.
When combined and paired with an algorithm that detects muscular degradation, the resulting AI was fed cough audio recordings. It turned out that the AI could detect Alzheimer’s via forced coughing with impressive accuracy. When the novel coronavirus pandemic began, the researchers tried the same principles.
“The sounds of talking and coughing are both influenced by the vocal cords and surrounding organs. This means that when you talk, part of your talking is like coughing and vice versa. It also means that things we easily derive from fluent speech, AI can pick up simply from coughs, including things like the person’s gender, mother tongue, or even emotional state. There’s, in fact, sentiment embedded in how you cough,” Brian Subirana told MIT News. “So we thought, why don’t we try these Alzheimer’s biomarkers [to see if they’re relevant] for COVID.”
The researchers went to work, creating a website where participants could register a series of forced coughs, as well as a survey of symptoms and an official COVID-19 test result. The scientists collected more than 70,000 recordings with some 200,000 coughs. Among those, 2,500 came from people who tested positive, including asymptomatics.
The team used 2,500 coughs from healthy people and the 2,500 covid coughs to train the AI with more than 4,200 samples. Then the AI went to work on the remaining 1,000 samples. The researchers say in a paper on the matter that the experiment yielded “a striking similarity between Alzheimer’s and Covid discrimination.”
The COVID-19 model picked up 98.5% of coughs from people with COVID-19, which sounds extremely promising. “We think this shows that the way you produce sound changes when you have Covid, even if you’re asymptomatic,” Subirana says.
The team is working on incorporating the model into an app that could be used as an affordable or free screening tool if the FDA approves it. Even if the tool isn’t 100% accurate, it could still be good enough to nudge people towards getting a test. Combine it with rapid testing and contact tracing apps, and screening tools like this might have a dramatic impact.
It shouldn’t be surprising that asymptomatic people are affected by the virus even if they don’t think anything is wrong. Blood tests and CT scans show that the body is affected, according to earlier studies. They might not show symptoms, but their immune systems are still putting up an invisible fight. So it’s no wonder that some muscular degradation might temporarily impact vocal cords and coughs.
The MIT paper is available in full at this link.
