Asymptomatic patients have been considered as superspreaders since they are infected with COVID-19 without exhibiting the most common coronavirus symptoms. Fortunately, recent research revealed that artificial intelligence (AI) models could detect the difference between the cough of a symptomatic patient as compared to asymptomatic ones.
A study titled, "COVID-19 Artificial Intelligence Diagnosis using only Cough Recordings," published in the IEEE Journal of Engineering in Medicine and Biology showed that an artificial intelligence model can distinguish between symptomatic patients and asymptomatic ones. People will simply submit forced-cough recordings on a web browser or their devices like cellphones and laptops and they could get their "results" immediately.
The researchers "trained" the AI model on how to distinguish between coughs by feeding it with sample coughs and spoken words of tens of thousands of people. After training the model, the researchers then fed new cough recordings to determine accuracy. The model, surprisingly, identified 100 percent of asymptomatic patients. Out of the coughs of symptomatic ones, 98.5 percent of patients were positively identified to have COVID-19.
The team is now planning to incorporate the model into an app. What they are looking forward to is that if the model is approved by the Food and Drugs Administration (FDA), it could be used on a wide-scale and could be a viable tool for non-invasive pre-screening of asymptomatic patients. The user will then be able to log into the app daily, simply cough into the phone and can get his/her results instantly, making it more convenient and of course, less expensive for patients.
"The effective implementation of this group diagnostic tool could diminish the spread of the pandemic," said Brian Subirana, co-author and a research scientist at the MIT Auto-ID Lab. He highlighted it could be effective if everyone would use it before going to work, to class, or to a restaurant.
Initially, the AI framework was used by MIT researchers in detecting Alzheimer's among patients. It effectively-identified samples of cough from the patients using biomarkers like vocal cord strength, lung, sentiment, respiratory performance, and muscular degradation.
When the pandemic hit, Subirana and the team started to look into the possibility of using the same framework to detect COVID-19. The researchers have already collected more than 70,000 recordings, with each recording totaling approximately 200,000 forced cough samples. Around 2,500 confirmed COVID-19 associated recordings were submitted. They then got 2,500 random audio-recordings. Four thousand were used to train the AI model and the remaining 1,000 recordings were used to test accuracy in distinguishing COVID-19 patients versus healthy ones.
The model successfully picked out patterns among the patients that enabled it to identify asymptomatic coughs from a healthy cough. The team noted though that the research is not meant to diagnose symptomatic patients.