Health Tech Insider has kept an eye on research and development of AI diagnostic tools that use voice and sound samples since we first covered ResApp in 2016. A patient with a cough can use this smartphone app — which analyzes the sound of coughing and breathing — to assist physicians in diagnosing acute respiratory illness remotely. Not surprisingly, researchers in 2020 hope to detect the presence of COVID-19 through sound samples. But an estimated 20% of COVID-19 cases are asymptomatic, so how could a sound sample identify the disease in those individuals?
Researchers at MIT have answered that question with a new AI model, which has identified asymptomatic COVID-19 patients’ coughs with 100% accuracy. The team designed a unique AI algorithm based on their previous research, in which they trained an AI model to recognize specific variations in vocal quality associated with Alzheimer’s disease. The newest algorithm can detect differences in the sound of forced-cough recordings of asymptomatic patients.
Volunteers with a confirmed COVID-19 diagnosis, as well as healthy individuals, submitted forced-cough samples via computers and mobile devices. The research team used thousands of these samples, along with spoken-word recordings, to train the new model, which distinguishes sound markers that the human ear can’t discern.
The MIT team has published its findings in the IEEE Journal of Engineering in Medicine and Biology. The team’s next step is to integrate the model with a user-friendly app, which they hope will eventually offer fast, free prescreening for COVID-19. Individuals could presumably cough into the app every day; if the app identifies the disease’s presence, the user could schedule an in-person test to confirm. As asymptomatic cases are one of the most worrying vectors of transmission, such an app has the potential to reduce the spread of COVID-19 significantly.