Researchers from the Massachusetts Institute of Technology, Cambridge, have developed an artificial intelligence (AI) tool that can detect COVID-19 by listening to patients' coughs, regardless if they are symptomatic or not.
To build the tool, the researchers solicited audio recordings of patients coughing and accompanying information about their conditions via an online website. They collected a dataset of more than 70,000 recordings containing an average of three coughs per patient and an estimated 2,600 patients with a positive case, to date.
Using the COVID-19 cough recordings and an equal number of COVID-19 negative recordings randomly selected from the dataset of 5,320, the researchers developed, trained, and validated a model that listens for specific acoustic biomarkers related to muscular degradation, vocal cord changes, sentiment or mood changes, and changes in the lungs or respiratory tract.
The tool discriminated COVID-19 positive patients with 97.1% accuracy, 98.5% sensitivity, and 94.2% specificity. The model performed at 100% accuracy when detecting coughs from asymptomatic positive cases.
This tool can provide a free, non-invasive, real-time large-scale COVID-10 asymptomatic screening test to augment current approaches, the researchers say. Practical uses include daily screening of students, workers, and the public, or for pool testing to quickly alert groups to outbreaks.