New AI model can detect coronavirus by listening to a cough
MIT researchers claim to have developed a new AI model that can detect if an individual has - just by listening to them cough.The infection detection works even for people who are completely asymptomatic.
The model uses four biomarkers - vocal cord strength, lung and respiratory performance, sentiment, and muscular degradation - to decide if the person is infected with the virus.
"We developed an AI speech processing framework that leverages acoustic biomarker feature extractors to pre-screen for COVID-19 from cough recordings, and provide a personalised patient saliency map to longitudinally monitor patients in real-time, non-invasively, and at essentially zero variable cost," .
Coronavirus almost always impacts the lungs and vocal cords, making an infected person's cough sound slightly different to that of a healthy individual. The researchers say their AI model can detect these subtle differences, while the human ear cannot.
Researchers collected nearly 200,000 recordings of coughs to train their model, including 2,500 from confirmed coronavirus positive patients. The study participants were also asked to disclose if they had had any symptoms of the disease recently.
Researchers say their model accurately identified 98.5 per cent of coughs from individuals who tested positive for the virus, and 100 per cent from those who were asymptomatic.
They suggest incorporating the new AI technology into an app to create a free, large-scale, non-invasive and real-time Covid-19 screening tool to support measures to contain the spread of COVID-19 pandemic. Such a tool could be used to screen employees and students daily as they return to their offices and schools - although any such app would need approval.
The detailed findings of the research are published in the IEEE Journal of Engineering in Medicine and Biology.
Last month, Nvidia also announced that it was building a new .
The supercomputer, dubbed Cambridge-1, will analyse millions of molecules before deciding which are most likely to be useful in clinical trials, thus speeding up drug-discovery processes that would otherwise take many years to complete.