Australian computer scientists have developed a novel Artificial Intelligence-based model that can hear the effects of Covid-19 in the sound of a forced cough, even when people are asymptomatic, an advance that can pave the way for detecting the infectious disease via diagnostic mobile phone apps.
During the pandemic, many crowdsourcing platforms have been designed to gather respiratory sound audios from both healthy and Covid-19 positive groups for research purposes.
A team of researchers from RMIT University accessed datasets from two of these platforms — Covid-19 Sounds App and COSWARA — to train the algorithm using contrastive self-supervised learning, a method by which a system works independently to encode what makes two things similar or different.
With further development, their algorithm could power a diagnostic mobile phone app, said lead author Hao Xue, Research Fellow in RMIT’s School of Computing Technologies.
“We’ve overcome a major hurdle in the development of a reliable, easily-accessible and contactless preliminary diagnosis tool for Covid-19,” said Xue, Research Fellow in RMIT’s School of Computing Technologies.
“This could have significant benefit in slowing the spread of the virus by those who have no obvious symptoms. A mobile app that can give you peace of mind during community outbreaks or prompt you to seek a Covid test — that’s the kind of innovative tool we need to better manage this pandemic,” Xue added.
Xue said the method they developed could also be extended for other respiratory diseases like tuberculosis.
While this is not the first Covid cough classification algorithm to be developed, the RMIT model outperforms existing approaches.
According to co-author Professor Flora Salim, previous attempts to develop this type of technology, like those at MIT and Cambridge, relied on huge amounts of meticulously-labeled data to train the AI system.
“The annotation of respiratory sounds requires specific knowledge from experts, making it expensive and time-consuming, and involves handling sensitive health information,” she said.
Moreover, cough samples from one hospital or one region to train the algorithm also limits its performance outside that setting.
“What’s most exciting about our work is we have overcome this problem by developing a method to train the algorithm using unlabelled cough sound data. This can be acquired relatively easily and at a larger scale from different countries, genders and ages,” Salim noted.