Researchers look for ‘sound’ of corona from Odisha to Bengaluru

Researchers involved in two ongoing projects, one of which involves the governments of Odisha and Bihar, are spending hours collecting sounds of people coughing. Their aim is simple: to use artificial intelligence (AI) to determine if the coronavirus can be ‘heard’ in a person’s cough, and to then develop a tool that can place individuals on a Covid-19 risk scale by checking for these acoustic markers.
With ICMR’s go-ahead, both projects are using two methods to collect cough sounds — through crowdsourcing, in which anybody can record his/her cough and upload it on the projects’ servers; and through focused cough sound collection from hospitals and testing centres.
The ‘Cough against Covid’ launched by Mumbai-based non-profit, Wadhwani Institute of Artificial Intelligence, encourages people — Covid patients, those negative and those not tested — to ‘donate a cough’ by recording it on a phone. Supported by the Bill and Melinda Gates Foundation, the campaign aims to create a large dataset of around 30,000 cough sounds from participants in India and abroad. AI has been deployed to look for signs that can distinguish a Covid cough from other coughs. The project aims to create an open-source cough sound analysis tool that people can use — through an app or a bot — to self-screen.
“We hope this can amplify testing capacity and bring limited tests to people who need them the most,” said Rahul Panicker, chief research and innovation officer at the non-profit.
To ensure development of a robust AI tool, the backend dataset is being strengthened by Wadhwani AI (WAI) through collection of patient cough sounds at facilities in Bihar and Odisha. These activities are being undertaken by state health departments in collaboration with the Norway-India Partnership Initiative (NIPI), a bilateral health sector initiative of the Indian and Norwegian governments.
“The Bihar and Odisha governments are working with NIPI and the non-profit to conduct research around developing AI tool for triaging suspect Covid-19 cases using cough sound acoustics. This, if found useful, can possibly be integrated with the Aarogya Setu app,” said a senior official related with the project. “At designated sites, suspected patients are being requested by facility health staff to provide three ‘solicited’ cough sound samples which further get labelled upon availability of the Covid test results,” the official added.
Working on a similar hypothesis is IISc Bengalore. The institute’s ‘Coswara’ project also aims to bring out an AI-based cough sound analysis tool by October. ‘Coswara’ too is crowdsourcing data by asking healthy and Covid-afflicted persons to ‘donate a cough’. Till July 1, it had collected 1,100 cough sounds; of them, around 20 are of Covid patients. “Once we get to around 150 coughs from Covid patients, we will deploy mathematical modelling to come up with an initial screening tool, which we will present to ICMR,” said project member Sriram Ganapathy of IISc’s electrical engineering department.
IISc, too, is looking to improve its crowdsourced data by collecting more cough sounds from certified Covid patients. “Our proposal is being reviewed by boards of hospitals in Mumbai and Bengaluru and we are also handing out fliers outlining our aims. We hope patients, especially those at home, will send us their cough sound,” said Ganapathy.
Both teams stressed data security was paramount and that they are aiming to design a screening tool, not a diagnostic one. Doctors TOI spoke to said the research looked promising but the final tool would need to be very sophisticated. “The Covid cough is not as unique as, say, the whooping cough of Pertussis. So, it might be difficult to screen for SARS-CoV-2 using cough sound alone,” was the note of caution struck by Prasanta R Mohapatra, HOD of pulmonary medicine at AIIMS Bhubaneswar.
Inspired By