IISc experts to collect sound samples of Covid patients’ cough, breath
Over the next couple of months, researchers at the Bengaluru-based Indian Institute of Science (IISc) will collaborate with hospitals, nodal and district health centres across India to develop a database of sound samples — cough, breath, and voice — from Covid-19 patients as a first-line diagnostic tool.
Last week, IISc received an in-principal nod from the Indian Council for Medical Research (ICMR) to collect respiratory sound data by tying up with hospitals treating Covid-19 patients.
Researchers said marrying this database, christened Coswara, with machine learning techniques and expertise from doctors treating respiratory illnesses could serve as a pre-screening mobile or web-based tool to help diagnose individuals potentially infected with the Sars-Cov2 virus that causes Covid-19.
“Our data will be generated in an Indian context because voice and cough sounds are affected by various local conditions such as weather and pollution. Using data from an external world can therefore be problematic since our population data may not match algorithms developed to build the Covid-19 diagnostic tool. We would like to enrol hospitals and health centres in the five major cities with high infection rates,” said Sriram Ganapathy, assistant professor, department of electrical engineering, IISc, who is spearheading project Coswara.
The institute’s work assumes importance with the World Health Organisation (WHO) and America’s Centre for Disease Control (CDC) having listed dry cough, difficulty in breathing, chest pain or pressure, and loss of speech or movement as key symptoms of Covid-19, which start showing between two to 14 days after exposure to the virus. A recent modelling study of data collected from a pool of 7,178 Covid-positive individuals validated the presence of these symptoms, and proposed a real-time prediction and tracking approach. The study was published in internationally peer-reviewed journal, Nature Medicine.
At present, nasal or throat swab samples of symptomatic individuals are tested with reverse transcription polymerase chain reaction (RT-PCR) method — considered a gold standard for Covid-19 diagnosis — and the results are available between two and 48 hours.
“The limitations of the RT-PCR method, however, include violating physical distancing rules that can increase the chances of the spread of the infection in addition to it being time-consuming, expensive and the difficulties involved with large-scale deployment,” said Ganapathy. “There is a need for an alternate point-of-care diagnostic tool that can be widely deployed to supplement existing chemical testing capacities given the pandemic is expected to stay for a while.”
The eight-member team has so far remotely collected respiratory sound samples (https://coswara.iisc.ac.in/) from 1,000 participants, including healthy individuals, those with respiratory infections and Covid-19 patients via worldwide crowd sourcing using a website application to establish proof of their concept.
The sound data comprises nine categories: Breathing (shallow and deep); cough (shallow and heavy); three kinds of sustained vowel phonation; and counting 1 to 20 (normal and fast-paced).
Analysis of data using signal processing and machine learning techniques will be followed by building a mathematical model to develop biomarkers for Covid-19 from sound samples. As a mobile or web-based application, the user can record her/his voice samples for analysis and this will be used to predict whether or not the sample is similar to a Covid-19 infection. Researchers said the final deployment would be subject to validation with clinical findings, and authorisation or approval from ICMR.
In addition to IISc, at least three research groups at Carnegie Mellon in the United States of America, Cambridge University in the United Kingdom and École Polytechnique Federale de Lausanne in Switzerland are also working on similar Covid-19 pre-screening diagnostic tools.