Mumbai institute develops AI-based solution to screen COVID-19 infections

The Mumbai-based institute was researching this aspect since the beginning of the pandemic, collecting sound samples to build an algorithm to develop a tool 'cough against COVID' that uses artificial intelligence to analyse every sound sample. In a statement, the institute said it has secured a US patent for the tool and claimed that it can also identify asymptomatic people.

The sound created while coughing can possibly reveal whether a person is infected with COVID-19 virus or not, Wadhwani Institute for Artificial Intelligence claimed on Tuesday. The Mumbai-based institute was researching this aspect since the beginning of the pandemic, collecting sound samples to build an algorithm to develop a tool 'cough against COVID' that uses artificial intelligence to analyse every sound sample.

In a statement, the institute said it has secured a US patent for the tool and claimed that it can also identify asymptomatic people. India's novel coronavirus tally crossed 49 lakh on Tuesday with 83,809 new cases.

Wadhwani's chief research and innovation officer Rahul Panicker said the institute started thinking about a non-invasive AI-based COVID screening tool amid the challenges of testing. "We believe it (the patented tool) will help health care and civic authorities speed up testing and also focus their resources better, by filtering out patients with COVID-19 like symptoms but without the infection," he said.

The tool can increase the testing capacity of a healthcare system by 43 per cent without additional supplies, trained personnel, or physical infrastructure with the assumption of disease prevalence at 5 per cent, the statement added. Panicker said the institute will continue to work towards making the tool accessible to those who need it most and will also be sharing the research widely so that people around the world can benefit from it.

The institute said it has achieved a balance between high sensitivity and low specificity and the tool can be used as a "COVID-19 triaging tool" or a method to identify the order of treatment for a patient. The institute partnered with Norway India Partnership Initiative (NIPI), Doctors for you (DFY), AIIMS Jodhpur, and Municipal Corporation of Greater Mumbai (MCGM) to collect data of 3,621 individuals from Bihar, Odisha, Rajasthan and Maharashtra as part of the research, of which 2,041 tested positive for COVID-19.

Open-source non-COVID cough data-sets were also used to obtain 31,909 sounds segments, of which 27,116 were non-cough respiratory sounds like wheezes, crackles or breathing, or human speech, and 4,793 were cough sounds, it said. An end-to-end deep Convolutional Neural Networks-based framework was used to ingest audio samples and predict a binary classification label, which would indicate the probability of the presence of COVID-19, it explained.

"The findings demonstrated that solicited-cough sounds collected over a phone can be analysed by the 'Cough against COVID' AI tool and provide a statistically significant signal that is indicative of COVID-19 status," it said..