A new, global crowdsourced project by a Mumbai-based nonprofit research institute hopes to accelerate COVID screening, by building an AI that accurately recognises the disease by analysing cough sounds
Those with symptoms or who have tested positive for the infection, can record cough samples on a recording device, and submit it to coughagainstcovid.org
The act of coughing has never felt so unnerving, as it has since the novel Coronavirus took over our lives. The stigma aside, the anxiety that comes with being suspect, has made "the cough" the evil of all flu symptoms. Yet, it has become the silver lining for a new, exploratory project that hopes to build an Artificial Intelligence (AI) tool to assist in speedy screening of COVID-19 patients.
ADVERTISEMENT
Cough against COVID is a global data-crowdsourcing and open-innovation initiative by Mumbai-based nonprofit research outfit, Wadhwani Institute for Artificial Intelligence, that is seeing merit in analysing cough sounds, to tackle the enormous pressure that most testing labs across the globe are experiencing, due to rising number of COVID cases.
Since its launch in April, the team has been encouraging people with symptoms and those who have tested positive for the disease, to donate their cough sounds on coughagainstcovid.org, so that they can train an AI to accurately recognise a "COVID cough".
Led by Dr Rahul Panicker, who is principal investigator of Cough against COVID, and Chief Innovation and Research Officer at Wadhwani AI, the initiative has gathered enough steam to receive collaborative support from Stanford University, and Bill & Melinda Gates Foundation.
In an interview with mid-day, Dr Panicker and Amrita Mahale, product manager and strategy lead at Wadhwani AI, explain why this project could be a positive step in improving COVID testing and diagnosis.
Edited excerpts from the interview.
Amrita Mahale
The project is based on the assumption that the COVID cough has a unique sound? Is there any evidence to prove this?
Dr Panicker: There is evidence from past work that cough and voice have useful indicators for lung disease like pneumonia. There is also preliminary evidence that sound may have useful signal for detecting COVID. But this evidence is preliminary. The key here is also to have enough useful signal for practical use.
Mahale: Whether AI will really be able to pick up some, unique acoustic signatures of the cough from COVID-affected patients, is still an open question.We have collaborated with several doctors, AI researchers, engineers, and public health professionals, whose expertise has been crucial to making this informed assumption.
How are you currently collecting 'cough sound' data for the AI?
Mahale: We are collecting data at two levels. One is our crowdsourcing platform, coughagainstcovid.org. Those with symptoms or who have been tested positive for the disease, can record cough samples on their smartphone or a recording device, and submit it to us. In addition to the cough sounds, one has to submit basic demographic details like age, gender, location as well as the symptoms that they are experiencing. [If you have the COVID-19 test result on-hand, you can share that too]. Having contextual information such as the location of the person helps. For instance, if the person resides in a hotspot zone, we'd be able to surmise that s/he is more likely to be infected.
We are also doing a more focussed facility-based data collection, especially in India, at testing centres and isolation wards. And we have started some data collection already. The more the data we have, the better it is for the AI algorithm. All our data will be anonymised.
Safety is crucial to what we do. As there is already a lot of stigma around coughing, we have set guidelines about how these recordings need to be done. If one is recording it on their own, they need to make sure they are recording away from other people, and that they disinfect any surface around them after that. For facility data collection, we have stricter protocols, about distance from device when recording, and that the recording be done in a separate area. We have to be very careful.
How will the AI be trained to detect COVID sounds?
Mahale: The way the AI model works, is that you have to give it input, for it to tell you what the output is. In the training phase, [which will begin simultaneously with data collection], we train the algorithm with input [cough sound data] and output [results]; eventually it learns to predict the output, based on the given input. This is how you build any AI product—you first give it a lot of training data, where there is input and output, and eventually, the algorithm learns the co-relation between the two.
Dr Rahul Panicker
If and when the AI is ready, how will it help in screening of COVID patients?
Mahale: The COVID-19 testing strategy, especially in India, is currently very restrictive. So, if you have mild or moderate symptoms, you cannot get tested unless you have had contact with a COVID-19 positive case or if you live in a hotspot or containment area. I think developing a self-screening tool, will help allay some of the anxiety that people are experiencing. Once we get into the flu season [during the monsoon], we will have a large number of people with symptoms similar to COVID. But, even though you have a fever or cough, your probability of having COVID is very low. COVID shares symptoms with a large number of respiratory disorders. So, what we want to do is provide a lift. It will give you a more, accurate estimate of the probability of having COVID. [The screening results] will narrow down to people who are more likely [to be infected]. The focus here is on early screening, and not diagnosis.
It has been about a month since this project took off. What are the key challenges you've been facing?
Dr Panicker: The key challenge is working in the midst of a crisis. Everything from official permissions, to review on platforms like Google and Facebook, to collecting data in clinical facilities takes longer, as everyone is dealing with
the crisis.
How has the response to the project been? Are people receptive to the idea?
Mahale: There is a lot of curiosity and excitement around the project. A crowdsourced project is a citizen's science project, where people can contribute to scientific research. We have already received a lot of crowdsourced data from the US and the UK, where persons with mild, moderate COVID symptoms are in home quarantine. I think, with the uncertainty around, people want to do their bit and volunteer.
Catch up on all the latest Mumbai news, crime news, current affairs, and a complete guide from food to things to do and events across Mumbai. Also download the new mid-day Android and iOS apps to get latest updates.
Mid-Day is now on Telegram. Click here to join our channel (@middayinfomedialtd) and stay updated with the latest news
Subscribe today by clicking the link and stay updated with the latest news!" Click here!
