Covid test is not a cough nut to track, finds IISc

Coswara tool to detect covid through voice samples is 70-80% accurate; that dreaded nasal poking may not be required soon

One silver lining to the covid pandemic is the number of technological innovations coming out of Bengaluru.

IISc researchers who have been working on a sound-based covid-19 diagnostic tool named Coswara, have reported about 70 to 80 per cent accuracy compared to the RT-PCR test, the gold standard for covid detection.

(Bangalore Mirror had reported about another initiative of IISc — ARTPARK — which had developed a robotic nurse as a tribute to the hundreds of fearless of ASHA workers who have been testing, tracing and helping covid infectees).

Coswara detects covid based on respiratory, cough and speech sounds. Work on Coswara has been in progress since April 2020 and the researchers involved said that the tool has the potential to evolve into an effective diagnostic tool based on the recent results.

“So far, the results of the models developed on the Coswara data are seen to provide about 70 to 80 per cent accuracy against the RT-PCR gold standard. For example, the widely used rapid antigen testing (RAT) is about 75 per cent accurate compared to the RT-PCR standard. Given the highly simplistic nature of sound-based testing, the tool has the potential to evolve into effective diagnostic equipment,” said Sriram Ganapathy, head of the Coswara project.

He added that the next steps in the development of the tool involve improving the machine-learning algorithms for better detection and also for reducing false positives.

“The development of the backend model will be integrated in the existing web tool. Once the models are finalised, we will proceed to seek formal approvals from agencies like Indian Council of Medical Research (ICMR). Following approval, the tool will be made available to users, free of cost. The exact timelines are difficult to predict as they are contingent on government agency approvals,” he added.


What does an infection sound like? IISc knows

The team, led by Sriram Ganapathy, aims to create a data set composed of voice samples from a mix of healthy individuals as well as from those affected with the virus using a webpage ( and a mobile app.

A few months ago, Massachusetts Institute of Technology (MIT) had developed an Artificial Intelligence (AI) model to detect asymptomatic covid patients through their cough recordings.

MIT’s AI model to detect covid patients through cough recordings was said to be similar to Coswara’s. That apart, scientific groups based in South Africa have used Coswara data to build a diagnostic tool for covid-19, Sriram said.

“Further, the data we collected and analyzed has also been released to researchers across the world as a special session challenge in an upcoming conference. This ongoing challenge, termed the DiCOVA challenge is witnessing widespread participation from the scientific community in the biomedical field from India and across the world. The progress made in this special session (concluding by end of March) would further ramp up the efforts in tool development and advance the detection accuracy,” he added.

Asked about the relevance of a tool like Coswara in the current times when the vaccination drive is on, the team pointed to the recent spike in cases in Maharashtra and in South Africa where the vaccination drive has been stopped.

“South Africa has formally stopped the vaccination drive as the vaccines were ineffective to the variants seen there. Hence, we find that the need for simple, fast, and cost-effective testing will be of interest in the coming months and years,” said the team.

Source link


Sagar Biswas

Leave a Reply

Close Bitnami banner