2 min read.Updated: 25 Sep 2020, 11:26 AM ISTSeethalakshmi S
The app captures audio and self-reported data such as gender, age and other health indicators, and uploads it to a secure cloud server for analysis using artificial intelligence technology
Bengaluru: Research teams from IT major Wipro and cancer care specialists HCG have partnered to study cough patterns of covid-19. They’re working on an app that can distinguish the covid cough from other types of cough caused by seasonal bronchitis and pneumonia.
Cough patterns in respiratory illnesses show variations and have been used earlier to diagnose illnesses. Since the SARS-CoV2 virus, which causes covid-19, targets the respiratory system, cough patterns from infected individuals could be distinct is the reasoning of the researchers.
“Cough is a common symptom in covid-19. We believe specific features to covid-19 cough could be used to identify cases. Successful identification and preventive measures could help us contain the spread with early detection," said Dr Anand Subash, head and neck oncologist, who is leading research for the AI Cough Recorder project.
The app captures audio and self-reported data such as gender, age and other health indicators, and uploads it to a secure cloud server for analysis using artificial intelligence technology. The signals are analysed by an algorithm to identify audio signatures specific to people who are infected with the virus. The idea is to help doctors speed up decision making in times of emergencies.
The app is not a diagnostic tool but can help “identify essential cough signatures that can quickly, inexpensively, and detect high-risk patients". This non-intrusive patient screening technique could ease the burden on healthcare professionals.
The Wipro research team is working with clinicians and speech pathologists from HCG and covid hospitals like Bangalore Medical College and Research Hospital and Bhaktivedanta Hospital in Mumbai.
The researchers are focusing more on classifying covid and non-covid coughs. “In future, we will be developing models to identify covid cough, normal cough and other diseases such as pneumonia, which will help the medical community make decisions quickly. As a data security measure, we are working on encrypting the data and processing it in the cloud. This is of high importance as far as data and subject privacy is concerned. The model size is 9MB which can be easily embedded in an Android or iOS app," a researcher at Wipro Technovation Centre explained.
“A number of machine learning covid-19 model projections have been developed using the accepted cough audio samples. Each model examines the audio files, to determine if the tester shows an audio signature that indicates if he/she is infected with the virus," says Dr U S Vishal Rao, a surgical oncologist at HCG.
A set of 1,553 validation samples was used. The performance of the model was 95% with just two false negatives and 75 false positives (negative samples identified as positive).
Over time, the team hopes to build a large covid-19 cough dataset. “This modelling data can then be leveraged to continually improve model projections and predictions. It can be used for screening in public places, airports, malls, restaurants to arrest disease spread," says Dr Subhas.
Subscribe to Mint Newsletters
* Enter a valid email
* Thank you for subscribing to our newsletter.
Never miss a story! Stay connected and informed with Mint.
our App Now!!