Joint Wipro-HCG Hospital effort can reduce load on healthcare sector, says Dr Vishal Rao
What with a second wave of Covid-19 infections expected in the country, a cough-based app to detect the presence of the virus in a carrier has come not a day too soon. HCG Hospital, in association with I-T major Wipro, has developed the novel ‘pre-screening and triaging’ tool which uses Deep Learning algorithms and Machine Learning with an accuracy of 84%. The app can process upto 1,000 samples per second, and can give a result in less than 5 seconds.
Dr Vishal Rao, Associate Dean (Academics), HCG, and member of the state Covid task force, said they had given the app to the government. “It can be useful for schools, businesses and offices alike in opening up the economy. In anticipation of a second wave of infections, this application can help with a more directed and focused approach. This will help categorize high-risk contacts, who can then be effectively isolated,” Dr Rao said, and added that the app would help reduce the financial and manpower burden on the healthcare sector.
A smartphone serves as the recording device and the risk categorization is handed out to the user in a matter of a few seconds. https://covidassist.wipro.com
How it works
There are 4 key steps to the ‘Covigilant’ app — Cough Detect, Cough Segment, Covid Detect and finally Explanation of Inference.
• For step 1, a CNN classifier is deployed. Once detected, the cough segments are separated based on the power and frequency content of the cough signal.
• For detection of Covid , a Mel spectrogram is generated from the audio to compute MFCC coefficients. This is them used as inputs to DNN for inferences.
• The output result is based on the majority voting of inference from each cough segment.
• Finally, a DNN model is used to understand the MFCC coefficients responsible for both positive and negative inferences.