Researchers from the
Massachusetts Institute of Technology (MIT) have recently developed an
Artificial Intelligence (AI) model to detect asymptomatic
covid patients through their cough recordings.
The MIT findings, which were published in the IEEE Journal of Engineering and Biology, states that the AI model distinguished asymptomatic patients from healthy individuals through forced cough recordings, which were collected via mobile phones.
A team of Indian Institute of Science (IISc) researchers who have been working on a tool for diagnosis of covid based on respiratory, cough and speech sounds since April, say the recent MIT findings is exactly in line with their research on the sound-based diagnostics tool, Coswara.
“The MIT group’s efforts used voice samples with cell-phone recordings along with personally reported health status using a web link. Their study seems to report a good accuracy of covid diagnostics based on cough samples analysed using machine learning models,” Sriram Ganapathy, head of the IISc research team, told Bangalore Mirror.
He added that the Coswara team’s study relies on similar efforts and they had started work on this much before the MIT team began theirs. “Our study, which started from April 2020, is also based on recording of a participant’s health status and acoustics data for different categories of speech, breathing and cough sounds. The data collection is ongoing along with the analysis. Through this, the MIT results provide a proof-of-concept to advance our efforts towards a tool,” he added.
The voice-based analysis has the advantage of quick, cost-effective and safe diagnostics using only a smartphone
–Sriram Ganapathy, head of IISc research team
Over the last two months, the Coswara team has been looking at data using various signal analysis tools. Ganapathy said the team had looked at a pool of 100-odd parameters and their properties for each signal classes (coughing/breathing/speech).
“This analysis also went on to identify noisy samples and remove them out for improved training of machine systems. Further, the analysis came up with a subset of parameters that can meaningfully quantify healthy individuals against covid patients. These parameters are now being used to train machine learning models to illustrate the specificity and sensitivity results. The statistical analysis performed thus far indicates that multiple parameters from different signals have the potential to detect covid infection with good accuracy,” he said.
So far, the team has collected over 1,400 samples, which includes about 100 covid-positive samples and more than 50 samples of individuals with respiratory disorders.
Ganapthy added that AI has a vital role in covid detection and they are much safer compared to other methods of analysis that detect the virus.
“Analysis of X-ray and CT-scan images have shown high accuracy for detection of covid infection. However, the process requires
the participants to visit a health facility, thereby compromising the safety of the subject and the healthcare workers. The voice-based analysis has the advantage of quick, cost-effective and safe diagnostics using only a smartphone,” Ganapathy added.