Nov 04, 2020 11:12:00

AI that can detect new coronavirus infection from 'coughing sound' will be developed, 100% detection of asymptomatic patients

There are many

asymptomatic infected people with the new coronavirus infection (COVID-19), and its infectivity is said to be the strongest before the onset . For this reason, there are concerns about infection from people who are not aware of the infection, but a new AI has been developed that can detect COVID-19 infection from the 'coughing sound' even in asymptomatic patients. If this AI can be implemented in apps in the future, it may be used as an early warning system for COVID-19.

COVID-19 Artificial Intelligence Diagnosis using only Cough Recordings --IEEE Journals & Magazine
https://ieeexplore.ieee.org/document/9208795

MIT Team's Cough Detector Identifies 97% of COVID-19 Cases Even in Asymptomatic People
https://www.sciencealert.com/ai-cough-analysis-could-detect-covid-19-even-if-you-re-asymptomatic

The research was led by a research team at the Massachusetts Institute of Technology (MIT). Originally, the research team developed an AI that detects Alzheimer's disease from speech and cough, and this time, we have created a new AI that can detect COVID-19 from patients even if they are asymptomatic.

“Speaking coughs are also affected by the vocal cords and the organs around them, which can make a person's speech more coughing or, conversely, like a speaking voice when coughing. AI Can get a lot of information from fluent speeches and can detect coughs as well as gender, native language, and emotions, 'said Brian Subirana of MIT, who conducted the study.

ResNet50, a convolutional neural network with 50 layers in depth, was used to develop AI, with a dataset of words spoken in different emotional states after training a model for thousands of hours of human speech. Given a dataset of how coughs alter lung and respiratory performance. After that, multiple models were combined to make it possible to separate 'strong cough' from 'weak cough', and AI was able to identify 97.1% of 2500 COVID-19 patients. When limited to asymptomatic patients, the identification rate was 100%.

Researchers believe that this technology can help identify 'healthy' and 'unhealthy' asymptomatic patients and serve as an 'early warning system.'

Testing for COVID-19 can be painful and can be expensive in some countries. However, AI-based diagnostics have the great advantage of being non-invasive, virtually free, and quick to perform. However, AI is not designed to diagnose COVID-19 patients who are already symptomatic, and will only determine whether asymptomatic patients need testing and quarantine.

In addition, in order to confirm the existence of other factors that improve the accuracy of AI, researchers will test using more diverse data sets in the future, and how to consider privacy in case it is released as an application. We are planning to consider whether to proceed.