Mobile phones use AI to hear “sounds” to identify COVID-19

Artificial intelligence (AI) can detect COVID-19 from people’s voices through a mobile phone app, according to a study announced at the European Respiratory Society International Conference in Barcelona, ​​Spain, on the 4th. infection, it is more accurate (up to 89%) than rapid antigen tests and is cheaper, faster and easier to use.

COVID-19 infection often affects the upper respiratory tract and vocal cords, causing changes in a person's voice. Wafaa Aljebawi, a researcher at the Institute of Data Science at Maastricht University in the Netherlands, explained that the research results show that simple voice recordings and AI algorithms can accurately determine who is infected with COVID-19. In addition, it also supports remote virtualization. The test takes less than a minute to produce results. Such tests can be used at testing sites at large gatherings to quickly screen people.

The research team used data from the University of Cambridge’s “COVID-19 Sound Bank” application, which contains 893 audio samples from 4,352 healthy and non-healthy participants, 308 of whom had COVID-19. Tested positive. The app is installed on the user's phone, and participants report basic information about demographics, medical history, and smoking status, and then are asked to record a number of sounds, including coughing 3 times, breathing deeply through the mouth 3-5 times, and reading on the screen. A short sentence 3 times.

The researchers used a speech analysis technique called mel spectroscopy, which identifies different speech characteristics such as loudness, power and changes over time.

To distinguish the voices of COVID-19 patients from those without the disease, researchers built different AI models. They found that long short-term memory (LSTM) models did the best job of classifying COVID-19 cases. LSTM is based on neural networks, which mimic the way the human brain works and identify underlying relationships in data. It can also store data in memory.

The overall accuracy of this AI-LSTM model is 89%, the ability to correctly detect positive cases (true positive rate or sensitivity) is 89%, and the ability to correctly identify negative cases (true negative rate or specificity degree) is 83%.

Researchers said that the sensitivity of the rapid antigen test is only 56%, but the specificity is as high as 99.5%. This means that rapid antigen tests incorrectly classified more positive infections as negative than were classified in this test. Using the AI-LSTM model, the researchers missed 11 out of 100 cases who went on to spread the virus, while the rapid antigen test missed 44. Intern reporter Zhang Jiaxin

The above is the detailed content of Mobile phones use AI to hear “sounds” to identify COVID-19. For more information, please follow other related articles on the PHP Chinese website!