A team of scientists, experts in artificial intelligence (AI), has developed a mobile application (app) that detects cases of COVID-19 in people’s voices more accurately and quickly than antigen tests used so far.
The system, which was presented this Monday at the International Congress of the European Respiratory Society in Barcelona, is also cheaper than antigen tests, which means that it could be used in low-income countries where these tests are expensive or difficult. to get
According to Wafaa Aljbawi, a researcher at the Institute of Data Science at the University of Maastricht (The Netherlands), this AI model has an accuracy of 89%, a percentage that in the case of tests varies depending on the brand.
“Our results are promising and suggest that voice recordings and fine-tuned AI algorithms can be highly accurate in determining which patients have COVID-19 infection,” he says.
“These tests are free and easy to interpret. In addition, they can be remote virtual tests and their response time is less than a minute, so they could be used, for example, at entry points for large gatherings to ensure rapid detection in the population, “he advances.
COVID-19 infection typically affects the upper respiratory tract and vocal cords, causing changes in a person’s voice.
From there, Aljbawi, Sami Simons, a pulmonologist at the Maastricht University Medical Center, and Visara Urovi, from the Data Science Institute, investigated whether it was possible to use AI to analyze voices and detect infections.
To do this, they used the open application “COVID-19 Sounds”, created by the University of Cambridge to study the symptoms of the coronavirus, a database that contains 893 audio samples from 4,352 healthy and unhealthy participants, 308 of whom tested positive. for the disease
The application is installed on the user’s mobile phone, the participants must give basic information, and data on their medical history and habits such as smoking, and then they are asked to record their breathing, their cough and their voice.
Using the voice analysis technique called Mel’s spectrogram analysis, which identifies different characteristics of the voice, such as loudness, power and variation, they were able to break down the different properties of the participants’ voices.
Then, to distinguish the voice of COVID-19 patients from healthy ones, the scientists built different artificial intelligence models and studied which one worked best to classify cases.
The Long-Short Term Memory (LSTM) model, based on neural networks that mimic the way the human brain operates and recognize underlying relationships in the data, achieved an accuracy of 89% correctly detecting positive cases and 83% % in the negatives.
The results of the study will be validated in a larger study with the 53,449 audio samples from 36,116 participants.
In a second study, Henry Glyde of the University of Bristol has shown that AI (via an app called “myCOPD”) could predict exacerbations (severe flare-ups) in patients with chronic obstructive pulmonary disease (COPD).
“MyCOPD” is an interactive application developed by patients and doctors that has been available for use in the UK National Health Service since 2016 and currently helps more than 15,000 COPD patients to manage their disease.
The researchers collected 45,636 records from 183 patients between August 2017 and December 2021 (45,007 records of stable disease and 629 exacerbations) and used this data to train AI models.
“The most recent AI model we developed has a sensitivity of 32% and a specificity of 95%. This means that the model is very good at telling patients when they are not going to experience an exacerbation, which can help them avoid unnecessary treatment,” Glyde concludes.
