Smartphone app can detect COVID-19 infection from voice • Earth.com

COVID-19 has affected people in almost every country and region around the world. Some of these are resource-poor countries, COVID No tests are available, nor are wastewater tests to warn of an impending wave of disease. In these countries, an affordable, accessible, non-invasive method of identifying people with COVID-19 could help manage and control the spread of the disease more effectively.

This is what inspired scientists from Maastricht University in the Netherlands to design a mobile phone app that identifies COVID-19 patients by the sound of their voice. One of the main symptoms of this disease is inflammation of the upper airways and vocal cords, usually resulting in a change in the patient’s voice. Scientists wondered if these symptoms could be used as an accurate diagnostic method and set out to collect data from the voices of sick and healthy people.

Their results were recently presented at the European Respiratory Society International Conference in Barcelona, ​​Spain.

Wafaa Aljbawi of the Maastricht University Data Science Institute and her supervisor Dr. Sami Simons, a respiratory specialist at Maastricht University Medical Center, and Dr. Visara Urovi, also of the Data Science Institute, used data from the Cambridge University crowd. Used. -sourcing COVID-19 sound app It included 893 audio samples from 4,352 participants, of which 308 tested positive for COVID-19. The speech was then analyzed using a technique called mel-spectrogram analysis. It identifies various audio features such as loudness, power, and aging.

“In this way, many characteristics of the participants’ voices can be decomposed,” Aljbawi said. “We built different artificial intelligence models to distinguish between the voices of COVID-19 patients and non-disease voices, and evaluated which one was the most suitable for classifying COVID-19 cases.”

Experts have found that one model called Long-Short Term Memory (LSTM) outperforms the others. Based on neural networks, LSTMs mimic the way the human brain works and recognize underlying relationships in data. It works in sequence and can store data in memory, making it suitable for modeling signals collected over time, such as speech.

at a meeting of European Respiratory SocietyAljbawi reported that the AI ​​model he decided to use was able to accurately identify positive cases of COVID-19 in 89% of cases, but the accuracy of the lateral flow test depends on the brand of test kit used. It was very different. Also, the lateral flow test was significantly less accurate in detecting her COVID infection in people who showed no symptoms.

“These promising results suggest that simple voice recordings and fine-tuned AI algorithms may be able to achieve high accuracy in identifying patients infected with COVID-19. ‘ she said. “Such a test can be provided free of charge and is easy to interpret. Moreover, it allows remote virtual testing and takes less than a minute. Allows screening.”

The app is installed on the user’s mobile phone, and to take advantage of it, the user first reports basic information about demographics, medical history, smoking status, and records breath sounds. These include three coughs, three to five deep breaths through the mouth, and his three readings of short sentences on the screen.

The app has an overall accuracy of 89%, equating to its ability to correctly detect positive cases (true positive rate or “sensitivity”). The ability to correctly identify negative cases (true negative rate or “specificity”) was 83%.

In contrast, the lateral flow test has a sensitivity of 56% but a higher specificity of 99.5%. This means that the test misclassifies an infected person as negative for her COVID-19 more often than the scientist’s AI model.

“These results demonstrate a significant improvement in diagnostic accuracy for COVID-19 compared to state-of-the-art tests such as the lateral flow test,” said Aljbawi. “While the lateral flow test has a sensitivity of only 56%, it has a high specificity of 99.5%. It is important because it means that AI LSTM models can miss 11 out of 100 cases that spread infection, whereas lateral flow testing can miss 44 out of 100 cases.”

“The high specificity of the lateral flow test means that only 1 in 100 people will be misrepresented as positive for COVID-19 when they are not actually infected. The LSTM test would result in 17 out of 100 uninfected people being misdiagnosed as positive.However, the test is virtually free, so if the LSTM test shows a positive We can invite you for a PCR test,” Aljbawi added.

Researchers say the results need to be validated with more participants. Since the start of this project, 53,449 speech samples of him from 36,116 participants have been collected and can be used to improve and validate the accuracy of the model. We are also doing further analysis to understand which voice parameters influence the AI ​​model.

The researchers conclude that deep learning can indeed detect subtle changes in the voices of COVID-19 patients. The diagnostic method is free, readily available, non-contact, non-invasive, and can be used in combination with other testing methods to rapidly and accurately diagnose cases of COVID-19. .

—

To Alison Bosman, Earth.com staff writer