COVID-19 has affected people in almost all countries and territories in the world. Some of these are low-resource countries where free COVID testing is not available, nor is the testing of waste water to give warning of imminent waves of the disease. In these countries, an affordable, accessible, non-invasive method of identifying those who have COVID-19 would help to manage and control the spread of the disease more effectively.
This is what led scientists from the Maastricht University in the Netherlands to design a mobile phone app that identifies COVID-19 sufferers from the sound of their voices. One of the main symptoms of this disease is inflammation in the upper respiratory tract and vocal cords, which usually leads to changes in the patient’s voice. The scientists wondered whether these symptoms could be used as an accurate method of diagnosis, and set about collecting data from the voices of ill and well people.
Their results were presented recently at the European Respiratory Society International Congress in Barcelona, Spain.
Ms. Wafaa Aljbawi from the Institute of Data Science at Maastricht University, and her supervisors – Dr Sami Simons, pulmonologist at Maastricht University Medical Centre, and Dr Visara Urovi, also from the Institute of Data Science – used data from the University of Cambridge’s crowd-sourcing COVID-19 Sounds App that contains 893 audio samples from 4,352 participants, 308 of whom had tested positive for COVID-19. They then analyzed the voice sounds using a technique called Mel-spectrogram analysis, which identifies different voice features such as loudness, power and variation over time.
“In this way we can decompose the many properties of the participants’ voices,” said Ms. Aljbawi. “In order to distinguish the voice of COVID-19 patients from those who did not have the disease, we built different artificial intelligence models and evaluated which one worked best at classifying the COVID-19 cases.”
The experts found that one model, called Long-Short Term Memory (LSTM), out-performed the other models. LSTM is based on neural networks, which mimic the way the human brain operates and recognizes the underlying relationships in data. It works with sequences, which makes it suitable for modelling signals collected over time, such as from the voice, because of its ability to store data in its memory.
At the congress of the European Respiratory Society, Ms. Aljbawi reported that the AI model they had decided to use was accurate in identifying positive cases of COVID-19 in 89 percent of cases, whereas the accuracy of lateral flow tests varied widely depending on the brand of test kit used. Also, lateral flow tests were considerably less accurate at detecting COVID infection in people who showed no symptoms.
“These promising results suggest that simple voice recordings and fine-tuned AI algorithms can potentially achieve high precision in determining which patients have COVID-19 infection,” she said. “Such tests can be provided at no cost and are simple to interpret. Moreover, they enable remote, virtual testing and have a turnaround time of less than a minute. They could be used, for example, at the entry points for large gatherings, enabling rapid screening of the population.”
The app is installed on a user’s mobile phone and then, to make use of it, the user first reports some basic information about demographics, medical history and smoking status, and records some respiratory sounds. These include coughing three times, breathing deeply through the mouth three to five times, and reading a short sentence on the screen three times.
The app’s overall accuracy was 89 percent, the same as its ability to detect positive cases correctly (the true positive rate or ‘sensitivity’). Its ability to identify negative cases correctly (the true negative rate or ‘specificity’) was 83 percent.
In contrast, lateral flow tests have a sensitivity of 56 percent, but a higher specificity rate of 99.5 percent. This would mean that the tests misclassify infected people as COVID-19 negative more often than the scientists’ AI model.
“These results show a significant improvement in the accuracy of diagnosing COVID-19 compared to state-of-the-art tests such as the lateral flow test,” said Ms. Aljbawi. “The lateral flow test has a sensitivity of only 56 percent, but a higher specificity rate of 99.5 percent. This is important as it signifies that the lateral flow test is misclassifying infected people as COVID-19 negative more often than our test. In other words, with the AI LSTM model, we could miss 11 out 100 cases who would go on to spread the infection, while the lateral flow test would miss 44 out of 100 cases.”
“The high specificity of the lateral flow test means that only one in 100 people would be wrongly told they were COVID-19 positive when, in fact, they were not infected, while the LSTM test would wrongly diagnose 17 in 100 non-infected people as positive. However, since this test is virtually free, it is possible to invite people for PCR tests if the LSTM tests show they are positive,” added Ms. Aljbawi.
The researchers say that their results need to be validated with larger numbers of participants. Since the start of this project, 53,449 audio samples from 36,116 participants have now been collected and can be used to improve and validate the accuracy of the model. They are also carrying out further analysis to understand which parameters in the voice are influencing the AI model.
The researchers conclude that deep-learning can indeed detect subtle changes in the voice of COVID-19 patients. This method of diagnosis is free, readily available contactless and non-invasive, and can be used, in conjunction with other testing methods, to diagnose COVID-19 cases rapidly and accurately.