COVID-19 AI testing is finally finding its voice

8 Sep 2022

Written by Ellen Williams

Could lateral flow tests (LFTs) be a thing of the past? Researchers from Maastricht University (The Netherlands) have developed an app to reliably detect COVID-19 using vocal recognition alone. The AI model used has proven to be more accurate than LFTs at detecting the virus, and offers a cheap and easy-to-use alternative to traditional COVID-19 testing platforms.

COVID-19 LFTs have been rolled out in their billions, contributing to the global plastic waste crisis whilst still under delivering on test outcome accuracy. The need for precise, low-cost alternatives to LFTs, which are easily distributed, is critical for sustainable COVID-19 monitoring, particularly for low-income countries.

As discussed at the European Respiratory Society (ERS) International Congress (Barcelona, Spain), research is now turning to machine learning to guide COVID-19 testing though a new mobile phone app. In most cases of COVID-19, the infected person’s voice is altered due to the virus’s effect on the upper respiratory tract and vocal cords; the AI model uses these changes to predict whether an individual has the virus.

You may be interested in:

Using 893 audio samples from the University of Cambridge’s (UK) crowd-sourcing COVID-19 Sounds App, healthy and non-healthy participant data was used with a voice analysis technique called mel-spectrogram analysis. Participants were asked to provide basic information regarding their medical history, smoking status and demographics, then recorded a range of audio samples, which included spoken voice, coughing and breathing samples. The voice analysis technique could identify different vocal features over time, including how the participant catches their breath when speaking and the intervals of their breathing patterns.

The app had an overall accuracy of 89%; the model’s sensitivity was 89% and the specificity was 83%. Although the specificity rate appears low, the researchers suggest that the app could be used to recommend individuals for PCR testing to confirm any positive results.

Wafaa Alijbawi, researcher at the Institute of Data science, Maastricht University, told the congress:

“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. 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.”

Alijbawi continued:

“These results show a significant improvement in the accuracy of diagnosing COVID-19 compared to state-of-the-art tests such as the LFT. The LFT has a sensitivity of only 56%, but a higher specificity rate of 99.5%. This is important as it signifies that the LFT 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 LFT would miss 44 out of 100 cases”.

Although these results are based on a small sample size, the app now holds 53,449 audio samples from 36,116 participants. With more samples being added every day, the model is continually being improved to maximise its accuracy and validity. This exciting research signifies a great change in the way we monitor viral infections in the future and makes significant progress towards reducing the amount of plastic waste produced by mass testing.

Source: Eureka Press Release: https://www.eurekalert.org/news-releases/963516