Typically, SARS-CoV-2 targets an infected person’s nose, throat and vocal cords first, which is one of the reasons it is considered an upper respiratory disease. In a new observational study, Dutch data science researchers have found a way to exploit this fact to develop better testing options.
The scientists created an AI model that can detect SARS-CoV-2 infection in people’s voices through a mobile phone app even when no symptoms are present. The model is accurate 89% of the time—which is significantly higher than current lateral flow/rapid antigen tests.
The researchers used data from the University of Cambridge’s crowd-sourced COVID-19 Sounds App, which contains 893 audio samples from 4,352 healthy and non-healthy participants, 308 of whom had tested positive for COVID-19. Through the app, participants report basic information, such as demographics, medical history and smoking status, and then are asked to record respiratory sounds. These include coughing three times, breathing deeply through their mouth three to five times, and reading a short sentence on the screen three times.
In order to distinguish the voice of COVID-19 patients from those not infected, the Dutch team built multiple artificial intelligence models to test.
According to their research, presented at the European Respiratory Society International Congress, the Long-Short Term Memory (LSTM) outperformed other models. LSTM is based on neural networks, which mimic the way the human brain operates and recognizes the underlying relationships in data.
LSTM’s overall accuracy was 89%, its ability to correctly detect positive cases (sensitivity) was 89%, and its ability to correctly identify negative cases (specificity) was 83%.
Comparatively, current lateral flow tests have an average sensitivity of 56%, but a high specificity rate of 99.5% on average—although accuracy varies depending on brand.
“The lateral flow test is misclassifying infected people as COVID-19 negative more often than our [AI] 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,” explained researcher Wafaa Aljbawi at the Institute of Data Science, Maastricht University, The Netherlands.
In addition to improved accuracy, the AI model is also quicker and less expensive than lateral flow tests, which means it can be more easily accessed in low-income countries where PCR tests can be difficult to distribute.
“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,” said Aljbawi. “They could be used, for example, at the entry points for large gatherings, enabling rapid screening of the population.”
Aljbawi and colleagues say their results need to be validated with large numbers. Since the start of this project, 53,449 audio samples from 36,116 participants have been collected and can be used to improve and validate the accuracy of the model. The researchers are also carrying out further analysis to understand exactly which parameters in the voice are influencing the AI model.
Further research presented at the European Respiratory Society International Congress showed AI models are suited for even more than COVID-19 diagnosis. Henry Glyde, an engineering researcher at the University of Bristol, also developed an AI model that could be harnessed via a popular app to predict when patients with chronic obstructive pulmonary disease (COPD) might suffer a flare-up.
The myCOPD app is a cloud-based interactive app developed by patients and clinicians that is available to use in the UK’s National Health Service. It was established in 2016 and, so far, has over 15,000 COPD patients using it to help manage their disease.
Glyde and his team collected 45,636 records for 183 patients between August 2017 and December 2021. Of these, 45,007 were records of stable disease and 629 were exacerbations. Exacerbation predictions were generated one to eight days before a self-reported exacerbation event. Glyde and colleagues used this data to train AI models on 70% of the data and test it on 30%.
Glyde’s most recent AI model boasts a sensitivity of 32% and a specificity of 95%, meaning it is very good at telling patients when they are not about to experience an exacerbation, but not as good at warning them when they are about to experience one. While it will help patient avoid unnecessary treatment, Glyde said greatly improving the 32% sensitivity is the focus of the next phase of his research.
“This study is the first of its kind to model real world data from COPD patients, extracted from a widely deployed therapeutic app. As a result, exacerbation predictive models generated from this study have the potential to be deployed to thousands more COPD patients after further safety and efficacy testing,” said project lead James Dodd, associate professor in respiratory medicine at the University of Bristol. “Further study is required into patient engagement to determine what level of accuracy is acceptable and how an exacerbation alert system would work in practice. The introduction of sensing technologies may further enhance monitoring and improve the predictive performance of models.”