Artificial intelligence (AI) may be used to detect COVID-19 infection in people’s voices through a cellular phone app, in accordance with research to be presented on Monday at the European Respiratory Society International Congress in Barcelona, Spain.
The AI model found in this research is more accurate than lateral flow/rapid antigen tests and is cheap, fast and simple to use, this means it could be found in low-income countries where PCR tests are costly and/or difficult to distribute.
Ms. Wafaa Aljbawi, a researcher at the Institute of Data Science, Maastricht University, HOLLAND, told the congress that the AI model was accurate 89% of that time period, whereas the accuracy of lateral flow tests varied widely with respect to the brand. Also, lateral flow tests were considerably less accurate at detecting COVID infection in individuals who showed no symptoms.
“These promising results claim that simple voice recordings and fine-tuned AI algorithms could achieve high precision in determining which patients have COVID-19 infection,” she said. “Such tests could be provided free and are easy to interpret. Moreover, they enable remote, virtual testing and also have a turnaround time of significantly less than a minute. They may be used, for instance, at the entry points for large gatherings, enabling rapid screening of the populace.”
COVID-19 infection usually affects top of the respiratory track and vocal cords, resulting in changes in someone’s voice. Ms. Aljbawi and her supervisors, Dr. Sami Simons, pulmonologist at Maastricht University INFIRMARY, and Dr. Visara Urovi, also from the Institute of Data Science, made a decision to investigate if it had been possible to utilize AI to investigate voices to be able to detect COVID-19.
They used data from the University of Cambridge’s crowd-sourcing COVID-19 Sounds App which has 893 audio samples from 4,352 healthy and non-healthy participants, 308 of whom had tested positive for COVID-19. The app is installed on the user’s cellular phone, the participants report some basic information regarding demographics, health background and smoking status, and are asked to record some respiratory sounds. Included in these are coughing 3 x, breathing deeply through their mouth 3 to 5 times, and reading a brief sentence on the screen 3 x.
The researchers used a voice analysis technique called Mel-spectrogram analysis, which identifies different voice features such as for example loudness, power and variation as time passes.
“In this manner we are able to decompose the countless properties of the participants’ voices,” said Ms. Aljbawi. “To be able to distinguish the voice of COVID-19 patients from those that didn’t have the condition, we built different artificial intelligence models and evaluated which worked best at classifying the COVID-19 cases.”
They discovered that one model called Long-Short Term Memory (LSTM) out-performed another models. LSTM is founded on neural networks, which mimic what sort of mind operates and recognizes the underlying relationships in data. It works together with sequences, that makes it ideal for modeling signals collected as time passes, such as for example from the voice, due to its capability to store data in its memory.
Its overall accuracy was 89%, its capability to correctly detect positive cases (the real positive rate or “sensitivity”) was 89%, and its own capability to correctly identify negative cases (the real negative rate or “specificity”) was 83%.
“These results show a substantial improvement in the accuracy of diagnosing COVID-19 in comparison to state-of-the-art tests like the lateral flow test,” said Ms. Aljbawi. “The lateral flow test includes a sensitivity of only 56%, but an increased specificity rate of 99.5%. That is important since it signifies that the lateral flow test is misclassifying infected people as COVID-19 negative more regularly than our test. Basically, with the AI LSTM model, we’re able to miss 11 out 100 cases who continue to spread the infection, as the lateral flow test would miss 44 out of 100 cases.
“The high specificity of the lateral flow test implies that only 1 in 100 people will be wrongly told these were COVID-19 positive when, actually, these were not infected, as the LSTM test would wrongly diagnose 17 in 100 non-infected people as positive. However, since this test is virtually free, you’ll be able to invite people for PCR tests if the LSTM tests show they’re positive.”
The researchers say that their results have to be validated with good sized quantities. Because the start of the project, 53,449 audio samples from 36,116 participants have been collected and may be used to boost and validate the accuracy of the model. Also, they are undertaking further analysis to comprehend which parameters in the voice are influencing the AI model.
In another study, Mr. Henry Glyde, a Ph.D. student in the faculty of engineering at the University of Bristol, showed that AI could possibly be harnessed via an app called myCOPD to predict when patients with chronic obstructive pulmonary disease (COPD) might suffer a flare-up of these disease, sometimes called acute exacerbation. COPD exacerbations can be quite serious and so are connected with increased threat of hospitalization. Medical indications include shortness of breath, coughing and producing more phlegm (mucus).
“Acute exacerbations of COPD have poor outcomes. We realize that early identification and treatment of exacerbations can improve these outcomes therefore we wished to determine the predictive ability of a trusted COPD app,” he said.
The myCOPD app is really a cloud-based interactive app, produced by patients and clinicians and can be acquired to utilize in the UK’s National Health Service. It had been established in 2016 and, up to now, has over 15,000 COPD patients deploying it to greatly help them manage their disease.
The researchers collected 45,636 records for 183 patients between August 2017 and December 2021. Of the, 45,007 were records of stable disease and 629 were exacerbations. Exacerbation predictions were generated someone to eight days before a self-reported exacerbation event. Mr. Glyde and colleagues used these data to teach AI models on 70% of the info and test drive it on 30%.
The patients were “high engagers,” who was simply utilizing the app weekly over months as well as years to record their symptoms along with other health information, record medication, set reminders, and also have usage of up-to-date health insurance and lifestyle information. Doctors can measure the data with a clinician dashboard, enabling them to supply oversight, co-management and remote monitoring.
“The newest AI model we developed includes a sensitivity of 32% and a specificity of 95%. Which means that the model is great at telling patients if they are not going to experience an exacerbation, which might help them in order to avoid unnecessary treatment. It really is less proficient at telling them if they are going to experience one. Improving this is the focus of the next thing of our research,” said Mr. Glyde.
Speaking prior to the congress, Dr. James Dodd, Associate Professor in respiratory medicine at the University of Bristol and project lead, said: “To your knowledge, this study may be the to begin its kind to model real life data from COPD patients, extracted from the widely deployed therapeutic app. Consequently, exacerbation predictive models generated out of this study have the potential to be deployed to thousands more COPD patients after further safety and efficacy testing. It could empower patients to possess more autonomy and control over their health. That is also a substantial benefit for his or her doctors therefore a system may likely reduce patient reliance on primary care. Furthermore, better-managed exacerbations could prevent hospitalization and alleviate the responsibility on medical care system. Further study is necessary into patient engagement to find out what degree of accuracy is acceptable and how an exacerbation alert system works used. The introduction of sensing technologies may further enhance monitoring and enhance the predictive performance of models.”
Among the limitations of the analysis is the few frequent users of the app. The existing model takes a patient to input a COPD assessment test score, complete their medication diary and report they’re having an exacerbation accurately days later. Usually, only patients that are highly engaged with the app, deploying it daily or weekly, can offer the quantity of data necessary for the AI modeling. Furthermore, because you can find a lot more days the users are stable than if they are experiencing an exacerbation, there exists a significant imbalance between your exacerbation and non-exacerbation data available. This results in even more difficulty in the models correctly predicting events after training with this imbalanced data.
“A recently available partnership between patients, clinicians and caregivers to create research priorities in COPD discovered that the highest-rated question was how exactly to identify improved ways to prevent exacerbations. We’ve centered on this question, and we’ll be working closely with patients to create and implement the machine,” concluded Mr. Glyde.
Chair of the ERS Science Council, Professor Chris Brightling, may be the National Institute for Health insurance and Care Research (NIHR) Senior Investigator at the University of Leicester, U.K., and had not been involved with the study. He commented: “Both of these studies also show the potential of artificial intelligence and apps on cell phones along with other digital devices to produce a difference in how diseases are managed. Having more data designed for training these artificial intelligence models, including appropriate control groups, along with validation in multiple studies, will enhance their accuracy and reliability. Digital health using AI models presents a thrilling opportunity and will probably impact health care.”
More info:  Abstract no: OA1626, “Creating a multivariate prediction model for the detection of COVID-19 from crowd-sourced respiratory voice data”, presented by Wafaa Aljbawi in “Digital medicine for COVID-19” session, 08.15-09.30 hrs CEST on Monday 5 September 2022, https://k4.ersnet.org/prod/v2/Front/Program/Session?e=377&session=14843 Abstract no. PA2728, “Exacerbation predictive modelling using real-world data from the myCOPD app”, presented by Henry Glyde, thematic poster “Digital health interventions in respiratory practice”, 13.00-14.00 hrs CEST on Monday 5 September 2022, https://k4.ersnet.org/prod/v2/Front/Program/Session?e=377&session=14775
Citation: Cellular phone app accurately detects COVID-19 infection in people’s voices by using AI (2022, September 5) retrieved 5 September 2022 from https://medicalxpress.com/news/2022-09-mobile-app-accurately-covid-infection.html
This document is at the mercy of copyright. Aside from any fair dealing for the intended purpose of private study or research, no part could be reproduced minus the written permission. This content is provided for information purposes only.