Home > Pulmonology > ERS 2022 > COVID-19: What Is New? > Accurate voice-based COVID-19 diagnostic test in development

Accurate voice-based COVID-19 diagnostic test in development

Presented By
Ms Wafaa Aljbawi, Maastricht University, the Netherlands
Conference
ERS 2022
Doi
https://doi.org/10.55788/d7edae4e

A long short-term memory (LSTM) model incorporating speech recognition demonstrated to provide a potentially cheaper, faster, and more precise diagnostic test for COVID-19 than a rapid antigen test (RAT).

Ms Wafaa Aljbawi (Maastricht University, the Netherlands) and colleagues aimed to develop and implement an effective, automated, and non-invasive diagnostic system for COVID-19 [1]. The research team used a database from Cambridge University that included data on the demographics, medical history, symptoms, and COVID-19 test results of registered participants. Furthermore, voice recordings were obtained from 893 participants, who recorded the sentence ‚ÄėI hope my data can help to manage the virus pandemic.‚Äô Of these voice recordings, 308 belonged to COVID-19-positive individuals. After the noise was removed from the recordings, the edited audio files were entered into an LSTM model, predicting COVID-19 status based on these recordings.

The LSTM model had an accuracy of 89%, a sensitivity of 89%, and a specificity of 83%. Ms Aljbawi emphasised that the sensitivity of 89% is high compared with that of a RAT test, which has an average sensitivity of 56.2%. In other words, the LSTM model may miss 11 out of 100 cases of COVID-19, whereas a RAT may miss 44 out of 100 cases. In contrast, the specificity of a RAT test is higher at 99.5%, compared with 83% of the LSTM model. This means that only 1 out of 100 individuals would be wrongly diagnosed with COVID-19 if they used a RAT, whereas the current LSTM model may wrongly detect COVID-19 in 17 out of 100 non-infected individuals.

‚ÄúWe would rather misdiagnose patients with COVID-19 than miss cases of COVID-19,‚ÄĚ commented Ms Aljbawi. Although the LSTM model needs to be validated with more data and in independent cohorts, it has the potential to provide a more precise test than a RAT.

  1. Aljbawi W, et al. Developing a multivariate prediction model for the detection of COVID-19 from crowd-sourced respiratory voice data. Digital medicine for COVID-19, OA1626, ERS International Congress 2022, Barcelona, Spain, 4‚Äí6 September.

Copyright ©2022 Medicom Medical Publishers



Posted on