TECH
Researchers from the Netherlands have presented an artificial intelligence algorithm that uses audio recordings of a person's voice to determine if they are infected with SARS-CoV-2. This approach is expected to significantly increase the accuracy of rapid tests, simplify and reduce the cost of the process.
Existing rapid tests for COVID-19 show different accuracy depending on the brand - on average, the sensitivity of such tests hovers around 50%. Scientists at Maastricht University have developed a simple and inexpensive alternative to this approach - AI that determines the infection by voice. Its accuracy reaches 89%.
Scientists have shown that it identifies SARS-CoV-2 in 89% and detects negative cases in 89%.
To train the AI, they used audio recordings of the voices of 4,352 people, 308 of whom were infected with SARS-CoV-2. Participants recorded and sent specific breath sounds to a smartphone app, including deep mouth breathing, coughing, and reading a short sentence. As a result, it was possible to build AI models that distinguished many characteristics of the voice and its change.
More than 36,000 people have already shared new voice recordings with scientists, so now the team plans to make the tests even more accurate. In the short term, if AI proves its potential in a large sample of participants, the scientists plan to introduce it into routine clinical practice. The emergence of such a simple technology should greatly simplify the COVID-19 diagnosis process, reduce the burden on medical staff, and reduce healthcare system costs.
Previously, other scientists have come up with an AI that diagnoses Parkinson's disease - a common and incurable neurodegenerative disease - by breathing.
mundophone
No comments:
Post a Comment