London: AI classifiers trained on audio recordings cannot accurately predict whether someone has COVID-19 analyzing the sound of their coughs, according to the study led by the UK’s Alan Turing Institute.
As first reported in a paper by researchers led by the Massachusetts Institute of Technology, there were claims that AI could detect the difference in cough sounds between those with and without covid-19 with an accuracy of up to 98.5 percent, reports The Register.
The result led to efforts to create an algorithm-driven app to give people a cheap and easy way to detect the new coronavirus. He Department of Health and Social Assistance in the UK he even awarded Fujitsu two contracts totaling more than £100,000 to develop the so-called “cough in a boxinitiative in 2021, according to the report.
The software would collect audio recordings of users’ coughing for analysis on its covid-19 app.
Researchers from the Alan Turing Institute and the Royal Statistical Society, commissioned by the UK Health Security Agency, conducted an independent review of audio-based artificial intelligence technology as a screening tool for Covid-19.
They found that even the most accurate cough detection model outperformed a model based on user-reported systems and demographic data such as age and gender.
The researchers examined data from more than 67,000 people recruited through the National Health Service’s Test and Trace and REACT-1 programs, which asked participants to submit the results of their Covid-19 nose and throat swab tests. , as well as recordings of them coughing, breathing and talking. .
In an attempt to see if coughing could be used as a biomarker, the researchers trained an AI model based on audio recordings and test results, according to the report.
“But as we continued to analyze the results, it seemed that the accuracy was likely due to an effect in statistics called confounding, where models learn other variables that correlate with the actual signal, rather than the actual signal itself.” , Kieran Baker. , a doctoral student in statistics at King’s College London and a research assistant at the Alan Turing Institute, said.
The confusion was caused by the recruitment bias of the Test and Trace system, which required participants to have at least one symptom in order to participate. The researchers conducted additional tests, matching participants of similar ages and genders, and only one had covid-19, the report mentions.
“When we tested these models on the pooled data, the models did not perform well, so we concluded that our models cannot detect a bioacoustic marker of covid-19 from this data,” Baker said.
