On Monday, scientists announced that artificial intelligence (AI) has enabled the development of a smartphone app that can accurately diagnose COVID-19 infection in people’s speech.
Since PCR testing can be prohibitively expensive and/or difficult to deliver in low-income countries, the researchers claimed its app was more accurate than other antigen tests and might be used there instead.
Wafaa Aljbawi, a researcher at the Institute of Data Science, Maastricht University, The Netherlands, said, “The positive results show that basic speech recordings and fine-tuned AI algorithms can possibly reach high precision in predicting which patients have COVID-19 infection.”
“In addition, they allow for digital testing to be conducted from afar and have a turnaround time of less than a minute. For instance, they could be set up at the entrances to huge events to speed up the screening process “during the European Respiratory Society’s International Congress in Barcelona, Spain.
Voice alterations are a common symptom of COVID-19 infection, which affects the upper respiratory system and vocal cords.
It was determined that Aljbawi and her superiors would look into whether or not COVID-19 might be detected by using artificial intelligence to analyze sounds.
For this study, they analyzed information from the crowdsourced COVID-19 Sounds App developed at the University of Cambridge, which includes 893 audio samples from 4,352 healthy and unhealthy users, including 308 who tested positive for COVID-19.
Mel-spectrogram analysis was utilized by the researchers because it isolates and characterizes individual characteristics of the human voice, including its volume, intensity, and temporal change.
Allawi elaborated, “we constructed multiple artificial intelligence models and analyzed which one worked best at identifying the COVID-19 cases in order to separate the voice of COVID-19 patients from the voice of individuals who did not have the disease.”
Specifically, they discovered that a model with the catchy acronym LSTM (Long-Short Term Memory) performed better than the rest.
Like the human brain, LSTM uses neural networks to analyze data and identify hidden patterns. With an overall accuracy of 89%, it properly identified 89% of positive instances and 83% of negative ones.
According to Aljbawi, “these findings demonstrate a substantial increase in the accuracy of diagnosing COVID-19 in comparison to state-of-the-art diagnostics like the lateral flow test.”
Their findings, the researchers add, need to be confirmed with larger samples.
