A new algorithm developed by MIT in Massachusetts, USA, can correctly identify people infected with COVID-19 based on the coughing sound. The algorithm has a success rate of 98.5% in identifying people with symptoms and 100% in differentiating from healthy people.
The algorithm can identify sounds that the human ear cannot hear
In a way, Brian Subirana, one of the scientists who developed the algorithm, explained in an article from theJournal of Engineering in Medicine and Biology»That the cough of people infected with COVID differs from that of healthy people and that these differences, which are inaudible to humans – but can be recognized using the algorithm they have developed – also exist in asymptomatic patients.
The algorithm is still in the testing state as it still requires government approval before it can be turned into a real medical or scientific application. Once approved, this technology can be used for day-to-day detection in study and work centers, as well as for use in public transport, and for rapid detection of sources of infection and new outbreaks.
For the development of the algorithm, MIT collected more than 70,000 cough audio samples. Among them were 2,500 of people with a confirmed case of coronavirus. This machine learning system is the same one used in artificial intelligence that is able to detect cancer based on what they learn from x-rays and mammograms.
Other companies and organizations such as pharmaceutical company Novoic, Carnegie Mellon University, and Cambridge University are working on similar algorithms. Such is the case at the University of Cambridge, which reported in July 2020 that their algorithm had one 80% success rate in identifying positive cases of Covid-19, in this case based on a combination of coughing and breathing sounds. The University of Cambridge has a data set of 30,000 records for this algorithm.