The application developed for smartphones evaluates the sound of the cough of the infected person and records the probability of being Covid-19. Photo: Bigstock
A smartphone app that uses artificial intelligence algorithms can help with future Covid-19 diagnoses. With an accuracy of between 80% and 85% – and with clinical tests performed in Brazil – the app assesses the sound of the infected person’s cough and looks for a characteristic pattern of Sars-CoV-2 infection.
Follow Semper Familia on Instagram!
People with Covid-19 tend to cough the same way, according to Amil Khanzada, founder of the international non-profit organization Virufy, who is the developer of the app. Khanzada noticed this pattern in the sound of coughing with researchers at Stanford University in the United States, and together they came to the application.
Clinical trials that assess the accuracy of the application are being conducted in Brazil, at the Hans Dieter Schmidt Regional Hospital, Joinville, Santa Catarina. The organization is, however, in negotiations with other hospitals in other regions of the country to extend the studies, according to information provided by Agência Brasil. Clinical studies are expected to be conducted within the next two to three months.
“As the company is not for profit, the idea is to make this application available for free, to facilitate detection (of the disease) by the sound of coughing alone”, according to Soraya Cavalcanti, manager of Virufy, in an interview with Agence Brazil.
App “learn” from the sound of cough
In order to be able to accurately detect which cough is linked to Covid-19, thousands of tests with coughs of people from Latin America, Europe and Asia have been carried out. From the “machine learning” algorithm [aprendizado de máquina, método de análise de dados que automatiza a construção de modelos analíticos], the app learns to differentiate the characteristics of coughs caused by Sars-CoV-2 from coughs with other causes.
The developers point out that the app is 80-85% accurate. “The more cough you give, the greater the probability of success of this algorithm. The tendency is that with the expansion of these clinical tests, this number will increase and, then, its confidence will become even greater,” Soraya explains to Brasil Agency.
According to Cavalcanti, the entity considers the application as a diagnostic aid tool, in order to understand the probability of contagion. If the result indicates a high probability, the person can proceed to isolation and see a doctor. If it is weak, the symptoms are monitored and retested.
How to participate ?
The project is divided into two parts, in Brazil. The first is the collection of cough sounds from people with symptoms similar to Covid-19, via this site.
These sounds will help train the algorithm to identify which ones are due to Covid-19 and which ones are due to other causes. The second part of the project will be to apply this knowledge in clinical research, with real patients, with or without Covid-19.
“When a patient comes to the Hans Dieter Schmidt Regional Hospital in Joinville to do the [exame de] PCR, will be invited to participate in the research and cough into a cell phone. We will collect this data from PCR and cough examinations and cross-reference the data, ”explains Diego Carvalho, the coordinator responsible for clinical tests, at Agência Brasil.
The idea is that two thousand patients make up the survey, to achieve an accuracy rate greater than 85%.