How artificial intelligence is fighting the coronavirus

The development of artificial intelligence (AI) is experiencing a boiling moment. According to the report Digital Society in Spain in 2019, by the Telefónica Foundation, one in 12 European startups that year focused their activity around this technology, a proportion that in 2013 was only one in 50. The majority work around to the category that the report defines as “necessary”: they develop systems around facial recognition, e-commerce search or image and medical diagnostics. This last typology, that of health technologies, brings together a fifth of emerging companies.

The trend has only increased in the context of covid-19. Experts predict a boom in the search for health and logistics solutions based on artificial intelligence. How is this technology helping to combat the pandemic?

Prediction of ICU beds for the Basque Health System

It is precisely in the field of health that some of the applications with the greatest practical impact are taking place., a Basque artificial intelligence company –considered among the 10 most important in the world in this field by various specialized publications–, has developed a platform so that health authorities can estimate the number of ICU beds needed throughout a week. An essential calculation for hospital logistics and to avoid the overflow scenes that were experienced in some territories.

“We believed that artificial intelligence could be important to combat the pandemic and we began to work hand in hand with the Basque Health Service,” explains Xabi Uribe-Etxebarria, founder and CEO of the company. The team began to shape a predictive model by feeding it with data from Italy, the only ones existing at the time, originating from regions similar in demography to the Basque Country. “First we calculate the data by province and by hospital,” continues Uribe-Etxebarria. “Later we created a platform that could be consulted virtually and we added functionalities. For example, how would the number of deaths and the number of infected evolve or where would the new outbreaks of the pandemic be ”.

The platform calculates the number of ICU beds needed within seven days and analyzes patterns and trends in the evolution of the virus, a fundamental aspect so that health workers can react in time. It also predicts the number of mild hospitalizations, which would allow, for example, to set up a specific hospital for this class of patients.

“The ICUs have not been saturated. The work of the health authorities and the health service has been tremendous. We have all gone out of our way to do this in weeks, ”Uribe-Etxebarría rejoices. develops conversational systems, recommendation, content analysis and detection of fake news and application of artificial intelligence in data privacy. “We are at the beginning of a new era in which we are going to coexist with AI. Many times we associate it with futuristic robots, but it is applied in everyday life and has the ability to solve urgent challenges for humanity ”, he argues. “It is present when we unlock the mobile with the fingerprint or facial recognition, when we pay for parking and the barrier is opened, or in health applications that detect cancers. It is a complement to the human, not a substitute ”.

Detect the virus with an x-ray

Francisco Herrera, professor of Computer Science and Artificial Intelligence at the University of Granada, set a simple goal for his work: that anyone, by means of a simple X-ray, could know if they had covid-19 or not.

To do this, together with the San Cecilio de Granada Clinical Hospital, his team works with a database of hundreds of X-rays, both of sick and healthy patients. A sea of ​​information from which the AI ​​will learn which X-rays correspond to sick patients, and to what degree, to later be able to analyze any new image.

“The hospital began to create a database of images of patients with a positive PCR test performed, thereby guaranteeing the quality of the sample. On the other hand, we had a set of healthy lung contrast, with a guarantee that there was no covid-19, ”Herrera added.

Currently, the model determines by 81% if a patient has coronavirus. The main advantage of the system is that it works with chest X-rays, a rapid test available throughout the health system, something that would improve diagnostic times. With a CT test, accuracies of more than 90% are obtained, the scientist warns, but it is much more expensive and is only available in large hospitals.

“We want anyone who lives in a town to come to the health center, have an X-ray and that image is analyzed by our system, which will respond with the probability of associated disease,” Herrera details. “Then the complementary tests will be requested and the relevant protocols will be activated.”

The more X-rays, explains the professor, the better the algorithm’s learning and the better the model’s behavior. Herrera underlines the importance of the quality of the images that feed the base. “If you take X-rays from different hospitals, the annotations or colors may vary,” he says. “You have to segment the image and keep the part that collects the lungs to avoid any type of noise.”

The project, at the moment, has seven research teams and six associated hospitals. Herrera hopes that other hospitals will join soon.

Leave a Reply

Your email address will not be published. Required fields are marked *