This team has developed a method that uses Artificial Intelligence techniques to help in the early detection of pathologies associated with COVID-19 in patient x-rays. This system, which is available to the scientific and medical community in open access, has been developed from a previous study in a hospital in Brighton (United Kingdom), adapted to the new coronavirus with the collaboration of the Fundación Instituto San José hospital in Madrid.

Specifically, the research team led by Verónica Sanz and Felipe Freitas (University of Aveiro, Portugal) began to develop a tool to facilitate the triage of patients with respiratory diseases in developing countries a couple of years ago. They used Artificial Intelligence techniques such as Machine Learning (which aims to make computers recognize certain patterns) applied to a large database of x-rays from the United States. This work, done in collaboration with Dr Andrew Elkins of Brighton Hospital, was published in the ‘Journal of Medical Artificial Intelligence’.

The appearance of COVID-19 led the researchers, together with Johannes Hirn, to consider applying this method to respiratory diseases caused by the new coronavirus. Using computer vision techniques, they apply specific Artificial Intelligence techniques to improve the speed and robustness of identification, something already studied in the work mentioned above. This is how the system is 'trained' to distinguish in the 2D image the characteristics of patients with IDOCs from other pathologies such as conventional pneumonia. "In the extensive database we now have, we achieved excellent identification results that have motivated us to continue our research and extend it to other clinical markers", assures Verónica Sanz.

Applications

Thus, in order to carry out a more detailed study of the disease, the researchers are working with the Fundación Instituto San José, a hospital specialized in Neurological and Traumatological Rehabilitation and a reference in Palliative Care located in Madrid. The idea is to include other clinical data to further refine the results. The prototype tool is public, open to the scientific and medical community.

"This study could help to make an earlier detection of the disease, or understand how the characteristics of the patient can determine what type of action is best," explains Sanz. "The prototype application we have developed can help in the triage of patients, but it does not replace a PCR diagnosis and the decision of a health professional", stresses the IFIC researcher.

On the other hand, in collaboration with Professor Mirko Salomón Alva Sánchez of the Federal University of Health Sciences of Porto Alegre (Brazil), the study will be extended to a hospital in the Brazilian city. The researchers have made an official request to the Brazilian government to carry out a clinical study of COVID-19 using this triage tool, which would improve its application and use in other health centres.

In addition, the line of work with which this application was initially developed remains open: the early detection of respiratory diseases in developing countries. Thus, in collaboration with several NGOs, researchers intend to fine-tune the system in other situations where this tool might fail to address a disease that is rare in developed countries, such as tuberculosis, and therefore not present in the training databases. With no incidence in developed countries, tuberculosis (lung infection caused by the bacterium Mycobacterium tuberculosis) remains one of the top 10 causes of death in the world, according to the WHO. In 2018, there were 10 million infections and 1.5 million deaths from this cause, including 251,000 children. Ending the TB epidemic by 2030 is one of the health-related targets of the Sustainable Development Goals.