The combination of X-rays and AI can serve as a quick diagnostic tool for COVID-19

X-rays, first used clinically in the late 1890s, could be a leading diagnostic tool for patients with COVID-19 using artificial intelligence, according to a team of researchers in Brazil who taught a computer program, through various machine learning methods to detect COVID-19. 19 in chest X-ray with an accuracy of 95.6 to 98.5%.

They published their results in IEEE / CAA Journal Automatica Sinica, a joint publication of the IEEE and the Chinese Automation Association.

Researchers have previously focused on the detection and classification of lung pathologies, such as fibrosis, emphysema and lung nodules, through medical imaging. Common symptoms of suspected COVID-19 infections include respiratory distress, coughing and, in more aggressive cases, pneumonia – all visible on medical imaging such as a CT scan or X-ray.

When the COVID-19 pandemic broke out, we agreed to use our expertise to address this new global problem. ”

Victor Hugo C. de Albuquerque, Stuy Cor corresponds to Author and Rresearcher, Laboratory for Image Processing, Signals and Applied Computing, Universidade de Fortaleza

Many medical facilities also have an inadequate number of tests and long processing times, Albuquerque said, so the research team focused on improving tools that are readily available in every hospital and are already commonly used in diagnosing COVID-19: X-ray devices.

“We decided to investigate whether COVID-19 infection could be automatically detected using X-rays,” Albuquerque said, noting that most X-rays are available within minutes, compared to the days required for diagnostic swab or saliva tests.

However, the researchers found a lack of publicly available chest X-rays to train their model of artificial intelligence to automatically identify the lungs of patients with COVID-19. They had only 194 X-rays of COVID-19 and 194 healthy X-rays, while usually thousands of images are needed to thoroughly learn the model for detecting and classifying a particular target.

To make up for this, they took a model trained on a large set of data from other X-rays and trained him to use the same lung detection methods that were probably infected with COVID-19. They used several different machine learning methods, two of which resulted in an accuracy score of 95.6% and 98.5%, respectively.

“Because X-rays are very fast and cheap, they can help triage patients in places where the health system has collapsed or in places that are far from major centers with access to more complex technologies,” Albuquerque said. “This approach to automatically detecting and classifying medical images can help physicians identify, measure weight, and classify disease.”

Further, Albuquerque said, the researchers plan to continue testing their method with larger data sets as soon as they become available, with the ultimate goal of developing a free online platform for classifying medical images.


Chinese Automation Association

Journal reference:

Ohata, EF, and others. (2021) Automatic detection of COVID-19 infection by chest X-rays through transfer learning. IEEE / CAA Journal Automatica Sinica.