X-rays combined with AI could prove effective in detecting Covid-19

The researchers previously focused on detecting and classifying lung pathologies, such as fibrosis, emphysema, and lung nodules, through medical imaging. Common symptoms presented by suspected Covid-19 infections include respiratory distress, cough, and, in more aggressive cases, pneumonia – all visible via medical imaging such as CT scans or X-rays.

“When the Covid-19 pandemic arose, we agreed to put our expertise to use to help deal with this new global problem,” said corresponding author Victor Hugo C de Albuquerque, a researcher in the Laboratory of Image Processing, Signals, and Applied Computing and with the Universidade de Fortaleza.

According to Albuquerque, many medical facilities have both an inadequate number of tests and lengthy processing times, so the research team focused on improving a tool that is readily available at every hospital and already frequently used in diagnosing Covid-19: X-ray devices.

“We decided to investigate if a Covid-19 infection could be automatically detected using X-ray images,” Albuquerque explained, noting that most X-ray images are available within minutes, compared to the days required for swab or saliva diagnostic tests.

The researchers found, however, a lack of publicly available chest X-rays to train their artificial intelligence model to automatically identify the lungs of Covid-19 patients – they had just 194 Covid-19 X-rays and 194 healthy X-rays, while it usually takes thousands of images to teach a model to detect and classify a particular target.

To tackle this, the researchers took a model trained on a large dataset of other X-ray images and trained it to use the same methods to detect lungs likely infected with Covid-19. They used several machine-learning methods, two of which resulted in a 95.6 per cent and a 98.5 per cent accuracy rating, respectively, the researchers said.

“Since X-rays are fast and cheap, they can help to triage patients in places where the health care system has collapsed or in places that are far from major centres with access to more complex technologies,” Albuquerque said. “This approach to detect and classify medical images automatically can assist doctors in identifying, measuring the severity and classifying the disease.”

Albuquerque concluded the team’s next plan is to continue testing their method with larger datasets as they become available, with the ultimate goal of developing a free online platform for medical image classification.

In January this year, the technology branch of the NHS, NHSX, acquired an extensive database of images from Covid-19 patients across the UK, which will be used to train an algorithm to inform a more accurate diagnosis of patients arriving in hospital.

The team at Brazil are not the only researchers aiming to combine AI with X-ray images to help detect Covid-19. Last November, researchers at Northwestern University developed AI that can detect signs of Covid-19 by looking at X-ray images of a patients’ lungs.