Researchers at Northwestern University have trained an artificial intelligence algorithm to automatically detect the signs of COVID-19 on a basic X-ray of the lungs, and it’s capable of outperforming a team of specialized readers.
The developers said the AI could be used to rapidly screen patients upon admission to a hospital, especially for reasons unrelated to coronavirus symptoms, and trigger protocols to help protect healthcare workers.
“It could take hours or days to receive results from a COVID-19 test,” said Ramsey Wehbe, a cardiologist and postdoctoral fellow in AI at the Northwestern Medicine Bluhm Cardiovascular Institute. “AI doesn’t confirm whether or not someone has the virus. But if we can flag a patient with this algorithm, we could speed up triage before the test results come back.”
Called DeepCOVID-XR, the machine learning program was able to spot COVID-19 in X-rays about 10 times faster than thoracic radiologists and 1% to 6% more accurately.
“We are not aiming to replace actual testing,” said Aggelos Katsaggelos, the Joseph Cummings Professor of Electrical and Computer Engineering at Northwestern and senior author of the team’s study published in the journal Radiology. “X-rays are routine, safe and inexpensive. It would take seconds for our system to screen a patient and determine if that patient needs to be isolated.”
RELATED: Caption Health nets Gates Foundation grant to bring AI guidance to lung ultrasound
Trained and tested on a data set of more than 17,000 X-ray images, the algorithm identified patterns in patients with COVID-19: Instead of a clear scan, their lungs appeared patchy and hazy as air sacs became inflamed and filled with fluid instead of air.
DeepCOVID-XR (Northwestern University)
These are similar to cases of pneumonia, heart failure or other pulmonary conditions, but the AI was able to tell the difference and spot the contagious disease. Still, there’s a limit to radiologic diagnosis, as not all carriers of COVID-19 may show signs of illness, especially during the early stages of an infection.
“In those cases, the AI system will not flag the patient as positive,” said Wehbe. “But neither would a radiologist.”
RELATED: GE Healthcare's new chest X-ray AI double-checks ventilator tube placement in COVID-19 patients
When put up against five experienced, fellowship-trained radiologists, DeepCOVID-XR was able to process a set of 300 test X-rays in about 18 minutes, compared to about two and a half to three and a half hours. The AI also delivered an accuracy rate of 82%, about on par with the group’s range of 76% to 81%.
Additionally, the researchers have made the algorithm publicly available, allowing others to train it with new data, with the goal of eventually getting the program into the clinic.
“Radiologists are expensive and not always available,” Katsaggelos said. “X-rays are inexpensive and already a common element of routine care. This could potentially save money and time—especially because timing is so critical when working with COVID-19.”