Machine Learning Tool Accurately Detects COVID-19 on X-Rays

A machine learning tool was able to detect COVID-19 in x-ray images about ten times faster and one to six percent more accurately than specialized thoracic radiologists, according to a study published in Radiology.

For more coronavirus updates, visit our resource page, updated twice daily by Xtelligent Healthcare Media.

In patients with COVID-19, chest x-rays look similar, with lungs appearing patchy and hazy rather than clear and healthy. However, pneumonia, heart failure, and other chronic illnesses in the lungs can look similar to COVID-19 on x-rays. Trained radiologists have to be able to tell the difference between COVID-19 and a less contagious disease.

“Many patients with COVID-19 have characteristic findings on their chest images,”  said Ramsey Wehbe, a cardiologist and postdoctoral fellow in AI at the Northwestern Medicine Bluhm Cardiovascular Institute.

“These include ‘bilateral consolidations.’ The lungs are filled with fluid and inflamed, particularly along the lower lobes and periphery.”

To improve the identification of COVID-19 in x-rays, researchers used 17,002 chest x-rays to develop, train, and test a machine learning algorithm. Of those images, 5,445 came from COVID-19-positive patients from sites across Northwestern Memorial Healthcare System.

The team then tested the machine learning algorithm against five experienced cardiothoracic fellowship-trained radiologists on 300 random test images. Each radiologist took approximately two-and-a-half to three-and-a-half to examine this set of images, while the machine learning model took about 18 minutes.

Additionally, the results showed that the radiologists’ accuracy ranged from 76 to 81 percent, while the algorithm performed slightly better with an accuracy of 82 percent.

The findings indicate the ability for the model to offer faster, more accurate diagnoses than are typically delivered in standard care.

“These are experts who are sub-specialty trained in reading chest imaging," said Wehbe. “Whereas the majority of chest X-rays are read by general radiologists or initially interpreted by non-radiologists, such as the treating clinician. A lot of times decisions are made based off that initial interpretation.” 

The tool could be a better alternative in hospitals and health systems during the resource-intensive pandemic.

“Radiologists are expensive and not always available,” said Aggelos Katsaggelos, an AI expert and senior author of the study.

“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.”

However, the platform isn’t meant to replace standard methods of diagnosis completely. Because not all COVID-19 patients show signs of illness – even on chest x-rays – the system will not diagnose those patients with the virus.

“In those cases, the AI system will not flag the patient as positive,” Wehbe said. “But neither would a radiologist. Clearly there is a limit to radiologic diagnosis of COVID-19, which is why we wouldn’t use this to replace testing.” 

The team expects that the machine learning tool could help physicians rapidly screen patients who are admitted into hospitals for reasons other than COVID-19. The model could facilitate faster, earlier detection of the virus, which could protect providers and patients by recognizing which patients need to isolate sooner.

The group also thinks that the algorithm could flag patients for isolation and testing who aren’t necessarily presenting with symptoms of COVID-19.

“We are not aiming to replace actual testing,” said Katsaggelos. “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.”

With the onset of COVID-19, the research team decided to shift their efforts to developing solutions that could help alleviate burdens on the healthcare system.

“When the pandemic started to ramp up in Chicago, we asked each other if there was anything we could do,” Wehbe said. “We were working on medical imaging projects using cardiac echo and nuclear imaging. We felt like we could pivot and apply our joint expertise to help in the fight against COVID-19.”

The researchers have made the algorithm publicly available so that others can keep training it with new data. The model is still in the research phase, but could potentially be used in clinical settings in the future.

“It could take hours or days to receive results from a COVID-19 test,” said Wehbe. “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.”

Machine learning and AI have played a large role in improving medical imaging analytics during the pandemic. In August, NIH launched the Medical Imaging and Data Resource Center (MIDRC), an project to leverage AI and medical imaging to enhance COVID-19 detection and treatment.

The effort is led by the National Institute of Biomedical Imaging and Bioengineering (NIBIB).

“This major initiative responds to the international imaging community’s expressed unmet need for a secure technological network to enable the development and ethical application of artificial intelligence to make the best medical decisions for COVID-19 patients,” said Krishna Kandarpa, MD, PhD, director of research sciences and strategic directions at NIBIB.

“Eventually, the approaches developed could benefit other conditions as well.”