A team of researchers at Northwestern University, Illinois, USA, have innovated a platform that uses artificial intelligence (A.I.) to detect COVID-19 at higher levels of accuracy than specialized thoracic radiologists.
Image Credit: Northwestern University
The breakthrough will likely be instrumental in preventing the spread of the virus, particularly amongst hospital patients and staff, by providing a rapid and accurate method for screening patients admitted to hospital either with or without COVID-19 symptoms by merely scanning their chest x-rays.
The technology will act as an early warning system, highlighting those who need to self-isolate even before symptoms develop and, additionally, picking up on those who may have never been aware of their need to isolate.
An early warning system
The study, published this month in the journal Radiology, proposes that the A.I. could be used as a safe and inexpensive screening method that would not replace actual testing but act as an extra route to highlighting those who need to isolate who otherwise may be missed, or would recognize their symptoms much later, after exposing many more people to the virus.
The platform can detect signs of the virus in seconds, enabling people to begin isolating before getting results from testing, which generally takes hours or even days.
Dr. Ramsey Wehbe, a cardiologist and postdoctoral fellow in A.I. highlights the role of the system not as a diagnostic tool but rather as a method of flagging those who show signs of COVID-19 and may need to isolate and receive further treatment.
A.I. 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."
Differentiating COVID-19 from pneumonia and other illnesses
COVID-19 infections can often be recognized by characteristic patterns displayed in their chest X-rays. Frequently, the images appear patchy or hazy, include 'bilateral consolidations,' and show signs of inflammation and extra fluid.
However, it is difficult even for specialized medical professionals to differentiate chest X-rays of COVID-19 patients from those of patients with pneumonia, heart failure, and other illnesses, as the lungs can look very similar.
To overcome this issue, the team of scientists developed and trained their A.I. algorithm on over 17,000 chest X-rays sourced from the most extensive published clinical dataset available from the time of the pandemic. 5,445 of the images in the data set were taken of patients who'd tested positive for COVID-19.
The accuracy of the A.I. system, known as DeepCOVID-XR, was compared with experienced cardiothoracic fellowship-trained radiologists on their analysis of 300 test images. The results revealed that the A.I. was not only significantly faster at analyzing the images; it was also more accurate.
On average, radiologists spend two-and-a-half to three-and-a-half hours to analyze the set of images, whereas the A.I. took around 18 minutes. Additionally, the accuracy of the radiologists' reading was between 76 and 81%, whereas the A.I. achieved 82%.
It should be noted that the A.I. system, while fast and accurate at reading the characteristics of COVID-19 from X-rays, would not be able to highlight all patients with the virus via these images as not every COVID-19 patient manifests symptoms on their lungs, especially early on.
The newly developed algorithm has been made publicly available with the intention of allowing other research teams to train the system with new data and develop its capabilities. The A.I. will likely be an invaluable tool to be used alongside other methods in preventing the spread of COVID-19.
Journal reference:
- Wehbe, R.M., Sheng, J., Dutta, S., Chai, S., Dravid, A., Barutcu, S., Wu, Y., Cantrell, D.R., Xiao, N., Allen, B.D., MacNealy, G.A., Savas, H., Agrawal, R., Parekh, N. and Katsaggelos, A.K. (2020). DeepCOVID-XR: An Artificial Intelligence Algorithm to Detect COVID-19 on Chest Radiographs Trained and Tested on a Large US Clinical Dataset. Radiology, p.203511. doi: 10.1148/radiol.2020203511