Machine learning tools accurately detect COVID-19 with X-rays

November 25, 2020 -Machine learning tools were able to detect COVID-19 on X-ray images about 10 times faster than a professional thoracic radiologist, 1-6 percent more accurately. Survey Was announced in Radiology..

For the latest information on coronavirus, Resource page, Updated twice daily by Xtelligent Healthcare Media.

In patients with COVID-19, chest x-rays are similar and the lungs are clearly unhealthy and appear spotted and vague. However, pneumonia, heart failure, and other chronic lung disorders may appear to be similar to COVID-19 on x-ray. A trained radiologist must be able to understand the difference between COVID-19 and a less contagious disease.

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

“These include” bilateral integration. ” The lungs are filled with fluid and become inflamed, especially along the lower lobes and periphery. “

read more: Artificial intelligence identifies asymptomatic COVID-19 infections

To improve the identification of COVID-19 on x-rays, researchers used 17,002 chest x-rays, Machine learning algorithm.. Of these images, 5,445 were from COVID-19-positive patients from sites throughout the Northwestern Memorial Health System.

The team then tested the machine learning algorithm on five experienced cardiothoracic fellowship-trained radiologists with 300 random test images. Each radiologist took about 2.5 to 3.5 minutes to examine this series of images, while the machine learning model took about 18 minutes.

In addition, the results showed that the radiologist’s accuracy ranged from 76 to 81%, while the algorithm was slightly better with 82% accuracy.

The findings show the ability of the model to provide a faster and more accurate diagnosis than is normally provided with standard treatment.

“These are professionally trained professionals in reading breast images,” Wehbe said. “Most of the chest x-rays are read by a general radiologist or initially interpreted by a non-radiologist, such as the treating clinician. Often decisions are made based on that initial interpretation. Will be given. ”

read more: Predictive analysis determines outcomes for patients with COVID-19 ED

This tool has the potential to be a better alternative in hospitals and healthcare systems during resource-intensive pandemics.

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

“X-rays are cheap and are already a common element of routine care. This can save you money and time, especially when using COVID-19, when timing is very important. Because there is. “

However, this platform is not intended to completely replace standard diagnostic methods. Not all COVID-19 patients show signs of illness even on chest x-rays, so the system does not diagnose them with the virus.

“In such cases, the AI ​​system does not flag the patient positive,” Wehbe said. “But the radiologist didn’t do that either. Obviously, the radiodiagnosis of COVID-19 is limited, so we don’t use it as a substitute for testing.”

read more: One-third of tissue uses artificial intelligence in medical imaging

Teams have machine learning tools Can help Physicians will quickly screen patients who are hospitalized for reasons other than COVID-19. This model can protect providers and patients by facilitating faster and earlier detection of the virus and recognizing which patients need to be quarantined earlier.

The group also believes that the algorithm can flag patients who do not necessarily present with COVID-19 symptoms for isolation and examination.

“We are not aiming to replace the actual test,” Katsaggelos said. “X-rays are routine, safe and inexpensive. It takes a few seconds for our system to screen patients and determine if they need to be quarantined.”

With the onset of COVID-19, the research team decided to focus on developing solutions that would help ease the burden on the healthcare system.

“When the pandemic began to occur in Chicago, we asked each other if there was anything we could do,” Webe said. “We were working on a medical imaging project using echocardiography and nuclear imaging. We felt that we could pivot and apply our joint expertise to help fight COVID-19.”

Researchers have made the algorithm publicly available so that others can continue to train it with new data. This model is still in the research stage, but may be used in clinical practice in the future.

“It can take hours or even days to receive the results of the COVID-19 test,” Wehbe said. “AI doesn’t check if someone is infected with the virus, but if this algorithm can flag the patient, it can speed up triage before the test results come back.”

Machine learning and AI have played a major role in improving medical image analysis during pandemics. August, NIH Release The Medical Imaging and Data Resource Center (MIDRC) is a project that leverages AI and medical imaging to enhance the detection and treatment of COVID-19.

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

“This key initiative enables the development and ethical application of artificial intelligence to meet the stated requirements of the international imaging community for a secure technology network to make the best medical decisions for COVID-19 patients. Meet unmet needs, “says Krishna Kandarpa, MD, Researcher in Science and Strategic Direction at NIBIB.

“Ultimately, the developed approach may also benefit other conditions.”