AI system is ideal for radiologists to discover COVID-19 in the lungs

A new artificial intelligence (AI) platform developed by researchers at Northwestern University is 10 times faster than a professional cardiothoracic radiologist and can detect COVID-19 in the lungs with a little more accuracy. Survey Released today Radiology..

Researchers used DeepCOVID-XR, a machine learning algorithm that analyzes chest x-rays, on 17,002 x-ray images collected from February to April (of which 5,445 show signs of COVID-19). Trained and tested.

DeepCOVID-XR analyzed each of 300 randomly selected test images in about 18 minutes when played against five experienced cardiothoracic radiologists, with individual radiologists 2.5- It was 3.5 hours. DeepCOVID-XR was 82% accurate, while radiologists were 76% to 81% individually and 81% as a team.

“These are specialists trained in the field of reading chest images, but most of the chest x-rays are read by a general radiologist or first by a non-radiologist, such as a treating clinician. Is interpreted by the lead author, Ramsey Wehbe, MD, in the northwest. news release.. “Most of the chest x-rays are read by a general radiologist or first interpreted by a non-radiologist, such as a treating clinician. Often decisions are made based on that first interpretation. I will. “

In connection with the standard reverse transcription-polymerase chain reaction COVID-19 test, the AI ​​platform was 82% accurate in classifying test X-ray images. DeepCOVID-XR also had 71% sensitivity compared to 60% for one radiologist and 92% specificity compared to 75% for two radiologists.

Sensitivity and specificity are performance indicators of the test, sensitivity refers to the percentage of positive test results that are correctly identified, and specificity refers to the percentage of negative results that are correctly identified.

Shortening the time to diagnosis and quarantine

Although the AI ​​platform is still in research and not clinically available, the authors can quickly screen patients hospitalized for conditions other than coronavirus and promptly test and quarantine COVID-19 if necessary. Said that it could be used someday to make it.

In the release, research co-author Dr. Aggelos Katsaggelos said the team was not trying to replace the COVID-19 test with an AI platform (diagnosis could not be confirmed), but rather hours or hours to diagnosis and isolation. He said he was trying to shorten the day. As a result, patients do not spread the virus to healthcare professionals or other patients. “It will take a few seconds for our system to screen patients and determine if they need to be quarantined,” he said.

However, many people with the coronavirus are asymptomatic, and in most cases they do not show signs of it on x-rays until the second half of the disease, so the system misses cases. “In such cases, the AI ​​system does not flag the patient positive,” he said. “But radiologists don’t.”

According to Webe, DeepCOVID-XR helps distinguish COVID-19 from pneumonia, heart failure, and other illnesses that show similar hazy patches on x-rays. “Many patients with COVID-19 have characteristic findings in chest images,” he said. “These include” bilateral integration. ” The lungs are filled with fluid and become inflamed, especially along the lower lobes and surroundings. “

Also, according to Katsaggelos, in contrast to radiologists, x-rays are routine, safe, inexpensive, and always available. “This can potentially save money and time-especially because timing is very important when working with COVID-19,” he said.

In conclusion, the authors conducted a prospective assessment that included patients who had not been investigated for COVID-19 and added other clinical data to the platform to enhance its performance and risk of confirmed clinical outcomes for patients. He said he plans to adapt to the forecast. COVID-19 (New Coronavirus Infection) (# If there is no character limit, add parentheses when it first appears.

Northwestern has released DeepCOVID-XR for other research teams to use the data to train. “By providing the DeepCOVID-XR algorithm codebase as an open source resource, we hope that researchers around the world will use clinical images from their institutions to further improve, fine-tune, and test the algorithm. The author writes.