AI detects COVID-19 in X-ray images more reliably than humans

A team of researchers from Inselspital, University Hospital Bern and University of Bern has developed a new method of chest X-ray image analysis using artificial intelligence (AI). The AI-supported analysis was compared with a conventional evaluation by X-ray doctors. The AI-supported analysis provided significantly more reliable results, especially in the detection of COVID-19 lung diseases.

The University of Bern and Inselspital, University Hospital of Bern are investing heavily in securing and expanding their leading position in the field of artificial intelligence (AI) in medicine. The establishment of the Center for Artificial Intelligence in Medicine (CAIM) was announced in mid-November. A study on COVID-19 detection is now being published, which impressively underlines the importance of AI in medicine. In this way, AI-supported technologies could help avoid overloading health systems during a global pandemic in the future.

What exactly was examined?

First, an AI algorithm was “trained”. The training data included 7988 cases, 258 of them with COVID-19 and 5451 with other pneumonia. For the comparison between radiology professionals and the AI ​​analysis, 100 cases each were selected. Eleven radiologists with different expertise were involved. Correct determinations in three categories were compared: “normal lung”, “other pneumonia” and “COVID-19 disease”.

Clear results: AI significantly better than experts

On average for all three categories, the AI ​​correctly analyzed 94% of the images. The radiologists, on the other hand, came up with 61% correct determinations. The biggest difference was found in the determination of the COVID-19 diseases: the AI ​​was able to correctly assign 97% of the cases, the radiologists, however, 53%. Prof. Dr. Andreas Christe, Head of Radiology, comments: “Experience has shown that radiologists are good at recognizing abnormalities (“ normal lung ”versus“ diseased lung ”). They were almost as good at this as the AI. But the computer was far superior in dividing pneumonia into Covid and non-Covid cases. This suggests that the computer can see something in the images that cannot be seen by the human eye. More attention will be paid to this aspect in future research projects. In the interaction of an AI-supported image analysis and medical expertise, we get the most out of the new technologies. “

Interdisciplinary cooperation ensures rapid implementation in clinical practice

Various national and international research groups are trying to advance the diagnosis and prognosis of the COVID-19 disease in interdisciplinary cooperation. This is where the research team from the ARTORG Center for Biomedical Engineering Research at the University of Bern and the Inselspital can exploit the strengths of Bern as a location. Prof. Dr. Stavroula Mougiakakou, head of the research group “AI in Health and Nutrition” at the ARTORG Center explains: “Our AI experts have been working with doctors in radiology in a joint research group for seven years now. During this time we have developed a common language and a transdisciplinary understanding. “

To further develop and validate the Bern system, the team will carry out a multi-center study in close collaboration with the University Hospitals in Zurich and Lausanne and other European universities. For example, the combination of X-ray imaging and AI can help to divide patients into different risk groups for more severe or milder courses of COVID-19 when they are first diagnosed. 

Related Articles

Stay Connected


Latest Articles

Australian Government Selects ICEYE for Near Real-Time Flood and Bushfire Data...

ICEYE , a world-leading provider of natural catastrophe solutions and insights, has been selected by the Federal Government of Australia to supply flood and bushfire...