“CT scanners can accurately detect the damage done to the lungs by the coronavirus,” explains Erik Ranschaert, a radiologist from Flanders and chairman of the European radiologists’ association, EuSoMii. “The technology even allows us to spot abnormalities before the patient shows any symptoms. This got us wondering whether we could analyze CT scans using digital tools in order to reach a diagnosis faster. To pull that off, we needed to develop an algorithm, a piece of AI software that helps us mimic human abilities such as pattern recognition.”
You are here
Robovision (Flanders) deploys AI for COVID-19 testing
Consequently, Erik Ranschaert joined forces with Jonathan Berte, the founder of Robovision, a tech company from Ghent (Flanders) and one of Europe’s leading forces in artificial intelligence. Together, they set up an international consortium of 30 hospitals across Europe in collaboration with Laurens Topff – a researcher who graduated from the university of Leuven (KU Leuven) in Flanders and who currently works for the Dutch Cancer Institute.
“People are really good at recognizing patterns, but less so at quantitatively calculating lung damage – and that is exactly what our AI model will excel at,” says Jonathan Berte. “All participating hospitals will share anonymized CT scans of COVID-19 patients with a data center in Amsterdam. These images will be sent to radiologists, who will manually color in areas of lung damage on the scans. All this data will then be collected by a supercomputer in Germany, where AI will learn how to recognize lung damage on the scans.”
“Once it has been trained,” Jonathan Berte adds, “the AI algorithm will be able to act as a kind of superbrain and ‘independently’ analyze CT scans of new patients to help determine whether they are infected with coronavirus, how badly their lungs are damaged and whether they should be taken into intensive care.”
Robovision aims to have its AI model ready in just three weeks. Jonathan Berte: “Afterwards, it will be tested within a clinical context. But we hope all of this can be done even faster. Still, three weeks is like working at the speed of light, even in the world of artificial intelligence. Luckily, we can count on technical support from technology giants such as Microsoft, Google and Nvidia.”
Participants in the initiative, however, emphasize that the AI project is first and foremost a scientific research collaboration. “In this way, we distinguish ourselves from more commercial initiatives that have been launched around the globe. However, competition is certainly not a bad thing in the world of AI innovation,” says Erik Ranschaert. On this note, Icometrix, a tech company from Leuven (Flanders), is also working pro bono on a similar AI tool for radiologists in collaboration with around ten hospitals in Brussels, Leuven, Antwerp and other cities nearby.
The large-scale collaboration with Robovision has also attracted attention from the US, as the two main American radiology associations – RSNA and ACR – have expressed their desire to participate in the European initiative. Also interested in joining the AI innovation frenzy is the radiology lab of Stanford University, which operates under the leadership of Olivier Gevaert, a professor from Flanders.
Finally, it’s important to point out that the AI tool will not replace radiologists any time soon, but that it will primarily guide radiologists toward a faster diagnosis. This is especially valuable now that hospitals worldwide are reaching the limits of their capacity. “Radiologists will still need to interpret the scans in their totality. After all, AI can’t simultaneously look for COVID-19 damage and damage caused by a tumor, for example, so further human interpretation is still required,” says Erik Ranschaert.
“What’s more,” Jonathan Berte adds, “we are confident that the AI tool will prove to remain useful after the coronavirus crisis. In the coming months or years, we will have to face mutations of the virus and take these into account in analyses, so AI assistance will still be welcome in detecting these mutations and acting faster.”