It is easier to identify COVID-19 diseases in an advanced stage by means of X-ray and computer tomography (CT) of the lungs, rather than by conventional PCR tests of throat swabs. nic.at employee Aaron Kaplan and his brother Hamilton have now developed an AI (artificial intelligence) device in their Vienna AI laboratory "Deep Insights", which is able to diagnose the so-called corona virus with a sensitivity rate of more than 90 percent, which could support diagnosis in hospitals. We are of course very proud of our dedicated staff and congratulate them warmly!

The "COVID-19 image classifier" can be used in those more severe cases where the virus has already reached and attacked the lungs. The artificial intelligence of "Deep Insights" is a so-called "Convolutional Neuronal Network", which has been trained using thousands of anonymized images, and provides one of three possible results: A COVID-19 diagnosis, a pathological finding (such as pneumonia) or a normal finding. In principle, it would also be possible to use AI for simple x-rays, which could be produced in many more clinics and at considerably lower cost, although the detection rate is not yet as high as for CT images.

"The idea was born during cooperation with a radiologist from Vienna. Since 2018, I have been able to participate in a regular AI expert panel. This "fast.ai" group met regularly and we were able to acquire knowledge that is now important for the project. The combination of expertise (radiology) and the fast.ai meetup was ultimately decisive," reports Aaron.

Behind the siblings are several very busy weekends and long evenings: "But it definitely paid off. A sensitivity rate of more than 90 percent is a very good result in our eyes - especially if you want to keep the data set as balanced as possible and deliberately use a lot of tricky cases in testing. Now it is particularly important that this open source solution is actually used and makes a valuable contribution". The AI has been developed in cooperation with the Viennese company LifeTec, which contributed its medical expertise. Publication as an open source solution was done deliberately - the license allows extensive modifications, further distribution and of course testability in practice.

Aaron and his brother hope that their development will be tested  as quickly and extensively as possible in order to further improve it. At the same time, they also point out that artificial intelligence cannot replace a doctor: "Ultimately, humans must make the right diagnosis based on the wealth of experience they have built up over the years. AI does not know whether the patient currently has a high fever or cough, so this solution can only be an additional instrument in the medical profession's toolbox".

nic.at is proud of the enormous efforts of Aaron, who has acquired basic knowledge about AI (in his spare time!) and is certainly making an extremely valuable contribution to the current COVID-19 crisis. Of course this know-how is also very advantageous for our company, especially in the area of CERT (Computer Energy Response Team) and in Research & Development.