Based on simple X-ray images, artificial intelligence (AI) will be able to predict the risk of a hip fracture occurring in the next ten years. What just a short time ago sounded like science fiction is now reality and only one part of the cooperation between the Faculty of Medicine at Kiel University (CAU) and the University Hospital Schleswig-Holstein (UKSH) with the University of California, San Francisco (UCSF). A delegation from the CAU and UKSH joined Schleswig-Holstein's Minister President Daniel Günther in opening the new infrastructure for AI in the USA in the night of 8 June 2023 (German time).
A first application will be predicting hip fractures based on X-ray images. It is not possible for people to accurately predict whether a fracture will occur within the next ten years based on simple images. The AI at the CAU is currently creating a predictive accuracy that is better than that of current medical check-ups. Together with the UCSF data, the success rate is to be further improved. Other projects are also planned for the short and medium term. One example is the automated detection of the cause of strokes in the ER department: is there bleeding that needs to be stopped or does a blocked vein need to be reopened? It is vital for patients to get the right one of these extremely different treatments, urgently. AI can help doctors not to miss anything.
The "AI Exchange" project between the "Intelligent Imaging Lab" at the CAU and the "Center for Intelligent Imaging" at UCSF uses "federated learning" technology, which protects sensitive data in particular, such as X-ray images. Prof. Dr Claus-Christian Glüer leads the project and explains:
"With this technology, all sensitive medical data remains on site, at the UKSH or at UCSF, so that data protection is guaranteed."
Together with Prof. Dr Jan-Bernd Hövener, he set up the Intelligent Imaging Lab and the Section Biomedical Imaging in the Radiology Department. Hövener explains further:
"Instead of sending the data back and forth, we train the networks at each location within the respective firewalls. After a certain time, we combine the results of the local networks so that we can take advantage of the large amounts of data from both sites and protect our data."
The AI networks at both locations will initially be trained independently at each site. These are sent to a central server at the UKSH and merged, which results in a joint new network that is more experienced than the individual ones. This is then sent back again and again to the local training sites until it is optimally trained.
- Kiel University (08.06.2023): Predicting disease risks in a timely manner: medical AI from the universities in Kiel and San Francisco launches