Train medical AI models without exchanging data

AI tools have so far only been used hesitantly in routine clinical practice. One reason is that data exchange between hospitals is severely restricted by legal and ethical hurdles, especially in Germany. One solution to this problem is swarm learning. With swarm learning, several institutions can jointly train medical AI models without exchanging data. By using decentralized artificial intelligence and swarm learning. In cancer research, privacy laws and ethical hurdles make it difficult to share sensitive patient data between different research institutions, even though many patients are in principle in favor of their data being used for research purposes. Swarm learning makes it easier to meet privacy requirements. Swarm learning is a special form of machine learning in which models are trained without exchanging actual data between participants. The coordination and merging of models is done via a blockchain, eliminating the need for a central instance. The DECADE project builds on this method to use SL-based AI technology to solve real-world clinical problems related to colorectal cancer.

The project partners will use SL to develop AI algorithms for diagnosing and subtyping colorectal cancer and predicting disease progression. In doing so, they are setting a precedent for the use of SL in medicine that can serve as a template for any AI system in the healthcare sector. After all, more powerful AI systems could help physicians detect bowel cancer at an earlier stage and treat it more effectively. This could support medical staff and improve the care and treatment of colorectal cancer patients.

Publication: Swarm learning for decentralized artificial intelligence in cancer histopathology