Collaborating towards trustworthy AI in European health care

FLUTE is one of several projects contributing to the overall goal of improving health care in the EU. While FLUTE is concerned with the technology of federated learning and the disease of prostate cancer, other projects focus on other, complementary aspects. Still, more is needed than everybody addressing its own challenges. Only when the several initiatives collaborate we will be able to get to fully integrated solutions. The European Commission is increasing its efforts to facilitate such collaborations.

The power of collaboration for Federated Learning implementation in healthcare

Over the last decades, AI models have succeeded increasingly well to assist in health care, e.g., to provide decision support for diagnosis or treatment choice, or to provide early alerts of potential complications. To train such models, large amounts of data are needed, often from more patients than one can find in a single hospital. The FLUTE project aims at developing a platform for federated learning, exploiting secure multi-party computation to allow multiple hospitals to learn a statistical model by exchanging encrypted messages in such a way that nothing else than the privacy-preserving statistical model can be extracted from them. To exploit this new technology, collaboration with other projects is needed.

First, developing such platform is a major challenge. FLUTE builds on the software developed by the TRUMPET project. TRUMPET’s platform provides basic federated learning functionalities, which FLUTE aims to scale up to large volumes of data, larger numbers of data owners and more advanced models.

Second, FLUTE collaborates with other projects (HealthData4EU cluster) funded in similar Horizon Europe calls which are study complementary problems and are building complementary components, e.g., the SecureD project too is working on methods for synthetic data generation and scaling up multi-party computation.

Third, FLUTE initiated discussions with the EUCAIM project, which coordinates the exploitation of cancer data in Europe. In the long run, we hope that this collaboration will double the amount of data available to the project to train prostate cancer models. In particular, the FLUTE project aims at assessing MRI images to determine whether clinically significant prostate cancer may be present and whether a biopsy is needed.

In conclusion, the FLUTE project will fully engage in the Horizon Europe eco-system to leverage results of existing initiatives to maximize the impact of its contributions.  Members of the FLUTE consortium recently both participated in networking meetings organized by the European Commission to this extent, and had meetings with other projects to set up concrete collaborations.

Jan Ramon - INRIA