PRED-IRM

Project description:

PRED-IRM is a clinical research project in oncology aiming to support a significant advance in the treatment of glioblastoma, an aggressive form of brain cancer.

The project develops an AI-based predictive model trained on multi-centric clinical data, including MRI scans and radiotherapy treatment plans. Its goal is to enable early prediction of patient response to combined radiotherapy and chemotherapy, supporting more precise and personalised treatment strategies.

PRED-IRM builds on prior research conducted at the Centre François Baclesse and the GREYC CNRS laboratories, extending an initial predictive model into a robust multi-center model.

To ensure patient privacy and regulatory compliance, the project relies on federated learning. This approach enables collaborative model training across participating institutions without centralising sensitive health data: patient data remains securely stored at each site, while only model updates are exchanged.

The project brings together five cancer research centers from the Unicancer network and is supported by the Fonds AMGEN France pour la Science et l’Humain.

Fed-BioMed’s position in the project:

Fed-BioMed provides the technological foundation for secure, privacy-preserving, and regulation-compliant federated learning within PRED-IRM.

Its interactive capabilities for model design, testing, and tuning are a key asset for the project, enabling researchers to iterate efficiently while maintaining strict data governance constraints.

The close collaboration between the PRED-IRM consortium and the Fed-BioMed team also contributes to the evolution of the platform. It helps identify and develop new features that address the needs of real-world clinical federated learning deployments in oncology.