Fed-BioMed for Federated-PET project

Project description:

FEDERATED-LUNG is a multicentric research project that builds upon the foundations of FEDERATED-PET to further advance the use of artificial intelligence and medical imaging in the management of metastatic lung cancer.

The project focuses on developing new predictive biomarkers for response to immunotherapy and chemo-immunotherapy by leveraging multimodal data, including PET-CT imaging (18FDG) and clinical information. It aims to improve patient stratification and support more personalised treatment strategies in oncology.

In continuity with FEDERATED-PET, FEDERATED-LUNG involves a large network of French hospitals and expands the clinical scope to patients treated with combined immunotherapy and chemotherapy. The project introduces a broader range of AI methodologies, combining deep learning approaches with more interpretable and robust methods such as radiomics and machine learning.

A key objective is to develop complementary models, including tumour segmentation models and radiomics-based predictors, enabling the extraction of clinically relevant imaging biomarkers. These approaches aim to balance performance, interpretability, and robustness, while addressing the limitations of purely deep learning-based models that require large annotated datasets.

FEDERATED-LUNG also emphasises data annotation and harmonisation efforts, supporting the creation of high-quality datasets for AI development. By combining different modelling strategies and cross-validating them across datasets, the project seeks to establish reliable tools for predicting treatment response and guiding future clinical trials.

Overall, FEDERATED-LUNG represents a next step toward clinically actionable, AI-driven decision support systems in oncology, with strong translational perspectives for improving patient care.

Fed-BioMed position in the project:

Within FEDERATED-LUNG, Fed-BioMed continues to play a central role as the core infrastructure enabling federated learning across the network of participating hospitals.

Building on its deployment in FEDERATED-PET, Fed-BioMed supports the secure and distributed training of AI models on multimodal clinical and imaging data, allowing institutions to collaborate without sharing sensitive patient data. This ensures compliance with data protection regulations while enabling large-scale, real-world analysis.

In FEDERATED-LUNG, Fed-BioMed facilitates the integration of diverse modelling approaches—including deep learning, radiomics, and machine learning—within a unified federated framework. It enables the orchestration of experiments across multiple centres, supports data harmonisation workflows, and ensures the scalability of the infrastructure as new hospitals join the project.

In this context, Fed-BioMed strengthens its role as a key enabler of collaborative, privacy-preserving AI in healthcare, helping transition federated learning from experimental settings to more mature, clinically oriented applications.