FedFound: a federated foundation model for lifespan brain morphological connectome analysis

The brain morphological connectome derived from structural MRI reflects inter-regional morphological relationships, providing a powerful representation for characterizing individual variability and detecting abnormalities across the lifespan. However, these abnormal alterations are subtle and complex, posing significant challenges for accurate and generalizable diagnosis using machine learning. Here, we present FedFound, the first federated foundation model inspired by the structured educational and residency training pathway of radiologists, designed for robust and scalable analysis of lifespan brain morphological connectomes. Integrating heterogeneous neuroimaging datasets across sites and disorders (22,911 subjects aged 0 to 100 years), FedFound combines self-supervised pre-training and supervised federated disease-specific refinement, supporting multidisciplinary knowledge aggregation through distributed optimization. Across nine diagnostic tasks spanning neurodevelopmental, neuropsychiatric, and neurodegenerative disorders, FedFound demonstrates superior performance and interpretability, revealing both shared and disorder-specific morphological patterns across etiologies. FedFound provides a robust foundation for lifespan neuroimage-based diagnosis that complements clinical expertise, while establishing a scalable and generalizable paradigm for integrating heterogeneous neuroimaging data across institutions, populations, and diseases to advance medical foundation models.