Multi-Site Federated Brain Disease Diagnosis Method Based on Brain Topology Prototypes

The multi-site analysis provides the potential for improving diagnostic accuracy by integrating data from various sites. However, despite its above promise, multi-site applications face challenges such as data leakage and the non-independence and non-identical distribution of data. In this paper, we propose a method for multi-site joint brain disease diagnosis using brain network topological prototypes. Leveraging the federated learning (FL) framework, our approach enables collaborative analysis of functional magnetic resonance imaging (fMRI) data across multiple sites without sharing raw data, thus ensuring data privacy. By fusing brain networks constructed from multi-site topology prototypes with local Pearson coefficients, we develop a contrastive learning-based model for brain disease diagnosis capable of identifying brain diseases and extracting topological structure prototypes. Experimental results demonstrate that our method outperforms existing federated learning algorithms on multi-site datasets related to Autism Spectrum Disorder (ASD) and Attention Deficit Hyperactivity Disorder (ADHD).