Linking public and private neuroimaging data with decentralized analysis via NeuroFLAME Vaults

Abstract

Large-scale datasets and global collaborations are increasingly enabling exciting discoveries in neuroimaging research; however, barriers such as data privacy regulations and unequal access to computational infrastructure continue to hinder progress. Federated analysis offers a promising solution by allowing decentralized analysis without centralized data sharing, but logistical challenges often impede its practical implementation. In this work, we present NeuroFLAME Vaults, a novel framework that builds upon the principles of federated analysis to streamline the privacy-preserving, decentralized analysis of neuroimaging data. NeuroFLAME Vaults provide continuously accessible, preprocessed datasets hosted at local sites, enabling researchers to participate in collaborative studies without transferring sensitive data or coordinating synchronous computation. We describe the design and implementation of the platform, demonstrating its utility through multiple vaults derived from public and private datasets, including structural and functional MRI data from studies on schizophrenia, substance use, and neurodevelopment. Vault-based analyses leverage standardized pipelines and ridge regression models to enable cross-site modeling and data harmonization. Our results replicate known findings in neurodevelopment, psychiatric disorders, and substance use, validating the approach and showcasing its potential for inclusive and scalable neuroimaging research. NeuroFLAME Vaults significantly reduce the technical and logistical burdens of collaborative analysis, while upholding FAIR data principles, offering a powerful tool to accelerate global neuroscience.