Trustworthy and Privacy-Preserving Collaborative AI Methods for Multicenter Neurodata

This Research Topic invites methodological, technical, and translational contributions that advance trustworthy, privacy-preserving, and collaborative AI for distributed multicenter neurodata. The emphasis is on deployable and reproducible approaches spanning neuroimaging, neurophysiology, and multimodal neuroscience data ecosystems.Large-scale neuroscience initiatives and clinical research networks have made multicenter neurodata the norm rather than the exception. Neuroimaging, neurophysiology, and increasingly behavioral, wearable, and digital phenotyping data are routinely collected across heterogeneous scanners, sensors, acquisition protocols, and patient populations. This creates major opportunities for robust and clinically useful AI, but also introduces a persistent bottleneck: the most valuable neurodata are distributed, highly heterogeneous, and subject to strict privacy and governance constraints.Collaborative AI provides a compelling response by enabling model development across institutions without requiring raw data centralization. Yet many current multicenter pipelines remain difficult to trust. Harmonization is often treated as an ad hoc preprocessing step, privacy protection is frequently appended rather than designed in, and cross-site evaluation is still commonly reduced to aggregate performance reporting. As a result, many promising methods do not transfer reliably across sites, time, or deployment conditions.This Research Topic seeks contributions that treat trustworthiness, privacy preservation, and collaboration as inseparable methodological requirements. We welcome submissions that formalize collaborative AI workflows, address site heterogeneity and non-IID learning, quantify privacy and security risks under explicit threat models, and provide rigorous evidence for reproducibility, robustness, calibration, fairness, and real-world feasibility in multicenter neurodata settings.Topics of Interest • Federated and distributed learning protocols for neuroimaging, EEG/MEG, iEEG, and other neurodata under non-IID and imbalanced conditions• Cross-site generalization, domain shift modeling, leave-one-site-out evaluation, and prospective drift assessment• Privacy-aware harmonization, confound control, and preprocessing strategies compatible with collaborative training• Threat modeling and leakage assessment, including membership inference, model inversion, and reconstruction risk in collaborative neurodata settings• Differential privacy, secure aggregation, privacy accounting, and other privacy-preserving learning mechanisms tailored to neuroscience workflows• Robust and secure collaborative training, including defenses against unreliable, malicious, or adversarial participants• Confidential computing and cryptographic approaches, such as MPC, homomorphic encryption, and trusted execution environments, with feasibility analysis• Uncertainty quantification, calibration, abstention policies, reliability monitoring, and interpretable failure analysis across sites• Auditable and interpretable collaborative AI, including provenance tracking, traceable preprocessing, governance-aware logging, and reproducibility artifacts• Fairness and bias analysis across demographic, clinical, and site-specific subgroups, together with privacy-compatible mitigation strategies• Communication-efficient and resource-efficient collaboration, including compression, asynchronous training, and edge-cloud or on-premise coordination• Benchmark suites, open toolchains, standardized reporting checklists, and validation protocols supporting independent verification• Deployment studies in multicenter clinical research or population neuroscience that explicitly analyze operational constraints and translational relevanceSubmission Focus and Article TypesSubmissions should make a clear methodological, technical, benchmarking, or translational contribution to trustworthy collaborative AI for multicenter neurodata. Purely incremental applications without rigorous validation, privacy analysis, or multicenter relevance are less suitable for this Research Topic.In line with the Medicine and Public Health section of Frontiers in Artificial Intelligence, relevant article categories may include Original Research, Methods, Review, Mini Review, Systematic Review, Technology and Code, Brief Research Report, Perspective, and Study Protocol, among other article types supported by the section. Topic editors: Sibo Qiao, Cong Liu Source: https://www.frontiersin.org/research-topics/81264/trustworthy-and-privacy-preserving-collaborative-ai-methods-for-multicenter-neurodata