NEXUS-MI: Communication-Aware Federated Personalization for Gateway-Coordinated Motor-Imagery Brain-Computer Interfaces

EEG-based motor-imagery brain-computer interfaces require personalization across subjects and recording sessions. NEXUS-MI introduces a gateway-coordinated federated framework that jointly considers personalization and communication-aware synchronization. Raw EEG and personalized classifier heads remain local while an edge coordinator maintains a shared backbone. Offline replay evaluations use BCI Competition IV Dataset 2a and OpenBMI, with limited Session 2 calibration and held-out testing. The study compares an ideal-link reference with heterogeneous communication policies governing gateway participation, buffered and stale updates, and backbone downloads. Communication-aware coordination reduces server-to-client backbone traffic by approximately 42% on both datasets, while revealing subject-level accuracy differences that can be obscured by cohort averages.