Test the ability to design an end-to-end federated learning architecture for cross-institutional medical imaging that preserves patient privacy and resists Byzantine faults under limited bandwidth and heterogeneous hardware. Assess selection and integration of privacy mechanisms (DP, secure aggregation, HE/MPC), robust aggregation and anomaly detection, communication-efficient protocols (compression, sparsification, async updates), strategies for client heterogeneity and resource constraints (personalization, split learning, model distillation, client selection), plus evaluation criteria (privacy-utility tradeoffs, robustness, latency, resource usage) and a validation/deployment plan considering regulatory constraints.