NeuroNimbus: Enabling Privacy-Preserving Real-Time Clinical Intelligence via Federated and Cloud-Orchestrated Deep Learning
Vijay Govindarajan, Pratik Patel · IEEE Access · 2025
We present NeuroNimbus, a federated and cloud-orchestrated deep learning framework for privacy-preserving, real-time clinical decision support. NeuroNimbus keeps data local at hospitals and uses a lightweight cloud service to coordinate model updates with privacy guarantees. On MIMIC-III and eICU, NeuroNimbus holds edge inference latency at 84 ms, reduces communication rounds by 34% and transmitted bytes by 38% versus fixed scheduling, and matches near-centralized AUC for mortality and sepsis prediction. The system scales in simulation to 50 heterogeneous clients without network congestion. Baselines include FedAvg, FedProx, and simple heuristic schedules; ablations isolate the effect of our RL orchestration (FCPO).