<em>safeMEDInet</em>: Federated AI Systems for Security and Privacy-Preserving Threat Detection in Distributed Healthcare

The explosive growth of digital healthcare data and networked Internet of Medical Things (IoMT) devices has heightened vulnerabilities inside healthcare networks, hence exposing sensitive medical systems to sophisticated cyber assaults. The safeMEDInet framework offers a secure, federated artificial intelligence (AI) architecture that allows decentralized healthcare institutions to cooperatively identify and address problems without disclosing raw patient data. safeMEDInet utilizes federated learning along with privacy-preserving techniques, such as differential privacy, homomorphic encryption, and Byzantine-resilient aggregation, to guarantee confidentiality, integrity, and adherence to regulations in remote settings. The proposed framework integrates a hybrid CNN-LSTM model for spatiotemporal intrusion detection with secure model synchronization and encrypted parameter sharing to ensure robust accuracy against various cyber-attacks, including ransomware, unauthorized access, and distributed denial-of-service (DDoS) intrusions. Empirical assessments utilizing MIMIC-IV, HealthData.gov, and WHO datasets reveal that safeMEDInet achieves a detection accuracy of 96.8% with robust privacy assurances (ε = 1.9) and sustains an accuracy of 88.4% despite 30% Byzantine interference, surpassing traditional federated and centralized systems. The findings confirm safeMEDInet's capacity to guarantee high detection reliability, low processing cost, and mathematical assurance of privacy resilience. This research positions safeMEDInet as a pivotal advancement towards safe, scalable, and ethically governed Healthcare 5.0 ecosystems, incorporating AI-driven privacy, federated cooperation, and blockchain-supported data integrity for next-generation medical cybersecurity.